Aiming to identify precise evidence sources from visual documents, visual evidence attribution for visual document retrieval–augmented generation (VD-RAG) ensures reliable and verifiable predictions from vision-language models (VLMs) in multimodal question answering. Most existing methods adopt end-to-end training to facilitate intuitive answer verification. However, they lack fine-grained supervision and progressive traceability throughout the reasoning process. In this paper, we introduce the Chain-of-Evidence (CoE) paradigm for VD-RAG. CoE unifies Chain-of-Thought (CoT) reasoning and visual evidence attribution by grounding reference elements in reasoning steps to specific regions with bounding boxes and page indexes. To enable VLMs to generate such evidence-grounded reasoning, we propose Look As You Think (LAT), a reinforcement learning framework that trains models to produce verifiable reasoning paths with consistent attribution. During training, LAT evaluates the attribution consistency of each evidence region and provides rewards only when the CoE trajectory yields correct answers, encouraging process-level self-verification. Experiments on vanilla Qwen2.5-VL-7B-Instruct with Paper‑ and Wiki‑VISA benchmarks show that LAT consistently improves the vanilla model in both single- and multi-image settings, yielding average gains of 8.23% in soft exact match (EM) and 47.0% in [email protected]. Meanwhile, LAT not only outperforms the supervised fine-tuning baseline, which is trained to directly produce answers with attribution, but also exhibits stronger generalization across domains.

FengXian Dong, Zhi Zheng, Xiao Han, Wei Chen, Jingqing Ruan, Tong Xu, Yong Chen, Enhong Chen. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Ning Li, Zheng Zhang, Zhenya Huang, Rui Li, Yi Zhan, Yinbo Luo, Qi Liu, Enhong Chen. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Junda Lin, Zhaomeng Zhou, Zhi Zheng, Shuochen Liu, Tong Xu, Yong Chen, Enhong Chen. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Derong Xu, Shuochen Liu, Pengfei Luo, Pengyue Jia, Yingyi Zhang, Yi Wen, Yimin Deng, Wenlin Zhang, Enhong Chen, Xiangyu Zhao, Tong Xu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Xuyang Zhi, Peilun Zhou, Chengqiang Lu, Hang Lv, Yiwei Liang, Rongyang Zhang, Yan Gao, Yiwu, Yao Hu, Hongchao Gu, Defu Lian, Hao Wang, Enhong Chen. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Large language models (LLMs) are revolutionizing user interactions in online map applications by enabling conversational interfaces with intelligent map agents. However, inherent ambiguities and nuances in human communication often lead to incomplete or unclear user instructions. While generating clarification questions can mitigate this issue, existing methods—which rely on prompt-based LLM assessments or supervised fine-tuning (SFT) of LLMs with limited annotated data—struggle to reliably determine when clarification is necessary, resulting in cold-start issues and reduced adaptability. To address this gap, we propose an uncertainty-aware dynamic planning framework for intent disambiguation in interactive agent systems, exemplified by Baidu Maps. Our framework leverages LLMs to dynamically generate agent action sequences while estimating query uncertainty. When uncertainty thresholds are exceeded, targeted clarification questions are triggered. Besides, to efficiently navigate the vast uncertainty space, we introduce a novel beam search-based pruning strategy. Crucially, the framework supports context-aware task planning without any additional model training. We evaluate our approach on Baidu Maps and a public robotic task, focusing on uncertainty management and adaptability. Experiments demonstrate that the framework achieves high precision across diverse LLMs without requiring an additional training phase, highlighting its versatility and robustness. Ablation studies confirm that the framework can significantly improve user intent recognition and task success rates.

Interactive recommendation is a typical information-seeking task that allows users to interactively express their needs through natural language and obtain personalized recommendations. Large language model-powered (LLM-powered) agents have become a new paradigm in interactive recommendations, effectively capturing users' real-time needs and enhancing personalized experiences. However, due to limited planning and generalization capabilities, existing formulations of LLM-powered interactive recommender agents struggle to effectively address diverse and complex user intents, such as intuitive, unrefined, or occasionally ambiguous requests. To tackle this challenge, we propose a novel ThoughtAugmented Interactive Recommender Agent system (TAIRA) that addresses complex user intents through distilled thought patterns. Specifically, TAIRA is designed as an LLM-powered multi-agent system featuring a manager agent that orchestrates recommendation tasks by decomposing user needs and planning subtasks, with its planning capacity strengthened through Thought Pattern Distillation (TPD), a thought-augmentation method that extracts high-level thoughts from the agent's and human experts' experiences. Through comprehensive experiments conducted across multiple designed datasets, TAIRA exhibits significantly enhanced performance compared to existing methods. Notably, TAIRA's thought augmentation strategies endow the agent system with the ability to solve complex tasks while generalizing effectively on novel tasks, validating its potential as a foundational framework for agent systems, particularly in complex user intent scenarios.

DNNs have gained widespread adoption in feature interaction recommendation models. However, there has been a longstanding debate on their roles. On one hand, some works claim that DNNs possess the ability to implicitly capture high-order feature interactions. Conversely, recent studies have highlighted the limitations of DNNs in effectively learning dot products, specifically second-order interactions, let alone higher-order interactions. In this paper, we present a novel perspective to understand the effectiveness of DNNs: their impact on the dimensional robustness of the representations. In particular, we conduct extensive experiments involving both parallel DNNs and stacked DNNs. Our evaluation encompasses an overall study of complete DNN on two feature interaction models, alongside a fine-grained ablation analysis of components within DNNs. Experimental results demonstrate that both parallel and stacked DNNs can effectively mitigate the dimensional collapse of embeddings. Furthermore, a gradient-based theoretical analysis, supported by empirical evidence, uncovers the underlying mechanisms of dimensional collapse. The code is accessible for reproduction. https://github.com/USTC-StarTeam/Dimensional-Collapse-Analysis

Learning transferable representations from unlabeled time series is crucial for improving performance in data-scarce classification. Existing self-supervised methods often operate at the point level and rely on unidirectional encoding, leading to low semantic density and a mismatch between pre-training and downstream optimization. In this paper, we propose TimeMAE, a self-supervised framework that reformulates masked modeling for time series via semantic unit elevation and decoupled representation learning. Instead of modeling individual time steps, TimeMAE segments time series into non-overlapping sub-series to form semantically enriched units, enabling more informative masked reconstruction while reducing computational cost. To address the representation discrepancy introduced by masking, we design a decoupled masked autoencoder that separately encodes visible and masked regions, avoiding artificial masked tokens in the main encoder. To guide pre-training, we introduce two complementary objectives: masked codeword classification, which discretizes sub-series semantics via a learned tokenizer and masked representation regression, which aligns continuous representations through a momentum-updated target encoder. Extensive experiments on five datasets demonstrate that TimeMAE outperforms competitive baselines, particularly in label-scarce scenarios and transfer learning scenarios. Our codes are publicly available at https://github.com/Mingyue-Cheng/TimeMAE.

Time series forecasting (TSF) traditionally relies on fast-thinking paradigms that map historical observations directly to future sequences of continuous values. While effective, such approaches often frame forecasting as a pattern-matching problem and tend to overlook explicit reasoning over temporal dynamics and contextual factors, which are critical for modeling long-range dependencies and non-stationary behaviors in real-world scenarios. Recent slow-thinking large language models (LLMs), such as OpenAI o1 and DeepSeek-R1, demonstrate strong inference-time multi-step reasoning abilities. This raises a fundamental question: can slow-thinking LLMs reason over temporal dynamics to support accurate TSF, even without task-specific training? To investigate this question, we present TimeReasoner, a systematic empirical study that reformulates TSF as a conditional reasoning process performed entirely at inference time. TimeReasoner integrates hybrid instructions consisting of task directives, timestamps, sequential values, and optional contextual features, and induces multi-step temporal reasoning in pretrained slow-thinking LLMs through chain-of-thought prompting and rollout-based reasoning strategies. Extensive experiments across diverse TSF benchmarks show that slow-thinking LLMs consistently outperform prior baselines or achieve competitive training-free forecasting performance. Beyond accuracy, we analyze how different inference-time reasoning strategies influence forecasting behaviors, highlighting both the potential and limitations of slow-thinking paradigms for TSF.

Search-based recommendation is one of the most critical application scenarios in e-commerce platforms. Users' complex search contexts—such as spatiotemporal factors, historical interactions, and current query's information—constitute an essential part of their decision-making, reflecting implicit preferences that complement explicit query terms. Modeling such rich contextual signals and their intricate associations with candidate items remains a key challenge. Although numerous efforts have been devoted to building more effective search methods, existing approaches still show limitations in integrating contextual information, which hinders their ability to fully capture user intent. To address these challenges, we propose a context-aware reasoning-enhanced generative search framework for better understanding the complicated context. Specifically, the framework first unifies heterogeneous user and item contexts into textual representations or text-based semantic identifiers and aligns them. To overcome the lack of explicit reasoning trajectories, we introduce a self-evolving post-training paradigm that iteratively combines supervised fine-tuning and reinforcement learning to progressively enhance the model's reasoning capability. In addition, we identify potential biases in existing RL algorithms when applied to search scenarios and present a debiased variant of GRPO to improve ranking performance. Extensive experiments on search log data collected from a real-world e-commerce platform demonstrate that our approach achieves superior performance compared with strong baselines, validating its effectiveness for search-based recommendation.

High-speed imaging, which captures the fleeting dynamics of moving objects at extreme frame rates, has become an indispensable tool across a wide range of scientific disciplines. Yet, the pursuit of high temporal resolution often comes at the cost of significant image degradation, due to the inherent limitations of imaging sensors and the extreme conditions of ultra-short exposure and massive data throughput. As a result, high-speed cameras often produce images marred by strong noise and severe color distortions. In this work, we propose a deep image signal processing (ISP) paradigm that enables high-speed cameras to maintain extremely high frame rates while achieving image quality comparable to that of digital single-lens reflex (DSLR) cameras. To this end, we make two key contributions: 1) constructing RHID, the first large-scale real-world high-speed imaging ISP dataset, comprising 282,912 RAW images captured by high-speed cameras and corresponding sRGB images captured by DSLRs, featuring complex degradations intrinsic to high-speed acquisition; and 2) proposing a misalignment-robust ISP learning framework (MisISP), equipped with a prior mapper-guided image alignment module (PMIA) and a spectrum-guided weakly-aligned image supervisory loss, which effectively addresses inherent pixel misalignments caused by heterogeneous sensor characteristics. Extensive experiments demonstrate that our paradigm substantially advances the performance of existing deep ISP models for high-speed imaging, achieving remarkable improvements in noise suppression, brightness enhancement, and color preservation.

Enabling machines to solve mathematical problems is a vital endeavor in developing intelligence that emulates human-like thinking and reasoning. However, most existing approaches focus on reconstructing human comprehension of problems, which are still far from enough since they neglect the fundamental human ability to learn knowledge from experiences. In this article, we focus on empowering models with the cognitive capacity to autonomously learn knowledge from mathematical problem-solving. We first propose a Cognitive Solver (CogSolver) that contains an intelligent BRAIN-ARM framework as the cognitive structure and operates the knowledge learning process in Store-Apply-Update steps inspired by two cognitive science theories. The BRAIN system stores three basic types of mathematical knowledge, and the ARM system applies them organically in answer reasoning process. After solving problems, the BRAIN updates its stored knowledge based on the ARM's feedback, with knowledge filters to eliminate redundancies and foster a more rational knowledge base. Our CogSolver carries out the above three steps iteratively, emulating a more human-like behavior. Furthermore, in order to overcome knowledge forgetting during the learning process, we extend CogSolver to CogSolver+ by incorporating an essential knowledge Recall mechanism, which is inspired by another prominent cognitive theory. We first discuss and fuse three crucial factors in simulating human memory replay. Then, we propose a influenced-based method with a theoretical guarantee of efficiency to consolidate the updated knowledge. Experiments on three math word problem benchmarks demonstrate the improvements of our CogSolver and CogSolver+ in answer reasoning and clearly illustrate how they acquire knowledge, leading to superior interpretability.

With the advancement of deep learning, deep recommendation models have achieved remarkable improvements in recommendation accuracy. However, due to the large number of candidate items in practice and the high cost of preference computation, these methods still suffer from low recommendation efficiency. The recently proposed tree-based deep recommendation models alleviate the problem by directly learning tree structure and representations under the guidance of recommendation objectives. To guarantee the effectiveness of beam search for recommendation accuracy, these models strive to ensure that the tree adheres to the max-heap assumption, where a parent node's preference should be the maximum among its children's preferences. However, they employ a one-versus-all strategy, framing the training task as a series of independent binary classification objectives for each node, which limits their ability to fully satisfy the max-heap assumption. To this end, we propose a Deep Tree-based Retriever (DTR for short) for efficient recommendation. DTR frames the training task as a softmax-based multi-class classification over tree nodes at the same level, enabling explicit horizontal competition and more discriminative top-k selection among them, which mimics the beam search behavior during training. To mitigate the suboptimality induced by the labeling of non-leaf nodes, we propose a rectification method for the loss function, which further aligns with the max-heap assumption in expectation. As the number of tree nodes grows exponentially with the levels, we employ sampled softmax to approximate optimization and thereby enhance efficiency. Furthermore, we propose a tree-based sampling method to reduce the bias inherent in sampled softmax. Theoretical results reveal DTR's generalization capability, and both the rectification method and tree-based sampling contribute to improved generalization. The experiments are conducted on four real-world datasets, validating the effectiveness of the proposed method.

Vision Foundation Models(VFMs) have achieved remarkable success in various computer vision tasks.However, their application to semantic segmentation is hindered by two significant challenges: (1) the disparity in data scale, as segmentation datasets are typically much smaller than those used for VFM pre-training, and (2) domain distribution shifts, where real-world segmentation scenarios are diverse and often underrepresented during pre-training. To overcome these limitations, we present Rein++, an efficient VFM-based segmentation framework that demonstrates superior generalization from limited data and enables effective adaptation to diverse unlabeled scenarios. Specifically, Rein++ comprises a domain generalization solution Rein-G and a domain adaptation solution Rein-A. Rein-G introduces a set of trainable, instance-aware tokens that effectively refine the VFM's features for the segmentation task. This parameter-efficient approach fine-tunes less than 1% of the backbone's parameters, enabling robust generalization. Building on the Rein-G, Rein-A performs unsupervised domain adaptation at both the instance and logit levels to mitigate domain shifts. In addition, it incorporates a semantic transfer module that leverages the class-agnostic capabilities of the segment anything model to enhance boundary details in the target domain. The integrated Rein++ pipeline first learns a generalizable model on a source domain (e.g., daytime scenes) and subsequently adapts it to diverse target domains (e.g., nighttime scenes) without any target labels. Comprehensive experiments demonstrate that Rein++ significantly outperforms state-of-the-art methods with efficient training, underscoring its roles an efficient, generalizable, and adaptive segmentation solution for VFMs, even for large models with billions of parameters.

Computerized Adaptive Testing (CAT) offers an efficient and personalized method for assessing examinee proficiency by dynamically adjusting test questions based on individual performance. Compared to traditional, non-personalized testing methods, CAT requires fewer questions and provides more accurate assessments. As a result, CAT has been widely adopted across various fields, including education, healthcare, sports, sociology, and the evaluation of AI models. While traditional methods rely on psychometrics and statistics, the increasing complexity of large-scale testing has spurred the integration of machine learning techniques. This paper aims to provide a machine learning-focused survey on CAT, presenting a fresh perspective on this adaptive testing paradigm. We delve into measurement models, question selection algorithm, bank construction, and test control within CAT, exploring how machine learning can optimize these components. Through an analysis of current methods, strengths, limitations, and challenges, we strive to develop robust, fair, and efficient CAT systems. By bridging psychometric-driven CAT research with machine learning, this survey advocates for a more inclusive and interdisciplinary approach to the future of adaptive testing.

The knowledge tracing (KT) aims to track changes in students' knowledge status and predict their future answers based on their historical answer records. Current research on KT modeling focuses on predicting student' future performance based on existing, unupdated records of student learning interactions. However, these approaches ignore the distractors (such as slipping and guessing) in the answering process and overlook that static cognitive representations are temporary and limited. Most of them assume that there are no distractors in the answering process and that the record representations fully represent the students' level of understanding and proficiency in knowledge. In this case, it may lead to many lack of synergy and incoordination issue in the original records. Therefore we propose a cognitive representation optimization for KT (CRO-KT) model, which utilizes a dynamic programming algorithm to optimize structure of cognitive representations. This ensures that the structure matches the students' cognitive patterns in terms of the difficulty of the exercises. Furthermore, we use the co-optimization algorithm to optimize the cognitive representations of the subtarget exercises in terms of the overall situation of exercises responses by considering all the exercises with co-relationships as a single goal. Meanwhile, the CRO-KT model fuses the learned relational embeddings from the bipartite graph with the optimized record representations in a weighted manner, enhancing the expression of students' cognition. Finally, experiments are conducted on three publicly available datasets respectively to validate the effectiveness of the proposed cognitive representation optimization model. The source code of CRDP-KT is available at https://github.com/bigdata-graph/CRO-KT.

The reliability of vision-based systems, such as traffic monitoring and intelligent driving, is typically compromised in hazy weather owing to diminished visibility. In this paper, we propose a novel efficient image dehazing framework guided by depth order, leveraging the consistency of depth perception to establish strong global constraints for enhanced haze removal. The consistent depth perception ensures that the regions that look farther or closer in hazy images also appear farther or closer in the corresponding dehazing results, substantially avoiding potential visual degradation. To this end, the depth order in hazy images is approximated by the reverse order of color difference between pixel values and global atmospheric light, offering an effective and efficient alternative for depth perception modeling. Subsequently, we have developed a depth order embedded transformation model to estimate the transmission maps jointly constrained by depth order and haze imaging model, ensuring that the depth order remains unchanged in corresponding dehazing results. This model harnesses the extracted depth order as a powerful global constraint for the dehazing process, facilitating the efficient use of global information and thus achieving superior image restoration. Extensive experiments demonstrate that the proposed method can better recover potential structure and vivid color with higher computational efficiency, offering an efficient solution for robust traffic monitoring against hazy weather.

Science Question Answering (SQA) is an important task for evaluating models' capability to reason with scientific knowledge. However, the extensive availability of scientific information (e.g., basic concepts in biology, physics, and chemistry) in pre-trained corpora may cause large language models (LLMs) to rely more on memorized information rather than actual reasoning when answering questions. This reliance persists even with techniques like Chain-of-Thought prompting, resulting in shallow understanding and limited reasoning based on scientific knowledge. Therefore, to enhance LLMs' capacity to comprehend and apply scientific knowledge, we propose a framework calledMulti-AgentCooperation-basedKnowledgeExemplification (MCKE). Specifically, MCKE leverages knowledge alongside questions to create exemplified knowledge, promoting deeper understanding through innovative knowledge representation. To better evaluate the model's ability to reason and apply knowledge, we introduceNovSciQA, a multiple-choice question answering dataset based on newly created scientific knowledge. This dataset covers multi-subject scientific knowledge and questions that do not exist in reality, making it impossible for the model to rely on memorized answer-related information to answer questions. Experimental results show that the MCKE framework outperforms baselines, and the NovSciQA dataset effectively assesses models' knowledge understanding and application. Our code and dataset are available inhttps://anonymous.4open.science/r/MCKE-NovSciQA.

People's social relationships are often manifested through their surroundings, with certain objects or interactions acting as symbols for specific relationships, e.g., wedding rings, roses, hugs, or holding hands. This brings unique challenges to recognizing social relationships, requiring understanding and capturing the essence of these contexts from visual appearances. However, current methods of social relationship understanding rely on the basic classification paradigm of detected persons and objects, which fails to understand the comprehensive context and often overlooks decisive social factors, especially subtle visual cues. To highlight the social-aware context and intricate details, we propose a novel approach that recognizes contextual social relationships (ConSoRs) from a social cognitive perspective. Specifically, to incorporate social-aware semantics, we build a lightweight adapter upon the frozen contrastive language-image pretraining (CLIP) to learn social concepts via our novel multimodal side adapter tuning mechanism. Furthermore, we construct social-aware descriptive language prompts (e.g., scene, activity, objects, and emotions) with social relationships for each image, and then compel ConSoR to concentrate more intensively on the decisive visual social factors via visual-linguistic contrasting. Impressively, ConSoR outperforms previous methods with a 7.6% gain on the people-in-social-context (PISC) dataset and a 9.8% increase on the people-in-photo-album (PIPA) benchmark. Furthermore, we observe that ConSoR excels at finding critical visual evidence to reveal social relationships. Code is available at https://github.com/starsholic/ConSoR.

Recent years have witnessed the prosperity of Knowledge Graph-Based Recommendation System (KGRS), which enriches the representation of users, items, and entities by structural knowledge with striking improvement. Nevertheless, its unaffordable computational cost still limits researchers from exploring more sophisticated models. We observe that the bottleneck for training efficiency arises from the knowledge graph, which is plagued by the well-known issue of knowledge explosion. Recently, some works have attempted to slim the inflated KG via summarization techniques, which summarize multiple real nodes into the single virtual one. However, these summarized virtual nodes may ignore collaborative signals and thus fail to figure out the redundant nodes related to recommendation task. To this end, in this article, we propose a novel approach called KGTrimmer for knowledge graph pruning tailored for recommendation, to remove the unessential nodes while minimizing performance degradation. Specifically, we design an importance evaluator from a dual-view perspective. For the collective view, we embrace the idea of collective intelligence by extracting community consensus based on abundant collaborative signals, i.e., nodes are considered important if they attract attention of numerous users. For the holistic view, we learn a global mask to identify the valueless nodes from their inherent properties or overall popularity. With the collective and holistic importance scores, we build an end-to-end importance-aware graph neural network, which injects filtered knowledge to enhance the distillation of valuable user-item collaborative signals. Ultimately, we generate a pruned knowledge graph with lightweight, stable, and robust properties to facilitate the following-up recommendation task. Extensive experiments are conducted on three publicly available datasets to prove the effectiveness and generalizability of KGTrimmer, where it can reduce the number of triplets in KG by up to 90% without compromising performance.

Graph Neural Networks (GNNs) have been widely used across various fields under the homophily assumption that connected nodes are similar. However, in heterophilic graphs, where connected nodes tend to have dissimilar features, existing GNNs still face some limitations. From the perspective of structure, shallow GNNs could not capture the high-order node information, whereas deep GNNs may suffer from the over-smoothing problem. From the perspective of feature, the useful information of high-order similar nodes is often weakened by low-order dissimilar nodes in the feature update phase. To address the above problems, we propose a Global Structure-aware and Feature-augmented Graph Neural Network (GSF-GNN) to alleviate the limitations from the perspectives of structure and feature. Specifically, from the structure perspective, we design a Structure-based Global Propagation (SGP) module to establish global connections among nodes and adaptively adjust edge weights for message propagation. From the feature perspective, we introduce a Feature-augmented Compensatory Update (FCU) module, which employs a multi-view feature updating mechanism to enhance node features from different perspectives. Our theoretical analysis formally demonstrates the effectiveness of GSF-GNN in heterophilic graphs. Experiments on heterophilic and homophilic benchmark datasets validate the effectiveness of GSF-GNN across various graph structures. Moreover, GSF-GNN achieves stable performance across multiple layers and effectively alleviates the over-smoothing problem. Our codes are available on https://github.com/huijieliu2023/GSF-GNN .

User modeling serves as a crucial foundation for researchers to capture useful potential characteristics, playing a pivotal role in various applications such as recommender systems. One common challenge in user modeling is the cold-start problem, where interactions are notably limited for new users. To tackle this issue, the paradigm of meta-learning has been introduced to user modeling, yielding promising results. Similar to a guidebook for a new traveler, meta-learning significantly influences decision-making for new users in critical scenarios, such as career recommendations. Consequently, the issue of fairness in meta-learning has become paramount. Several methods have been proposed to mitigate unfairness in meta-learning and have shown promising results. However, a fundamental question remains unexplored: What is the critical factor leading to unfairness in meta-learned user modeling? Through theoretical analysis that integrates the meta-learning paradigm with group fairness metrics, we identify group proportion imbalance as a critical factor. Subsequently, another question arises: How can we mitigate the influence of this factor to enhance fairness while ensuring accuracy? To this end, we introduce a novel F airness-aware A daptive S ampling framework for me T a-learning, abbreviated as FAST. Its core concept involves adaptively adjusting the sampling distribution for different user groups during the interleaved training process of meta-learning. Moreover, we provide theoretical guarantees demonstrating the convergence of FAST, showcasing its potential to effectively eliminate unfairness. Furthermore, to ensure model accuracy, we enhance FAST with FAST+ by introducing a hybrid sampling strategy at an individual level. This strategy prioritizes fairness and thoroughly explores important users during the sampling process, allowing for a better accuracy-fairness tradeoff. Finally, we conduct extensive experiments on real-world datasets, which demonstrate the effectiveness of both FAST and FAST+ frameworks.

Cross-Domain Recommendation (CDR) can effectively alleviate the data sparsity issue in the recommendation system by transferring the source domain knowledge to the target domain. Many CDR methods try to find a mapping of latent embeddings of the overlapping nodes (e.g., the overlapping users) from source domains to target domains. However, due to the randomness of parameter initialization and model training, the embeddings from different domains may differ a lot and the inter-domain mapping can be difficult to learn. To this end, in this work, we propose a novel Similarity-Preserving Cross-Domain Recommendation ( SimCDR ) framework. Instead of trying to learn the inter-domain mapping, SimCDR aims to preserve the intra-domain similarities of users, such that the users that are close in the source domain are still close in the target domain. Specifically, we first verify that the intra-domain similarities of user embeddings can represent domain knowledge well. Then, based on this finding, we propose the SimCDR framework, which adds the auxiliary task of predicting similar users to preserve the intra-domain similarities of the source domain user embeddings, i.e., the source domain knowledge. Besides being distinct in knowledge transfer, SimCDR also supports multiple source domains and embedding compression, making it more suitable for real-world applications. We conduct thorough experiments that verify the effectiveness of the proposed method.

Retrieval-Augmented Generation (RAG) utilizes external knowledge to augment Large Language Models’ (LLMs) reliability. For flexibility, agentic RAG employs autonomous, multi-round retrieval and reasoning to resolve queries. Although recent agentic RAG has improved via reinforcement learning, they often incur substantial token overhead from search and reasoning. This tradeoff prioritizes accuracy over efficiency. To address this issue, this work proposes TeaRAG, a T oken- e fficient a gentic RAG framework capable of compressing both retrieval content and reasoning steps. (1) First, the retrieved content is compressed by augmenting chunk-based semantic retrieval with a graph retrieval using concise triplets. A knowledge association graph is then built from semantic similarity and co-occurrence. Finally, Personalized PageRank is leveraged to highlight key knowledge within this graph, reducing the number of tokens per retrieval. (2) Besides, to reduce reasoning steps, Iterative Process-aware Direct Preference Optimization (IP-DPO) is proposed. Specifically, our reward function evaluates the knowledge sufficiency by a knowledge matching mechanism, while penalizing excessive reasoning steps. This design can produce high-quality preference-pair datasets, supporting iterative DPO to improve reasoning conciseness. Across six datasets, TeaRAG improves the average Exact Match by \(4\%\) and \(2\%\) while reducing output tokens by \(61\%\) and \(59\%\) on Llama3-8B-Instruct and Qwen2.5-14B-Instruct, respectively. Code is available at https://github.com/Applied-Machine-Learning-Lab/TeaRAG .

Recommendation systems have become ubiquitous tools in online platforms, providing personalized suggestions based on user–item interactions. To capture the dynamic higher-order connections between users and items, recommendation approaches based on dynamic graphs have garnered significant attention from researchers. However, existing recommendation methods based on dynamic graphs are often limited by data sparsity, which prevents them from achieving satisfactory performance. Fortunately, the rapid development of large language models (LLMs) with powerful text generation capabilities and extensive domain knowledge has offered new possibilities for addressing this challenge. However, how to effectively integrate LLMs with dynamic graphs remains unexplored. To bridge this gap, in this article, we propose a novel framework, that is, DynLLM, for applying LLMs to dynamic graph-based recommendation methods. Specifically, DynLLM harnesses the power of LLMs to generate multi-faceted user profiles based on the rich textual features of historical purchase records, which in turn supplement and enrich the underlying relationships between users and items. Along this line, to fuse the multi-faceted profiles with temporal graph embedding, we engage LLMs to derive corresponding profile embeddings and further employ a distilled attention mechanism to refine the LLM-generated profile embeddings for alleviating noisy signals, while also assessing and adjusting the relevance of each distilled facet embedding for seamless integration with temporal graph embedding from continuous time dynamic graphs (CTDGs). Extensive experiments on three real datasets have validated the superior improvements of DynLLM over a wide range of state-of-the-art baseline methods. The implementation code is available online at https://github.com/meteor-gif/DynLLM .

Talking face generation is a technique that synthesizes realistic, speech-synchronized facial animations from static images driven by audio or textual input. In recent years, notable advancements have been achieved in the generation of realistic lip-synchronized talking face videos. Although these methods have demonstrated highly realistic results in lip-sync generation, they rarely focus on the expression of emotions, which is essential for achieving emotionally rich and lifelike talking face videos. In this article, we propose a novel emotionally controllable audio-driven talking face generation framework, termed ECATFG. Specifically, the proposed ECATFG mainly consists of three modules: template video generator, landmark generator, and rendering module. Firstly, we integrate a customized motion transfer module into StyleGAN to generate template videos that encapsulate the target emotions and head movements. Subsequently, a Mamba-based landmark generator is proposed to accurately model the correlation between input audio and facial landmarks. Afterwards, in the rendering module, we employ a framework that combines both AdaIN and SPADE layers to transform audio-predicted landmarks into highly realistic images, guided by reference images and facial contour landmarks. Finally, comprehensive experiments validate the effectiveness of ECATFG, demonstrating its capability to generate high-quality talking face videos that not only achieve exceptional lip-synchronization accuracy but also effectively convey a wide range of emotions.

The autoregressive decoding paradigm endows large language models (LLMs) with superior language generation capabilities; however, its step-by-step decoding process inherently limits decoding speed. To mitigate these constraints, the prevalent “draft and validation” strategy enables parallel validation of candidate drafts, allowing LLMs to decode multiple tokens simultaneously during one model forward propagation. However, existing methodologies for obtaining drafts often incur additional overhead in communication or training process, or statistical biases from the corpus. To this end, we propose an innovative draft generation and maintenance approach that leverages the capabilities of LLM itself. Specifically, we extend the autoregressive decoding paradigm to a multi-branch drafting procedure, which can efficiently generate draft sequences without any additional models or training process, while preserving the quality of the generated content by maintaining LLM parameters. Experiments across various open-source benchmarks show that our method generates 2.0 to 3.2 tokens per forward step and achieves around 2 times improvement of end-to-end throughput compared to the autoregressive decoding strategy.

Complex question answering (QA) is a challenging task in artificial intelligence research which requires reasoning based on related knowledge. The retrieval-augmented generation (RAG) based on large language models (LLMs) have become one promising solution in QA. To facilitate RAG more effectively, the LLM needs to precisely evaluate knowledge required in QA. That is, first, the LLM needs to examine its knowledge boundary (what the LLM does not know) to retrieve external knowledge as supplement. Second, the LLM needs to evaluate the utility of the retrieved knowledge (whether it helps in reasoning) for robust RAG. To this end, in this paper, we propose a novel Question Answering with Knowledge Evaluation (KEQA) framework to promote the effectiveness and efficiency of RAG in QA. First, inspired by quizzes in classroom, we propose a quiz-based method to precisely examine the knowledge state of the uninterpretable LLM for QA. We ask indicative quizzes on each required knowledge, and inspect whether the LLM can consistently answer the quiz to examine its knowledge boundary. Second, we retrieve the unknown knowledge from external source, and evaluate its utility to pick the helpful ones for reasoning. We design a reasoning-based metric to evaluate utility, and construct a demonstration set in training data for reference to guide knowledge picking in inference. We conduct extensive experiments on four widely-used QA datasets, and the results demonstrate the effectiveness of the proposed method.

Large Language Models (LLMs) demonstrate remarkable capabilities, yet struggle with hallucination and outdated knowledge when tasked with complex knowledge reasoning, resulting in factually incorrect outputs. Previous studies have attempted to mitigate it by retrieving factual knowledge from large-scale knowledge graphs (KGs) to assist LLMs in logical reasoning and prediction of answers. However, this kind of approach often introduces noise and irrelevant data, especially in situations with extensive context from multiple knowledge aspects. In this way, LLM attention can be potentially mislead from question and relevant information. In our study, we introduce an Adaptive Multi-Aspect Retrieval-augmented over KGs (Amar) framework. This method retrieves knowledge including entities, relations, and subgraphs, and converts each piece of retrieved text into prompt embeddings. The Amar framework comprises two key sub-components: 1) a self-alignment module that aligns commonalities among entities, relations, and subgraphs to enhance retrieved text, thereby reducing noise interference; 2) a relevance gating module that employs a soft gate to learn the relevance score between question and multi-aspect retrieved data, to determine which information should be used to enhance LLMs' output, or even filtered altogether. Our method has achieved state-of-the-art performance on two common datasets, WebQSP and CWQ, showing a 1.9% improvement in accuracy over its best competitor and a 6.6% improvement in logical form generation over a method that directly uses retrieved text as context prompts. These results demonstrate the effectiveness of Amar in improving the reasoning of LLMs.

Cognitive diagnosis, which assesses the learners' competence from learners' interaction logs, plays a vital role in education. It provides a crucial reference for gauging learners' proficiency levels and tailoring future learning activities accordingly. Researchers have proposed numerous cognitive diagnosis models to address this task. Despite their success, these models continue to face the ill-posed problem because of the information loss caused by under-expressive interaction function and incomplete observations. In this paper, we address these challenges by proposing a novel cognitive diagnosis model, DMC-CDM, based on the theoretical premise that cognitive states can be captured with minimal information loss by maximizing the mutual information between observed and potential observations. Specifically, DMC-CDM incorporates a semantic extractor to provide a comprehensive semantic understanding of learners' interaction logs, thereby enhancing current collaborative-based cognitive state representations. It then consolidates multi-perspective observations to capture precise cognitive states by maximizing mutual information between these observations. We conducted extensive experiments on three datasets, and the experimental results demonstrate that our proposed model is both effective and beneficial for downstream applications in education.

Retrieval-augmented generation (RAG) is increasingly recognized as an effective approach to mitigating the hallucination of large language models (LLMs) through the integration of external knowledge. While numerous efforts, most studies focus on a single type of external knowledge source. However, in real-world applications, most situations involve diverse knowledge from various sources, yet this area has been less explored. The main dilemma is the lack of a suitable dataset containing multiple knowledge sources and pre-exploration of the associated issues. To address these challenges, we standardize a benchmark dataset that combines structured and unstructured knowledge across diverse and complementary domains. Based on this dataset, we further develop a plug-and-play RAG framework, PruningRAG, whose main characteristic is the use of multi-granularity pruning strategies to optimize the integration of relevant information while minimizing misleading context. It consistently improves performance across various existing RAG variants, demonstrating its robustness and broad applicability. Building upon the standardized dataset and PruningRAG, we also report a series of experimental results, as well as insightful findings. Our dataset and code are publicly available. https://github.com/USTCAGI/PruningRAG, with the aim of advancing future research in the RAG community.

Electrocardiogram (ECG) diagnosis aims to automatically classify ECG recordings into clinically meaningful categories, playing a vital role in medical decision-making. Deep learning methods, while promising, demand extensive annotated data and lack interpretability. Large Language Models (LLMs) offer potential in low-data scenarios and generating interpretable outputs, yet their application to ECG diagnosis, especially leveraging multimodal data (e.g., raw signals, derived features, and clinical knowledge), remains underexplored. To address these challenges, we propose ECG-Doctor, an interpretable and multimodal ECG diagnosis framework based on LLMs. ECG-Doctor comprises four key components: (1) ECG Knowledge Acquisition Module, which integrates external medical knowledge and Chain-of-Thought (CoT) reasoning to address the inability of LLMs to follow standardized ECG diagnostic procedures; (2) ECG Feature Extraction Module, which incorporates domain knowledge to overcome LLMs' limitations in comprehensively understanding structured ECG features; (3) ECG Waveform Analysis Module, which introduces time-series ECG models to equip LLMs with the capability to interpret and reason over raw ECG signal morphologies; (4) KNN-based ECG Retrieval Module, which retrieves the top-k most similar ECG samples and guides LLMs through in-context learning (ICL), enabling them to differentiate and learn from variations across ECGs. The outputs of these modules are aggregated and provided to the LLM as diagnostic context, enabling ICL to perform comprehensive ECG diagnosis. This design effectively simulates the diagnostic reasoning process of experienced electrocardiologists. Extensive experiments on the PTB-XL dataset demonstrate that ECG-Doctor is compatible with various LLMs and consistently outperforms existing baselines at both 100 Hz and 500 Hz sampling rates, showcasing its strong versatility and robustness. Furthermore, ECG-Doctor provides well-grounded diagnostic explanations, highlighting its superior interpretability.

In the quest for artificial general intelligence, Multi-modal Large Language Models (MLLMs) have emerged as a focal point in recent advancements. However, the predominant focus remains on developing their capabilities in static image understanding. The potential of MLLMs to process sequential visual data is still insufficiently explored, highlighting the lack of a comprehensive, high-quality assessment of their performance. In this paper, we introduce Video-MME, the first-ever full-spectrum, Multi-Modal Evaluation benchmark of MLLMs in Video analysis. Our work distinguishes from existing benchmarks through four key features: 1) Diversity in video types, spanning 6 primary visual domains with 30 subfields to ensure broad scenario generalizability; 2) Duration in temporal dimension, encompassing both short-, medium-, and long-term videos, ranging from 11 seconds to 1 hour, for robust contextual dynamics; 3) Breadth in data modalities, integrating multi-modal inputs besides video frames, including subtitles and audios, to unveil the all-round capabilities of MLLMs; 4) Quality in annotations, utilizing rigorous manual labeling by expert annotators to facilitate precise and reliable model assessment. With Video-MME, we extensively evaluate various state-of-the-art MLLMs, and reveal that Gemini 1.5 Pro is the best-performing commercial model, significantly outperforming the open-source models with an average accuracy of 75%, compared to 71.9% for GPT-4o. The results also demonstrate that Video-MME is a universal benchmark that applies to both image and video MLLMs. Further analysis indicates that subtitle and audio information could significantly enhance video understanding. Besides, a decline in MLLM performance is observed as video duration increases for all models. Our dataset along with these findings underscores the need for further improvements in handling longer sequences and multi-modal data, shedding light on future MLLM development. Project page: https://video-mme.github.io.

In this paper, we observe that the collaboration of various foundation models can perceive semantic and degraded information within images, thereby guiding the low-light enhancement process. Specifically, we propose a self-supervised low-light enhancement framework based on the multiple foundation models collaboration (dubbed FoCo), aimed at improving both the visual quality of enhanced images and the performance in high-level applications. At the feature level, FoCo leverages the rich features from various foundation models to enhance the model’s semantic perception during training, thereby reducing the gap between enhanced results and high-quality images from a high-level perspective. At the task level, we exploit the robustness-gap between strong foundation models and weak models, applying high-level task guidance to the low-light enhancement training process. Through the collaboration of multiple foundation models, the proposed framework shows better enhancement performance and adapts better to high-level tasks. Extensive experiments across various enhancement and application benchmarks demonstrate the qualitative and quantitative superiority of the proposed method over numerous state-of-the-art techniques.

Large language models (LLMs) rely on massive amounts of training data, however, the quantity of empirically observed data is limited.To alleviate this issue, lots of LLMs leverage synthetic data to enhance the quantity of training data.Despite significant advancements in LLMs, the efficiency and scalability characteristics of data synthesis during pre-training phases remain insufficiently explored.In this work, we propose a novel data synthesis framework, Cognitive Combination Synthesis (CCS), designed to achieve highly efficient and scalable data synthesis.Specifically, our methodology mimics human cognitive behaviors by recombining and interconnecting heterogeneous data from diverse sources thereby enhancing advanced reasoning capabilities in LLMs.Extensive experiments demonstrate that: (1) effective data organization is essential, and our mappingbased combination learning approach significantly improves data utilization efficiency; (2) by enhancing data diversity, accuracy, and complexity, our synthetic data scales beyond 100B tokens, revealing CCS's strong scalability.Our findings highlight the impact of data organization methods on LLM learning efficiency and the significant potential of scalable synthetic data to enhance model reasoning capabilities.

Few-Shot Fake News Detection (FS-FND) aims to distinguish inaccurate news from real ones in extremely lowresource scenarios. This task has garnered increased attention due to the widespread dissemination and harmful impact of fake news on social media. Large Language Models (LLMs) have demonstrated competitive performance with the help of their rich prior knowledge and excellent in-context learning abilities. However, existing methods face significant limitations, such as the Understanding Ambiguity and Information Scarcity, which significantly undermine the potential of LLMs. To address these shortcomings, we propose a Dual-perspective Knowledge-guided Fake News Detection (DKFND) model, designed to enhance LLMs from both inside and outside perspectives. Specifically, DKFND first identifies the knowledge concepts of each news article through a Detection Module. Subsequently, DKFND creatively designs an Investigation Module to retrieve inside and outside valuable information concerning to the current news, followed by another Judge Module to evaluate the relevance and confidence of them. Finally, a Determination Module further derives two respective predictions and obtain the final result. Extensive experiments on two public datasets show the efficacy of our proposed method, particularly in low-resource settings.

Recent advances in machine learning, particularly in Natural Language Processing (NLP), have produced powerful models trained on vast datasets. However, these models risk leaking sensitive information, raising privacy concerns. In response, regulatory measures such as the European Union's General Data Protection Regulation (GDPR) have driven increasing interest in Machine Unlearning techniques, which enable models to selectively forget specific data entries. Early unlearning approaches primarily relied on pre-processing methods, while more recent research has shifted towards training-based solutions. Despite their effectiveness, a key limitation persists: most methods require access to original training data, which is often unavailable. Additionally, directly applying unlearning techniques bears the cost of undermining the model's expressive capabilities. To address these challenges, we introduce the Iterative Contrastive Unlearning (ICU) framework, which consists of three core components: A Knowledge Unlearning Induction module designed to target specific knowledge for removal using an unlearning loss; A Contrastive Learning Enhancement module to preserve the model's expressive capabilities against the pure unlearning goal; And an Iterative Unlearning Refinement module that dynamically adjusts the unlearning process through ongoing evaluation and updates. Experimental results demonstrate the efficacy of our ICU method in unlearning sensitive information while maintaining the model's overall performance, offering a promising solution for privacy-conscious machine learning applications.

Zero-shot composed image retrieval (ZS-CIR) is a rapidly growing area with significant practical applications, allowing users to retrieve a target image by providing a reference image and a relative caption describing the desired modifications. Existing ZS-CIR methods often struggle to capture fine-grained changes and integrate visual and semantic information effectively. They primarily rely on either transforming the multimodal query into a single text using image-to-text models or employing large language models for target image description generation, approaches that often fail to capture complementary visual information and complete semantic context. To address these limitations, we propose a novel Fine-Grained Zero-Shot Composed Image Retrieval method with Complementary VisualSemantic Integration (CVSI). Specifically, CVSI leverages three key components: (1) Visual Information Extraction, which not only extracts global image features but also uses a pre-trained mapping network to convert the image into a pseudo token, combining it with the modification text and the objects most likely to be added. (2) Semantic Information Extraction, which involves using a pre-trained captioning model to generate multiple captions for the reference image, followed by leveraging an LLM to generate the modified captions and the objects most likely to be added. (3) Complementary Information Retrieval, which integrates information extracted from both the query and database images to retrieve the target image, enabling the system to efficiently handle retrieval queries in a variety of situations. Extensive experiments on three public datasets (e.g., CIRR, CIRCO, and FashionIQ) demonstrate that CVSI significantly outperforms existing state-of-the-art methods. Our code is available at https://github.com/yyc6631/CVSI.

Gaze estimation aims to infer a person’s gaze direction from images, which has wide applications in human-computer interaction. Recently, full-face-based gaze estimation methods have become increasingly prominent, owing to their efficiency and adaptability. However, irrelevant information in facial images limits estimation accuracy. To address this challenge, we propose DGRGaze, a difference-guided gaze estimation framework based on 6D rotation matrix representation. By incorporating an auxiliary task, namely predicting the gaze differences between facial image pairs, DGRGaze becomes more sensitive to gaze-related features. Additionally, a novel 6D rotation matrix representation is introduced to enhance the efficiency of network learning and resolve ambiguities in large-angle scenarios. Furthermore, we design a multi-task loss function based on the geodesic distance of the rotation matrix, enabling more precise quantification of gaze differences. Extensive experiments demonstrate our method achieves state-of-the-art performance on benchmark datasets, proving its effectiveness.

Function calling significantly extends the application boundary of large language models (LLMs), where high-quality and diverse training data is critical for unlocking this capability. However, collecting and annotating real function-calling data is challenging, while synthetic data from existing pipelines often lack coverage and accuracy. In this paper, we present ToolACE, an automatic agentic pipeline designed to generate accurate, complex, and diverse tool-learning data, specifically tailored to the capabilities of LLMs. ToolACE leverages a novel self-evolution synthesis process to curate a comprehensive API pool of 26,507 diverse APIs. Dialogs are further generated through the interplay among multiple agents, under the guidance of a complexity evaluator. To ensure data accuracy, we implement a dual-layer verification system combining rule-based and model-based checks. We demonstrate that models trained on our synthesized data---even with only 8B parameters---achieve state-of-the-art performance, comparable to the latest GPT-4 models. Our model and a subset of the data are publicly available at https://huggingface.co/Team-ACE.

The reasoning abilities are one of the most enigmatic and captivating aspects of large language models (LLMs). Numerous studies are dedicated to exploring and expanding the boundaries of this reasoning capability. However, tasks that embody both reasoning and recall characteristics are often overlooked. In this paper, we introduce such a novel task, code reasoning, to provide a new perspective for the reasoning abilities of LLMs. We summarize three meta-benchmarks based on established forms of logical reasoning, and instantiate these into eight specific benchmark tasks. Our testing on these benchmarks reveals that LLMs continue to struggle with identifying satisfactory reasoning pathways. Additionally, we present a new pathway exploration pipeline inspired by human intricate problem-solving methods. This Reflective Hypothesis Decomposition and Amendment (RHDA) pipeline consists of the following iterative steps: (1) Proposing potential hypotheses based on observations and decomposing them; (2) Utilizing tools to validate hypotheses and reflection outcomes; (3) Revising hypothesis in light of observations. Our approach effectively mitigates logical chain collapses arising from forgetting or hallucination issues in multi-step reasoning, resulting in performance gains of up to $3\times$. Finally, we expanded this pipeline by applying it to simulate complex household tasks in real-world scenarios, specifically in VirtualHome, enhancing the handling of failure cases. We release our code and all of results at https://github.com/TnTWoW/code_reasoning.

Remote Photoplethysmography (rPPG) is a non-contact technique for extracting physiological signals from facial videos. Recently, Transformer-based architectures have exhibited remarkable performance in rPPG estimation, owing to their excellent long-range spatial-temporal modeling capacities. However, challenges persist in applying Transformer-based rPPG methods, where quadratic computational costs and inadequate feature modeling diversity remain formidable. To address these challenges, we leverage the power of the Frequency Domain-based Vision Transformer and propose an end-to-end model PhysFFTFormer. Specifically, by integrating customized Frequency-Domain Spatiotemporal Self-Attention and Frequency-Domain Discriminative Feed-Forward modules, PhysFFTFormer efficiently captures spatial-temporal dependencies with reduced computational complexity. To extract rich and diverse feature representations, we design a dual-pathway architecture to utilize both raw and differential video frames. Furthermore, the Frequency-Domain Spatiotemporal Cross-Attention module is introduced to enhance information exchange and enable feature complementation between the two pathways. Extensive experiments on multiple benchmark datasets demonstrate PhysFFTFormer’s state-of-the-art performance, robustness, and potential for real-world non-contact health monitoring.

Although large language models (LLMs) show promise in solving complex mathematical tasks, existing evaluation paradigms rely solely on a coarse measure of overall answer accuracy, which are insufficient for assessing their authentic capabilities. In this paper, we propose CogMath, which comprehensively assesses LLMs’ mathematical abilities through the lens of human cognition. Specifically, inspired by psychological theories, CogMath formalizes human reasoning process into 3 stages: problem comprehension, problem solving, and solution summarization. Within these stages, we investigate perspectives such as numerical calculation, knowledge, and counterfactuals, and design a total of 9 fine-grained evaluation dimensions. In each dimension, we develop an “Inquiry-Judge-Reference” multi-agent system to generate inquiries that assess LLMs’ mastery from this dimension. An LLM is considered to truly master a problem only when excelling in all inquiries from the 9 dimensions. By applying CogMath on three benchmarks, we reveal that the mathematical capabilities of 7 mainstream LLMs are overestimated by 30%-40%. Moreover, we locate their strengths and weaknesses across specific stages/dimensions, offering in-depth insights to further enhance their reasoning abilities.

Recent advancements have significantly enhanced the performance of large language models (LLMs) in tackling complex reasoning tasks, achieving notable success in domains like mathematical and logical reasoning. However, these methods encounter challenges with complex planning tasks, primarily due to extended reasoning steps, diverse constraints, and the challenge of handling multiple distinct sub-tasks. To address these challenges, we propose HyperTree Planning (HTP), a novel reasoning paradigm that constructs hypertree-structured planning outlines for effective planning. The hypertree structure enables LLMs to engage in hierarchical thinking by flexibly employing the divide-and-conquer strategy, effectively breaking down intricate reasoning steps, accommodating diverse constraints, and managing multiple distinct sub-tasks in a well-organized manner. We further introduce an autonomous planning framework that completes the planning process by iteratively refining and expanding the hypertree-structured planning outlines. Experiments demonstrate the effectiveness of HTP, achieving state-of-the-art accuracy on the TravelPlanner benchmark with Gemini-1.5-Pro, resulting in a 3.6$\times$ performance improvement over o1-preview.

Owing to the capability of in-context learning, large language models (LLMs) have shown impressive performance across diverse mathematical reasoning benchmarks. However, we find that few-shot demonstrations can sometimes bring negative performance and their effectiveness on LLMs’ reasoning abilities remains unreliable. To this end, in this paper, we aim to theoretically analyze the impact of in-context demonstrations on LLMs’ reasoning performance. We prove that the reasoning efficacy (measured by empirical prediction loss) can be bounded by an LLM-oriented semantic similarity and an inference stability of demonstrations, which is general for both one-shot and few-shot scenarios. Based on this finding, we propose a straightforward, generalizable, and low-complexity demonstration selection method named LMS3. It facilitates to select the most pertinent samples for different LLMs and includes a novel demonstration rejection mechanism to automatically filter out samples that are unsuitable for few-shot learning. Through experiments on three representative benchmarks, two LLM backbones, and multiple few-shot settings, we verify that our LMS3 has superiority and achieves consistent improvements on all datasets, which existing methods have been unable to accomplish. Our code is available at https://github.com/Ljyustc/LMS3.

Click-Through Rate (CTR) prediction, a core task in recommendation systems, aims to estimate the probability of users clicking on items. Existing models predominantly follow a discriminative paradigm, which relies heavily on explicit interactions between raw ID embeddings. However, this paradigm inherently renders them susceptible to two critical issues: embedding dimensional collapse and information redundancy, stemming from the over-reliance on feature interactions over raw ID embeddings. To address these limitations, we propose a novel Supervised Feature Generation (SFG) framework, shifting the paradigm from discriminative "feature interaction" to generative "feature generation". Specifically, SFG comprises two key components: an Encoder that constructs hidden embeddings for each feature, and a Decoder tasked with regenerating the feature embeddings of all features from these hidden representations. Unlike existing generative approaches that adopt self-supervised losses, we introduce a supervised loss to utilize the supervised signal, i.e., click or not, in the CTR prediction task. This framework exhibits strong generalizability: it can be seamlessly integrated with most existing CTR models, reformulating them under the generative paradigm. Extensive experiments demonstrate that SFG consistently mitigates embedding collapse and reduces information redundancy, while yielding substantial performance gains across various datasets and base models. The code is available at https://github.com/USTC-StarTeam/GE4Rec.

The two-tower model has become prevalent in recommender systems for its computational efficiency and robust predictive capabilities. The model usually employs two independent neural networks to encode user and item data separately, and predicts the similarity score with inner product or cosine functions, depending on which the Top-k ranked item list is generated. The optimization process typically involves a multi-label classification objective, often guided by surrogate loss functions like Softmax and One-vs-All (OvA), to enhance the recommendation performance. Despite both Softmax and OvA losses being Bayes-consistent, empirical observations reveal a significant performance gap in evaluation metrics, suggesting limitations in Bayes-consistency for analyzing loss effectiveness. To address this, we introduce ℋ-consistency into the discussion, which provides non-asymptotic and hypothesis-specific guarantees for Top-k classification within the two-tower model's hypothesis space. Through theoretical analysis, we demonstrate that Softmax and Cosine Contrastive Loss exhibit ℋ-consistency, while the OvA loss does not, explaining the observed performance discrepancies. Our findings bridge the gap between theoretical properties and practical outcomes, offering deeper insights into the optimization of two-tower models and contributing to the development of more effective recommendation systems.

Sequential recommendation (SR), as an important branch of recommendation systems, has garnered significant attention due to its substantial commercial value. This has inspired some researchers to draw from the successful experiences of large language models to develop scaling laws for SR. However, the improvements brought by parameter expansion often reach a limit when the data scale is fixed. We have observed that existing deep learning sequence methods are typically seen as learning a unified pattern of user interactions, as they apply the same model for inference across different users, which often leads to the neglect of individual user behavior patterns. To address this, we propose conducting an independent analysis of each user's interaction sequence in SR. We initially developed the PCRec-simple, which uses KL divergence to perform a one-time optimization on each sequence after training, demonstrating that optimizing individual sequences can provide additional insights and overcome the performance bottleneck after scaling laws. Subsequently, we introduce PCRec, a sequential recommendation model that integrates real-time inference of hidden states into the model. It applies KL divergence optimization during the forward process, allowing for end-to-end optimization and addressing issues of robustness, parallelism, and optimization stability. Extensive experiments on real-world datasets show that PCRec significantly outperforms the current state-of-the-art methods. The code can be found at https://github.com/USTC-StarTeam/PCRec.

Click-through Rate (CTR) prediction is crucial for online personalization platforms. Recent advancements have shown that modeling rich user behaviors can significantly improve the performance of CTR prediction. Current long-term user behavior modeling algorithms predominantly follow two cascading stages. The first stage retrieves subsequence related to the target item from the long-term behavior sequence, while the second stage models the relationship between the subsequence and the target item. Despite significant progress, these methods have two critical flaws. First, the retrieval query typically includes only target item information, limiting the ability to capture the user's diverse interests. Second, relational information, such as sequential and interactive information within the subsequence, is frequently overlooked. Therefore, it requires to be further mined to more accurately model user interests.

Depression is an increasingly prevalent mental health issue worldwide, especially among the elderly, where effective early detection is crucial for timely intervention. While recent multimodal approaches demonstrate promise in leveraging visual and auditory cues for automatic depression recognition, existing methods often fail to extract fine-grained, depression-specific patterns from pre-extracted features and underutilize cross-modal interactions. To address these challenges, we propose DepFormer, a unified framework that incorporates a Bimodal Collaborative Transformer(BCT) for cross-modal representation learning and a personalized fusion module to enhance individual-specific modeling. The architecture comprises: (1) unimodal feature extraction, (2) bimodal collaborative representation learning via the BCT, (3) personalized feature fusion, and (4) final depression classification. Critically, the BCT employs symmetric bidirectional branches comprising an Audio-to-Video Transformer and a Video-to-Audio Transformer to enable mutual enhancement and complementary learning across modalities. Extensive experiments validate DepFormer's effectiveness, securing first place in the MPDD Challenge 2025 (Elderly Track), underscoring its strong practical potential.

Key-Value (KV) cache eviction---which retains the KV pairs of the most important tokens while discarding less important ones---is a critical technique for optimizing both memory usage and inference latency in large language models (LLMs). However, existing approaches often rely on simple heuristics---such as attention weights---to measure token importance, overlooking the spatial relationships between token value states in the vector space. This often leads to suboptimal token selections and thus performance degradation. To tackle this problem, we propose a novel method, namely **AnDPro** (**An**chor **D**irection **Pro**jection), which introduces a projection-based scoring function to more accurately measure token importance. Specifically, AnDPro operates in the space of value vectors and leverages the projections of these vectors onto an *``Anchor Direction''*---the direction of the pre-eviction output---to measure token importance and guide more accurate token selection. Experiments on $16$ datasets from the LongBench benchmark demonstrate that AnDPro can maintain $96.07\\%$ of the full cache accuracy using only $3.44\\%$ KV cache budget, reducing KV cache budget size by $46.0\\%$ without compromising quality compared to previous state-of-the-arts.

Time series forecasting plays a pivotal role in various real-world applications and has attracted significant attention in recent decades. While recent methods have achieved remarkable accuracy by incorporating advanced inductive biases and training strategies, we observe that instance-level variations remain a significant challenge. These variations—stemming from distribution shifts, missing data, and long-tail patterns—often lead to suboptimal forecasts for specific instances, even when overall performance appears strong. To address this issue, we propose a model-agnostic framework, PIR, designed to enhance forecasting performance through Post-forecasting Identification and Revision. Specifically, PIR first identifies biased forecast instances by estimating their predictive accuracy. Based on this, the framework revises the forecasts using contextual information, including covariates and historical time series, from both local and global perspectives in a post-processing fashion. Extensive experiments on real-world datasets with mainstream forecasting models demonstrate that PIR effectively mitigates instance-level errors and significantly improves forecasting reliability.

With the growing size of data and models in Large Recommendation Models, the time required for debugging has become increasingly prohibitive, underscoring the urgent need for effective guidance in parameter configuration. The Scaling Law (SL) offers analogous guidance in the Sequential Language domain, having achieved significant success by predicting model loss when scaling model size. However, the existing guidance from SL for Sequential Recommendation (SR) remains qualitative, which is because quantitative analysis of SL on SR encounters challenges with quality measurement on redundant sequences along with loss-performance discrepancy. In response, we introduce the Performance Law (P-Law) for SR models, which predicts model performance across various settings, intending to provide a quantitative framework for guiding the parameter optimization of future models. Initially, Performance Law utilizes Real Entropy to measure data quality, aiming to remove the low-quality influence of low-entropy redundant sequences. Subsequently, Performance Law investigates a fitting decay term, which facilitated the prediction of the major loss-performance discrepancy phenomena of overfitting, ultimately achieving quantitative performance prediction. Extensive experiment on various datasets demonstrates the effectiveness of Performance Law by displaying exceptional quantitative prediction ability against the original and modified qualitative SL. Additional application experiments on optimal parameter prediction and model expansion potential prediction also demonstrated the broad applicability of the Performance Law.

Despite their remarkable performance on various tasks, Large Language Models (LLMs) still struggle with logical reasoning, particularly in complex and multi-step reasoning processes. Among various efforts to enhance LLMs' reasoning capabilities, synthesizing large-scale, high-quality logical reasoning datasets has emerged as a promising direction. However, existing methods often rely on predefined templates for logical reasoning data generation, limiting their adaptability to real-world scenarios. To address the limitation, we propose **LogicTree**, a novel framework for efficiently synthesizing multi-step logical reasoning dataset that excels in both complexity and instantiation. By iteratively searching for applicable logic rules based on structural pattern matching to perform backward deduction, **LogicTree** constructs multi-step logic trees that capture complex reasoning patterns. Furthermore, we employ a two-stage LLM-based approach to instantiate various real-world scenarios for each logic tree, generating consistent real-world reasoning processes that carry contextual significance. This helps LLMs develop generalizable logical reasoning abilities across diverse scenarios rather than merely memorizing templates. Experiments on multiple benchmarks demonstrate that our approach achieves an average improvement of 9.4\% in accuracy on complex logical reasoning tasks.

In real-world scenarios, providing user queries with visually enhanced responses can considerably benefit understanding and memory, underscoring the great value of interleaved image-text generation. Despite recent progress, like the visual autoregressive model that unifies text and image processing in a single transformer architecture, generating high-quality interleaved content remains challenging. Moreover, evaluations of these interleaved sequences largely remain underexplored, with existing benchmarks often limited by unimodal metrics that inadequately assess the intricacies of combined image-text outputs. To address these issues, we present RAG-IGBench, a thorough benchmark designed specifically to evaluate the task of Interleaved Generation based on Retrieval-Augmented Generation (RAG-IG) in open-domain question answering. RAG-IG integrates multimodal large language models (MLLMs) with retrieval mechanisms, enabling the models to access external image-text information for generating coherent multimodal content. Distinct from previous datasets, RAG-IGBench draws on the latest publicly available content from social platforms and introduces innovative evaluation metrics that measure the quality of text and images, as well as their consistency. Through extensive experiments with state-of-the-art MLLMs (both open-source and proprietary) on RAG-IGBench, we provide an in-depth analysis examining the capabilities and limitations of these models. Additionally, we validate our evaluation metrics by demonstrating their high correlation with human assessments. Models fine-tuned on RAG-IGBench's training set exhibit improved performance across multiple benchmarks, confirming both the quality and practical utility of our dataset. Our benchmark is available at https://github.com/zry13/RAG-IGBench.

Click-through rate (CTR) prediction is a critical task in online advertising and recommender systems, relying on effective modeling of feature interactions.Explicit interactions capture predefined relationships, such as inner products, but often suffer from data sparsity, while implicit interactions excel at learning complex patterns through non-linear transformations but lack inductive biases for efficient low-order modeling.Existing two-stream architectures integrate these paradigms but face challenges such as limited information sharing, gradient imbalance, and difficulty preserving low-order signals in sparse CTR data.We propose a novel framework, Dynamic Low-Order-Aware Fusion (DLF), which addresses these limitations through two key components: a Residual-Aware Low-Order Interaction Network (RLI) and a Network-Aware Attention Fusion Module (NAF).RLI explicitly preserves low-order signals while mitigating redundancy from residual connections, and NAF dynamically integrates explicit and implicit representations at each layer, enhancing information sharing and alleviating gradient imbalance.Together, these innovations balance low-order and high-order interactions, improving model expressiveness.Extensive experiments on public datasets demonstrate that DLF achieves

In recommendation systems, the traditional multi-stage paradigm, which includes retrieval and ranking, often suffers from information loss between stages and diminishes performance. Recent advances in generative models, inspired by natural language processing, suggest the potential for unifying these stages to mitigate such loss. This paper presents the Unified Generative Recommendation Framework (UniGRF), a novel approach that integrates retrieval and ranking into a single generative model. By treating both stages as sequence generation tasks, UniGRF enables sufficient information sharing without additional computational costs, while remaining model-agnostic. To enhance inter-stage collaboration, UniGRF introduces a ranking-driven enhancer module that leverages the precision of the ranking stage to refine retrieval processes, creating an enhancement loop. Besides, a gradient-guided adaptive weighter is incorporated to dynamically balance the optimization of retrieval and ranking, ensuring synchronized performance improvements. Extensive experiments demonstrate that UniGRF significantly outperforms existing models on benchmark datasets, confirming its effectiveness in facilitating information transfer. Ablation studies and further experiments reveal that UniGRF not only promotes efficient collaboration between stages but also achieves synchronized optimization. UniGRF provides an effective, scalable, and compatible framework for generative recommendation systems.

For the advancement of time series classification, we can summarize that most existing methods adopt a common learning-to-classify paradigm - a classifier model tries to learn the relation between sequence inputs and target label encoded by one-hot distribution. Although effective, this paradigm conceals two inherent limitations: (1) one-hot distribution fails to reflect the comparability and similarity between labels, and (2) it is difficult to learn transferable representation across domains. In this work, we propose InstructTime, a novel attempt to reshape time series classification as a learning-to-generate paradigm. Relying on the generative capacity of the pre-trained language model, the core idea is to formulate the classification of time series as a multimodal understanding task. Specifically, firstly, a time series discretization module is designed to convert continuous inputs into a sequence of discrete tokens to solve the inconsistency issue across modality data. Secondly, we introduce an alignment projected layer before feeding the transformed token of time series into language models. Thirdly, prior to fine-tuning the language model for the target domain, it is essential to emphasize the necessity of auto-regressive pre-training across various modality inputs. Finally, extensive experimentation are conducted on several prevalent public benchmark datasets, indicating the superior performance of the InstructTime. Our code is at https://github.com/Mingyue-Cheng/InstructTime.

With the proliferation of images in online content, language-guided image retrieval (LGIR) has emerged as a research hotspot over the past decade, encompassing a variety of subtasks with diverse input forms. While the development of large multimodal models (LMMs) has significantly facilitated these tasks, existing approaches often address them in isolation, requiring the construction of separate systems for each task. This not only increases system complexity and maintenance costs, but also exacerbates challenges stemming from language ambiguity and complex image content, making it difficult for retrieval systems to provide accurate and reliable results. To this end, we propose ImageScope, a training-free, three-stage framework that leverages collective reasoning to unify LGIR tasks. The key insight behind the unification lies in the compositional nature of language, which transforms diverse LGIR tasks into a generalized text-to-image retrieval process, along with the reasoning of LMMs serving as a universal verification to refine the results. To be specific, in the first stage, we improve the robustness of the framework by synthesizing search intents across varying levels of semantic granularity using chain-of-thought (CoT) reasoning. In the second and third stages, we then reflect on retrieval results by verifying predicate propositions locally, and performing pairwise evaluations globally. Experiments conducted on six LGIR datasets demonstrate that ImageScope outperforms competitive baselines. Comprehensive evaluations and ablation studies further confirm the effectiveness of our design.

In the era of data-centric AI, the focus of recommender systems has shifted from model-centric innovations to data-centric approaches. The success of modern AI models is built on large-scale datasets, but this also results in significant training costs. Dataset distillation has emerged as a key solution, condensing large datasets to accelerate model training while preserving model performance. However, condensing discrete and sequentially correlated user-item interactions, particularly with extensive item sets, presents considerable challenges. This paper introduces TD3, a novel Tucker Decomposition based Dataset Distillation method within a meta-learning framework, designed for sequential recommendation. TD3 distills a fully expressive synthetic sequence summary from original data. To efficiently reduce computational complexity and extract refined latent patterns, Tucker decomposition decouples the summary into four factors: synthetic user latent factor, temporal dynamics latent factor, shared item latent factor, and a relation core that models their interconnections. Additionally, a surrogate objective in bi-level optimization is proposed to align feature spaces extracted from models trained on both original data and synthetic sequence summary beyond the naive performance matching approach. In the inner-loop, an augmentation technique allows the learner to closely fit the synthetic summary, ensuring an accurate update of it in the outer-loop. To accelerate the optimization process and address long dependencies, RaT-BPTT is employed for bi-level optimization. Experiments and analyses on multiple public datasets have confirmed the superiority and cross-architecture generalizability of the proposed designs. Codes are released at https://github.com/USTC-StarTeam/TD3.

The pursuit of model explainability has prompted the selective rationalization (aka, rationale extraction) which can identify important features (i.e., rationales) from the original input to support prediction results. Existing methods typically involve a cascaded approach with a selector responsible for extracting rationales from the input, followed by a predictor that makes predictions based on the selected rationales. However, these approaches often neglect the information contained in the non-rationales, underutilizing the input. Therefore, in our prior work, we introduce the Disentanglement-Augmented Rationale Extraction (DARE) method, which disentangles the input into rationale and non-rationale components, and enhances rationale representations by minimizing the mutual information between them. While DARE demonstrates strong performance in rationalization, it may still rely on shortcuts in the training distribution, leading to unfaithful rationales. To this end, in this paper, we propose Faith-DARE, an extension of DARE that aims to extract more reliable rationales by mitigating shortcut dependencies. Specifically, we treat the non-rationale features identified by DARE as environments that are decorrelated from the predictions. By shuffling and recombining these environments with rationales, we generate counterfactual samples and identify invariant rationales that remain predictive across shifted distributions. Extensive experiments on graph and textual datasets validate the effectiveness of Faith-DARE.

Evaluating AI systems, particularly large models, is an essential yet computationally expensive task. The use of extensive benchmarks often leads to substantial computational/human costs that may even exceed those of pretraining. The efficiency of AI model evaluation focuses on estimating the model's score on the full benchmark based on its responses to a smaller subset. Various empirical selection methods have been proposed to identify valuable subsets within these benchmarks. In this paper, we formally define and approximate the subset selection problem inherent in efficient evaluation. We prove that this problem actually optimizes a submodular function and that a unified subset can be identified using a simple greedy algorithm. Importantly, this approach is the first to provide theoretical guarantees of bias control and generalizability in score estimation. Using language models as a case study, experimental results across 11 different benchmarks validate its superiority in estimating model scores and maintaining ranking consistency. It can achieve accurate score estimation using no more than 30% of the full benchmark, thus facilitating efficient and sparse benchmark design.

As a critical psychological stress response, micro-expressions (MEs) are fleeting and subtle facial movements revealing genuine emotions. Automatic ME recognition (MER) holds valuable applications in fields such as criminal investigation and psychological diagnosis. The Facial Action Coding System (FACS) encodes expressions by identifying activations of specific facial action units (AUs), serving as a key reference for ME analysis. However, current MER methods typically limit AU utilization to defining regions of interest (ROIs) or relying on specific prior knowledge, often resulting in limited performance and poor generalization. To address this, we integrate the CLIP model's powerful cross-modal semantic alignment capability into MER and propose a novel approach namely MER-CLIP. Specifically, we convert AU labels into detailed textual descriptions of facial muscle movements, guiding fine-grained spatiotemporal ME learning by aligning visual dynamics and textual AU-based representations. Additionally, we introduce an Emotion Inference Module to capture the nuanced relationships between ME patterns and emotions with higher-level semantic understanding. To mitigate overfitting caused by the scarcity of ME data, we put forward LocalStaticFaceMix, an effective data augmentation strategy blending facial images to enhance facial diversity while preserving critical ME features. Finally, comprehensive experiments on four benchmark ME datasets confirm the superiority of MER-CLIP. Notably, UF1 scores on CAS(ME)$^{3}$reach 0.7832, 0.6544, and 0.4997 for 3-, 4-, and 7-class classification tasks, significantly outperforming previous methods.

Scene graph has long been treated as the basic tool to summarize the visual semantics from a structural perspective, which always confronts the challenge of capturing open entity and relation classes. Recently, some efforts have been made to enhance the open-vocabulary scene graph generation (OV-SGG) task via prompt tuning techniques. However, as only one prompt is utilized for all the classes, they may confuse the modeling for relations with similar spatial positions or semantics. To address these challenges, in this article, we propose a novel one-stage framework, named R elation-decoupled T ransformer framework with adaptive H ierarchical P rompt ( RTHP ), based on the vision-language model. Specifically, we first develop the dual entities/relations deformable attention module in the relation-decoupled transformer, which decouples relations into subject and object relations, and deploys on entity/relation queries separately. Along this line, we further design the Adaptive Hierarchical Prompt (AHPro) module to model the inherent hierarchical structure of relation classes, which enables categories situated in varying positions to have different prompts. In this way, confusing categories can be easily distinguished by their respective positions in the hierarchical structure. Extensive experimental results demonstrate that our RTHP framework achieves competitive performance for OV-SGG, validating its effectiveness on base classes and generalization capability on novel classes.

Modeling and process optimization of Multi-stage Continuous Production System (MCPS) is an important research topic in the field of intelligent manufacturing today. However, due to the inherent properties of MCPS, existing researches face difficulties in (1) coupling optimization across multiple stages under diverse constraints and objectives, and (2) accurately mapping controllable process variables to critical production outputs. In this work, we propose a novel multi-objective optimization method for MCPS based on feature fusion modeling, with main innovations including: 1) For MCPS modeling, we propose a Transformer network structure with a parallel dual-branch attention mechanism for linking process variables to production outputs. A multi-stage feature fusion prediction model based on DenseNet is developed, which not only improves prediction accuracy but also provides intermediate stage prediction results. 2) For process optimization of MCPS, we design a multi-constraint and multi-objective optimization model. Subsequently, we propose a dynamic multi-objective optimization algorithm framework to enhance the performance of the algorithm and improve the quality of solutions. Furthermore, we conducted experiments with a real Coke-Chem integrated production dataset, and the results show that our predictive model achieves an average MAE of 0.11, MSE of 0.04, and RMSE of 0.20, outperforming state-of-the-art methods and boosts solution coverage (up to 80%) and hypervolume (up to 2.5×) when integrated with classical algorithms like NSGA-II and GDE3.

Knowledge combination prediction involves analyzing current knowledge elements and their relationships, then forecasting how these elements, drawn from various fields, can be creatively combined to form new, innovative solutions. This process is critical for countries and businesses to understand future technology trends and promote innovation in an era of rapid scientific and technological advancement. Existing methods often overlook the integration of knowledge combinations from multiple views, along with their inherent heterophily and the dual “many-to-one” property, where a single knowledge combination can include multiple elements, and a single element may belong to various combinations. To this end, we propose a novel framework named Multi-viewHeterogeneousHyperGNN forHeterophilicKnowledgeCombinationPrediction (H3KCP). Specifically, H3KCP first constructs a hypergraph reflecting the dual “many-to-one” property of knowledge combinations, where each hyperedge may contain several nodes and each node can also belong to multiple hyperedges. Next, the framework employs a multi-view fusion approach to model knowledge combinations, considering heterophily and integrating insights from co-occurrence, co-citation, and hierarchical structure-based views. Furthermore, our analysis of H3KCP from a spectral graph perspective offers insights into its rationality. Finally, extensive experiments on real-world patent datasets and the Open Academic Graph dataset validate the effectiveness and efficiency of our approach, yielding significant insights into knowledge combinations. Our code and dataset are publicly available onhttps://github.com/huijieliu2023/H3KCP.

Predicting the scientific impact of research scholars is increasingly crucial for career planning, particularly for young scholars considering career transitions. However, predicting a scholar's future development, especially after they move to a different academic group, presents significant challenges. To tackle this issue, we propose a Future Publication Impact Prediction Network (FPIPN) based on graph neural networks. FPIPN leverages rich information from a heterogeneous academic graph for impact prediction. We employ a hierarchical attention mechanism to learn the significance of graph information and utilize a knowledge distillation strategy to assess future impact based on historical records. Extensive experiments on a real-world academic dataset showcase the effectiveness of our approach compared to state-of-the-art methods.

Due to cultural influences and the long history of language evolution, metaphorical expressions are pervasive in Chinese texts, particularly in ironical comments, poetry, and various literary works. Traditional Chinese text classification methods relying on surface-level features often fail to bridge the gap between literal features and label spaces in such texts. To address this problem, we propose L-MAM, a novel label-constrained contrastive learning approach for Chinese text classification based on large language models. Specifically, we first introduce an ambiguity recognition module (ARM) to quantify phonetic and syntactic ambiguities in Chinese characters, identifying expressions that benefit most from metaphorical associations. By leveraging large language model-driven prompts, metaphorical interpretations are extracted from texts, and classification labels are aligned with semantic definitions for deeper contextual understanding. Then, we design a contextual attention mechanism that dynamically adjusts weights based on character ambiguity, and a metaphorical attention mechanism that aligns metaphorical embeddings with label semantics for refined label-constrained associations. Additionally, we devise a contextual-metaphorical contrastive learning mechanism, which employs a triplet loss to differentiate literal and metaphorical aspects while enhancing interclass separability. Finally, we conduct extensive experiments on three real-world datasets, where the experimental results validate the effectiveness of L-MAM, offering new insights into Chinese text classification involving metaphorical comprehension.

In this work, we observe that the generators, which are pre-trained on massive natural images, inherently hold the promising potential for superior low-light image enhancement against varying scenarios. Specifically, for the low-light image enhancement process of a single image, we introduce the pre-trained generators to restore the details and colors degraded by low-light conditions, thereby improving the visual effect. Taking one step further, we introduce a novel optimization strategy, which backpropagates the gradients to the input seeds rather than the parameters of the low-light image enhancement model, thus intactly retaining the generative knowledge learned from natural images and achieving faster convergence speed. Benefiting from the pre-trained knowledge and seed-optimization strategy, the low-light image enhancement model can significantly regularize the visibility and fidelity of the enhanced result, thus rapidly generating high-quality images without training on any low-light dataset. Extensive experiments on various benchmarks demonstrate the effectiveness of the proposed method, showing its potential advantages over numerous state-of-the-art methods both qualitatively and quantitatively.

Recently, audiences have increasingly watched diverse dramas and movies on streaming platforms, prompting platform administrators to enhance their understanding of video semantics. In particular, capturing and interpreting social relations among characters is critical for content-driven intelligent services and enhancing user experience. However, most existing research has solely approached social relations recognition as a classification problem, without justifiable interpretation of prediction results within the video context. To address this issue, we study cognitive science research and the Chain-of-Thought (CoT) strategy. Based on these foundations, we propose SaMo-CoT, an approach that leverages large language models (LLMs) in step-by-step reasoning. This approach simulates human cognitive processes by combining social scenarios in videos and incorporating empirical social interaction knowledge. In this way, we enable verbal rationales for determining social relations in video understanding. Furthermore, we present an innovative doubly-right social relation recognition framework that predicts both correct social relation labels and correct scenario rationales. Specifically, we translate verbal CoT into multimodal CoT by leveraging scenario-aware prompts and contrastive learning. Extensive experiments demonstrate significant improvements in classification accuracy and interpretability compared to traditional approaches.

Knowledge discovery and utilization are two essential cognitive processes that enable humans to understand the world and extract new insights from their surroundings. These processes have motivated machine learning studies, particularly zero-shot (ZS) learning, which seeks to identify unseen concepts through the use of side information. Previous ZS studies primarily focused on utilizing existing knowledge to infer unseen events, yet they overlook the crucial process of knowledge discovery and the integrated modeling of these knowledge-aware processes. In this study, we present a comprehensive ZS learning approach that explores and evaluates the machine’s abilities of discovering and utilizing knowledge. More specifically, to emulate human-like knowledge discovery and utilization processes, we propose a novel visual-aware ZS knowledge graph completion task for evaluation, incorporating a traditional ZS image classification task. Technically, we develop a unified ZS learning paradigm named Cognitive Learner (CoLa) to foster the two knowledge-aware abilities. Including a knowledge representation learning (KRL) module and a knowledge adaptation (KA) module, CoLa adapts well to the two specified tasks with the corresponding data. Extensive experiments on large-scale datasets demonstrate CoLa models’ outstanding performance over compared methods in the two ZS tasks, illustrating their superior ability of discovering and utilizing knowledge.

Math word problem (MWP) serves as a critical milestone for assessing the text mining ability and knowledge mastery level of models. Recent advancements have witnessed large language models (LLMs) showcasing remarkable performance on MWP. However, current LLMs still frequently exhibit logical errors, which highlights their inability to fully grasp the knowledge required for genuine step-by-step mathematical reasoning. To this end, in this paper, we propose a novel Knowledge-guided Solver (KNOS) framework that empowers LLMs to simulate human mathematical reasoning, whose core idea is toInvoke-Verify-Injectnecessary knowledge to solve MWP. We draw inspiration from the dual-process theory to construct two cooperative systems: aKnowledge Systemand anInference System. Specifically, theKnowledge Systememploys LLMs as the knowledge base and develops a novelknowledge invokerthat can elicit their relevant knowledge to support the strict step-level mathematical reasoning. In theInference System, we propose aknowledge verifierand aknowledge injectorto evaluate the knowledge rationality and further guide the step-wise symbolic deduction in an interpretable manner based on human cognitive mechanism, respectively. Moreover, to tackle the potential scarcity issue of mathematics-specific knowledge in LLMs, we consider an open-book exam scenario and propose an improved version of KNOS called EKNOS. In EKNOS, we meticulously designknowledge selectorsto extract the most relevant commonsense and math formulas from external knowledge sources for each reasoning step. This knowledge is utilized to assist theknowledge invokerin better stimulating LLMs’ reasoning abilities. Both KNOS and EKNOS are flexible to empower different LLMs. Our experiments with GPT3, ChatGPT, and GPT4 not only demonstrate their reasoning accuracy improvement but also show how they bring the strict step-wise interpretability of mathematical thinking.

Knowledge Tracing (KT) predicts future performance by modeling students' historical interactions, and understanding students' affective states can enhance the effectiveness of KT, thereby improving the quality of education. Although traditional KT values students' cognition and learning behaviors, efficient evaluation of students' affective states and their application in KT still require further exploration due to the non-affect-oriented nature of the data and budget constraints. To address this issue, we propose a computation-driven approach,DynamicAffectSimulationKnowledgeTracing (DASKT), to explore the impact of various student affective states (such as frustration, concentration, boredom, and confusion) on their knowledge states. In this model, we first extract affective factors from students' non-affect-oriented behavioral data, then use clustering and spatiotemporal sequence modeling to accurately simulate students' dynamic affect changes when dealing with different problems. Subsequently, we incorporate affect with time-series analysis to improve the model's ability to infer knowledge states over time and space. Extensive experimental results on two public real-world educational datasets show that DASKT can achieve more reasonable knowledge states under the effect of students' affective states. Moreover, DASKT outperforms the most advanced KT methods in predicting student performance. Our research highlights a promising avenue for future KT studies, focusing on achieving high interpretability and accuracy.

Learning-based video denoisers have attained state-of-the-art (SOTA) performances on public evaluation benchmarks. Nevertheless, they typically encounter significant performance drops when applied to unseen real-world data, owing to inherent data discrepancies. To address this problem, this work delves into the model pretraining techniques and proposes masked central frame modeling (MCFM), a new video pretraining approach that significantly improves the generalization ability of the denoiser. This proposal stems from a key observation: pretraining denoiser by reconstructing intact videos from the corrupted sequences, where the central frames are masked at a suitable probability, contributes to achieving superior performance on real-world data. Building upon MCFM, we introduce a robust video denoiser, named MVDenoiser, which is firstly pretrained on massive available ordinary videos for general video modeling, and then finetuned on costful real-world noisy/clean video pairs for noisy-to-clean mapping. Additionally, beyond the denoising model, we further establish a new paired real-world noisy video dataset (RNVD) to facilitate cross-dataset evaluation of generalization ability. Extensive experiments conducted across different datasets demonstrate that the proposed method achieves superior performance compared to existing methods. Code and dataset are available athttps://github.com/mxxx99/MVDenoiser.

Open vocabulary 3D instance segmentation aims to align 3D instance segmentation results with natural language text, thereby achieving semantic prediction without relying on predefined class labels for specific scenes, which has been widely used in the field of multimedia. Current open vocabulary 3D instance segmentation methods mainly rely on 2D masks provided by various 2D segmentation foundation models. However, in complex scenes, the calculation of 2D masks often struggles to balance over-segmentation of large objects and under-segmentation of small objects. In this paper, we introduce OV-BIS, a novel zero-shot open vocabulary 3D instance segmentation method that leverages instance boundary information to improve 3D semantic segmentation performance. The key insight of our method is that the edge map as 3D boundary projection is suitable for multi-scale tasks and capable of compensating for the weakness of 2D masks in multi-scale adaptability for complex scenes. Our method aggregates multiview edge maps and 2D masks, iteratively guiding the merging of over-segmented point clouds with regions growing to cluster 3D primitives into distinct 3D instances. By projecting 3D instances onto images and using CLIP to calculate semantic features from multiple perspectives with an outliers filter, 3D semantic instance segmentation has been achieved. Experiments on multiple datasets demonstrate the superiority of our method.

Code summarization is designed to generate descriptive natural language for code snippets, facilitating understanding and increasing productivity for developers. Previous research often overlooks the semantic connection between code and its natural language description, resulting in a noticeable gap and suboptimal solution. To address this issue, we introduce a semantic-aligned code summarization framework that leverages crucial data flow information from code for semantic analysis, ensuring alignment between code and summaries. Specifically, we utilize a semantic extraction module (SEM) to decipher the meaning of code and align it with natural language through a semantic alignment module. In the SEM, we construct a code graph that includes data flow edges using static program analysis techniques. Then, on this well-constructed code graph, we innovatively adopt a walking algorithm guided by data flow to extract the semantics of the code. This walking algorithm understands code semantics by analyzing the information transfer between variables during the program execution process. In the semantic alignment module, we integrate a contrastive learning loss mechanism for semantic alignment, which cohesively maps the semantic domains of code and natural language into a unified vector space. We further theoretically analyzed that the data-flow-guided walking algorithm can ensure capturing semantically highly related nodes in shorter paths. Extensive experiments on two benchmark datasets demonstrate the efficacy and broad applicability of the framework.

Retrieval-augmented generation (RAG) is a technique that enhances the capabilities of large language models (LLMs) by incorporating external knowledge sources. This method addresses common LLM limitations, including outdated information and the tendency to produce inaccurate “hallucinated” content. However, evaluating RAG systems is a challenge. Most benchmarks focus primarily on question-answering applications, neglecting other potential scenarios where RAG could be beneficial. Accordingly, in the experiments, these benchmarks often assess only the LLM components of the RAG pipeline or the retriever in knowledge-intensive scenarios, overlooking the impact of external knowledge base construction and the retrieval component on the entire RAG pipeline in non-knowledge-intensive scenarios. To address these issues, this article constructs a large-scale and more comprehensive benchmark and evaluates all the components of RAG systems in various RAG application scenarios. Specifically, we refer to the CRUD actions that describe interactions between users and knowledge bases and also categorize the range of RAG applications into four distinct types—create, read, update, and delete (CRUD). “Create” refers to scenarios requiring the generation of original, varied content. “Read” involves responding to intricate questions in knowledge-intensive situations. “Update” focuses on revising and rectifying inaccuracies or inconsistencies in pre-existing texts. “Delete” pertains to the task of summarizing extensive texts into more concise forms. For each of these CRUD categories, we have developed different datasets to evaluate the performance of RAG systems. We also analyze the effects of various components of the RAG system, such as the retriever, context length, knowledge base construction, and LLM. Finally, we provide useful insights for optimizing the RAG technology for different scenarios. The source code is available at GitHub: https://github.com/IAAR-Shanghai/CRUD_RAG .

Programmatic Weak Supervision (PWS) is a recent data labeling paradigm, which employs several Labeling Functions (LFs) to provide weak labels and involves a Label Model (LM) for label aggregation. Despite the significant progress, there still remain some inherent challenges in PWS. From the view of labeling, LFs may wrongly label some data points. From the view of data, some data points themselves may be low-quality (e.g., ambiguous texts or blurred images). These largely stem from the lack of an explicit evaluation mechanism for LFs or data points. To this end, inspired by confident learning focusing on label quality, we propose a Confident PWS (CPWS) approach for high-quality data labeling. Specifically, several LFs are firstly utilized to provide weak labels for unlabeled data. Then, we develop an explicit Dual Evaluation Mechanism (DEM) to evaluate the quality of both LFs and data points, which not only employs data to evaluate trained models but also leverages trained models to evaluate data. Along this line, we further design a Distribution-Guided Pruning Strategy (DPS) to prune low-quality data and aggregate weak labels under the guidance of label class distribution. Extensive experiments on various benchmark datasets demonstrate the effectiveness and generalization ability of our proposed approach.

Recently, the dual-target cross-domain recommendation has been an emerging research problem, which aims to improve the performances of both source and target domains by transferring the preferences of overlapping users. Most of the existing work adopted a coarse-grained manner to detach general users’ preferences and associate them with domain-specific information for enhancing user representation learning, which fails to depict the differences in users’ diverse preferences and aggregate relevant preferences with improper propagation. To this end, in this article, we propose a multi-factor user representation pre-training framework, dubbed MF-GSLAE, with a focus on fine-grained preference learning and transferring. Specifically, we first propose a fine-grained factor representation pre-training paradigm. It projects the behavior records of both domains into several subspaces and introduces a compactness regularization to generate multiple fine-grained preference factors. Furthermore, we propose a multi-factor graph structure learning method within linear complexity to efficiently construct preference connections on different scales of users, which could aggregate the intrinsic relationship of user preferences in immediate embedding spaces to capture high-order information. Following the pre-training, we subsequently design a factor selection module with the bootstrapping mechanism to adaptively choose the corresponding domain-related preferences and transfer domain-shared information through partial overlapping factors for addressing the negative transfer problem. Finally, the optimization objectives of both domains are formalized in a multi-task learning framework and derive the learned user representation in an end-to-end training manner. Extensive experimental results on several publicly available datasets have not only demonstrated the effectiveness of the learned user representations with the comparison of state-of-the-art baselines but also indicated the interpretability and robustness. The code of our work is publicly available at https://github.com/USTC-StarTeam/MF-GSLAE .

The application of contrastive learning (CL) to collaborative filtering (CF) in recommender systems has achieved remarkable success. CL-based recommendation models mainly focus on creating multiple augmented views by employing different graph augmentation methods and utilizing these views for self-supervised learning. However, current CL methods for recommender systems usually struggle to fully address the problem of noisy data. To address this problem, we propose the G raph A ugmentation E mpowered C ontrastive L earning (GAECL) for recommendation framework, which uses graph augmentation based on topological and semantic dual adaptation and global co-modeling via structural optimization to co-create contrasting views for better augmentation of the CF paradigm. Specifically, we strictly filter out unimportant topologies by reconstructing the adjacency matrix and mask unimportant attributes in nodes according to the PageRank centrality principle to generate an augmented view that filters out noisy data. Additionally, GAECL achieves global collaborative modeling through structural optimization and generates another augmented view based on the PageRank centrality principle. This helps to filter the noisy data while preserving the original semantics of the data for more effective data augmentation. Extensive experiments are conducted on five datasets to demonstrate the superior performance of our model over various recommendation models.

A well-developed recommendation system can not only leverage multi-typed interactions (such as page view , add-to-cart , and purchase ) to better identify user preferences but also demonstrate high performance, low complexity, and strong interpretability. However, many existing solutions for multi-behavior recommendation fall short of intuitive modeling of real-world scenarios, leading to overly complex models with massive parameters and cumbersome components. In particular, they share two critical limitations: (1) Some pioneering models are built upon the strict assumption of cascade effects across behaviors, which contradicts multifarious behavior paths in practical applications. (2) Existing approaches fail to explicitly capture the unique idiosyncrasies of users and even neglect the inherent nature of items involved in the multi-behavior interactions. To this end, we propose a novel Directed Acyclic Graph Convolutional Network (DA-GCN) for the multi-behavior recommendation task. Specifically, we pinpoint the partial order relations within the monotonic behavior chain and extend it to personalized directed acyclic behavior graphs to exploit behavior dependencies. Then, a GCN-based directed edge encoder is employed to distill rich collaborative signals embodied by each directed edge. In light of the information flows over the directed acyclic structure, we propose an attentive aggregation module to gather messages from all potential antecedent behaviors, representing distinct perspectives to understand the terminated behavior. Thus, we obtain comprehensive representations for the follow-up behavior through learnable distributions over its preceding behaviors, explicitly reflecting personalized interactive patterns of users and underlying properties of items simultaneously. Finally, we design a customized multi-task learning objective for flexible joint optimization. Extensive experiments on public benchmarking datasets fully demonstrate the superiority of DA-GCN with significant performance improvement and computational efficiency over a wide range of state-of-the-art methods. Our code is available at https://github.com/xizhu1022/DA-GCN .

Collaborative filtering (CF) works have demonstrated the robust capabilities of Shallow Autoencoders on implicit feedback, showcasing highly competitive performance with other reasonable approaches (e.g., iALS and VAE-CF). However, despite their dual advantages of high performance and simple construction, EASE still exhibits several major shortcomings that must be addressed. To be more precise, the scalability of EASE is limited by the number of items, which determines the storage and inversion cost of a large dense matrix; the square-loss optimization objective does not consistently meet the recommendation task’s requirement for predicting personalized rankings, resulting in suboptimal outcomes; the regularization coefficients are sensitive and require recalibration with different datasets, leading to an exhaustive and time-consuming fine-tuning process. In order to address these obstacles, we propose a novel approach called Similarity-Structure Aware Shallow Autoencoder (AutoS \(^2\) AE) that aims to enhance both recommendation accuracy and model efficiency. Our method introduces three similarity structures: Co-occurrence, KNN, and NSW graphs, which replace the large dense matrix in EASE with a sparse structure, thus facilitating model compression. Additionally, we optimize the model by incorporating a low-rank training component into the matrix and applying a weighted square loss for improved ranking-oriented approximations. To automatically tune the hyperparameters, we further design two validation losses on the validation set for guidance and update the hyperparameters using the gradients of these validation losses. Both theoretical analyses regarding the introduction of similarity structures and empirical evaluations on multiple real-world datasets demonstrate the effectiveness of our proposed method, which significantly outperforms competing baselines.

Large Language Models (LLMs) have revolutionized natural language processing tasks, demonstrating their exceptional capabilities in various domains. However, their potential for graph semantic mining in job recommendations remains largely unexplored. This paper focuses on unveiling the capability of large language models in understanding behavior graphs and leveraging this understanding to enhance recommendations in online recruitment, including promoting out-of-distribution (OOD) applications. We present a novel framework that harnesses the rich contextual information and semantic representations provided by large language models to analyze behavior graphs and uncover underlying patterns and relationships. Specifically, we propose a meta-path prompt constructor that aids LLM recommender in grasping the semantics of behavior graphs for the first time and design a corresponding path augmentation module to alleviate the prompt bias introduced by path-based sequence input. By facilitating this capability, our framework enables personalized and accurate job recommendations for individual users. We evaluate the effectiveness of our approach on comprehensive real-world datasets and demonstrate its ability to improve the relevance and quality of recommended results. This research not only sheds light on the untapped potential of large language models but also provides valuable insights for developing advanced recommendation systems in the recruitment market. The findings contribute to the growing field of natural language processing and offer practical implications for enhancing job search experiences.

We study the problem of federated contextual combinatorial cascading bandits, where agents collaborate under the coordination of a central server to provide tailored recommendations to users. Existing works consider either a synchronous framework, necessitating full agent participation and global synchronization, or assume user homogeneity with identical behaviors. We overcome these limitations by considering (1) federated agents operating in an asynchronous communication paradigm, where no mandatory synchronization is required and all agents communicate independently with the server, (2) heterogeneous user behaviors, where users can be stratified into latent user clusters, each exhibiting distinct preferences. For this setting, we propose a UCB-type algorithm with delicate communication protocols. Through theoretical analysis, we give sub-linear regret bounds on par with those achieved in the synchronous framework, while incurring only logarithmic communication costs. Empirical evaluation on synthetic and real-world datasets validates our algorithm's superior performance in terms of regrets and communication costs.

Visualization recommendations, which aim to automatically match proper visual charts for specific data tables, can significantly simplify the data analysis process.Traditional approaches in this domain have primarily relied on rule-based or machine learning-based methodologies.These methods often demand extensive manual maintenance and yet fail to fully comprehend the tabular data, leading to unsatisfactory performance.Recently, Large Language Models (LLMs) have emerged as powerful tools, exhibiting strong reasoning capabilities.This advancement suggests their substantial promise in addressing visualization recommendation challenges.However, effectively harnessing LLMs to discern and rationalize patterns in tabular data, and consequently deduce the essential information for chart generation, remains an unresolved challenge.To this end, we introduce a novel Hierarchical Table Prompt-based reprogramming framework, named HTP.This framework aims to integrate multi-dimensional tabular data into LLMs through a strategically crafted prompt learning method while keeping the LLMs' backbone and weights unaltered.The HTP framework uniquely incorporates a four-level prompt structure, encompassing general, instance, cluster, and column levels.This multi-level approach is engineered to provide a comprehensive understanding of both general distribution and multifaceted fine-grained features of tabular data, before inputting the tabular data into the frozen LLM.Our empirical studies confirm that the HTP framework achieves state-of-the-art performance, marking an advancement in the field of data visualization and analysis.

Multimodal Entity Linking (MEL) is a crucial task that aims at linking ambiguous mentions within multimodal contexts to the referent entities in a multimodal knowledge base, such as Wikipedia. Existing methods focus heavily on using complex mechanisms and extensive model tuning methods to model the multimodal interaction on specific datasets. However, these methods overcomplicate the MEL task and overlook the visual semantic information, which makes them costly and hard to scale. Moreover, these methods cannot solve the issues like textual ambiguity, redundancy, and noisy images, which severely degrade their performance. Fortunately, the advent of Large Language Models (LLMs) with robust capabilities in text understanding and reasoning, particularly Multimodal Large Language Models (MLLMs) that can process multimodal inputs, provides new insights into addressing this challenge. However, how to design a universally applicable LLMs-based MEL approach remains a pressing challenge. To this end, we propose UniMEL, a unified framework which establishes a new paradigm to process multimodal entity linking tasks using LLMs. In this framework, we employ LLMs to augment the representation of mentions and entities individually by integrating textual and visual information and refining textual information. Subsequently, we employ the embedding-based method for retrieving and re-ranking candidate entities. Then, with only ~0.26% of the model parameters fine-tuned, LLMs can make the final selection from the candidate entities. Extensive experiments on three public benchmark datasets demonstrate that our solution achieves state-of-the-art performance, and ablation studies verify the effectiveness of all modules. Our code is available at https://github.com/Javkonline/UniMEL.

Zero-Shot Relational Learning (ZSRL), strives to predict relations that have not been observed during training, presenting a considerable challenge in terms of model generalization. Existing ZSRL methods usually utilize the prior knowledge of labels (e.g., text description, ontological schema) to enable knowledge transfer by learned features. Nonetheless, these methods remain limited to calculating the surface features exhibited by relations, failing to fully explore their underlying driving factors. This leads to insufficient discrimination between the shared and distinctive inherent components among relations, which consequently impedes the cognitive understanding required for advanced reasoning. In our study, we aim to identify and utilize shared factors that widely exist in the prior knowledge of classes to learn enhanced semantic representations via shared factors composition, and develop our Factor-based ZSRL framework (FZR) with Generative Adversarial Networks (GANs) to bridge inequality between seen and unseen classes. FZR is designed to restructure the semantic space in such a way that it captures the essence of relation formation, thereby facilitating superior knowledge transfer in zero-shot scenarios. We conduct extensive experiments and evaluate our model on real-world datasets, and the results clearly demonstrate the effectiveness of the proposed model in zero-shot relational learning tasks.

Knowledge Tracing (KT) is a crucial research task for dynamically monitoring students' knowledge states, particularly in online education systems. Recently, knowledge tracing has gained significant attention and in-depth research. Most existing methods rely on students' response data for question understanding and modeling, which helps better updating students' knowledge states. Meanwhile, question ID is utilized to indicate and represent questions. However, this presents a challenge when transitioning to new, cold-start questions that few students has answered before. Also, prior work has overlooked the semantic modeling of questions, which could better assist in modeling the transfer of students' knowledge states. In this paper, we explore leveraging the power of Large Language Models (LLMs) to help understand questions for knowledge tracing, which benefits mitigating cold-start and sparse problems and modeling the transfer of students' knowledge states in a sophisticated manner. Specifically, we first design an attribute estimation module to estimate the attribute of the questions (e.g., difficulty, ability requirements, expected response time) by prompting Large Language Models. Subsequently, we have developed a question embedding module that incorporates graph attention network to effectively utilizing these attributes. Extensive experiments on various datasets demonstrate that our model outperforms existing state-of-the-art models and effectively addresses the problems of cold-start and sparsity. In addition, due to the estimation of multiple attributes of the questions, our model exhibits superior interpretability.

Sequential recommendation has attracted increasing attention due to its ability to accurately capture the dynamic changes in user interests. We have noticed that generative models, especially diffusion models, which have achieved significant results in fields like image and audio, hold considerable promise in the field of sequential recommendation. However, existing sequential recommendation methods based on diffusion models are constrained by a prior distribution limited to Gaussian distribution, hindering the possibility of introducing user-specific information for each recommendation and leading to information loss. To address these issues, we introduce the Schrödinger Bridge into diffusion-based sequential recommendation models, creating the SdifRec model. This allows us to replace the Gaussian prior of the diffusion model with the user's current state, directly modeling the process from a user's current state to the target recommendation. Additionally, to better utilize collaborative information in recommendations, we propose an extended version of SdifRec called con-SdifRec, which utilizes user clustering information as a guiding condition to further enhance the posterior distribution. Finally, extensive experiments on multiple public benchmark datasets have demonstrated the effectiveness of SdifRec and con-SdifRec through comparison with several state-of-the-art methods. Further in-depth analysis has validated their efficiency and robustness.

Model editing aims to precisely alter the behaviors of large language models (LLMs) in relation to specific knowledge, while leaving unrelated knowledge intact. This approach has proven effective in addressing issues of hallucination and outdated information in LLMs. However, the potential of using model editing to modify knowledge in the medical field remains largely unexplored, even though resolving hallucination is a pressing need in this area. Our observations indicate that current methods face significant challenges in dealing with specialized and complex knowledge in medical domain. Therefore, we propose MedLaSA, a novel Layer-wise Scalable Adapter strategy for medical model editing. MedLaSA harnesses the strengths of both adding extra parameters and locate-then-edit methods for medical model editing. We utilize causal tracing to identify the association of knowledge in neurons across different layers, and generate a corresponding scale set from the association value for each piece of knowledge. Subsequently, we incorporate scalable adapters into the dense layers of LLMs. These adapters are assigned scaling values based on the corresponding specific knowledge, which allows for the adjustment of the adapter's weight and rank. The more similar the content, the more consistent the scale between them. This ensures precise editing of semantically identical knowledge while avoiding impact on unrelated knowledge. To evaluate the editing impact on the behaviours of LLMs, we propose two model editing studies for medical domain: (1) editing factual knowledge for medical specialization and (2) editing the explanatory ability for complex knowledge. We build two novel medical benchmarking datasets and introduce a series of challenging and comprehensive metrics. Extensive experiments on medical LLMs demonstrate the editing efficiency of MedLaSA, without affecting unrelated knowledge.

Sequential recommender systems offer personalized suggestions by modeling users' interactions chronologically to capture dynamic user interest. Existing approaches typically fail to adequately describe the dynamics of the entire recommender system, including shifts in both user interest and item availability. To address this, we propose a simple yet effective framework with three key perspectives, tailored to the dynamics of recommender system by fully exploiting the time information. Firstly, we propose a dynamic candidate set construction approach to prevent the model from learning future interactions. Secondly, assuming that user behaviors remain consistent over short terms but may evolve over long terms, we employ a interval-weighted optimization target to model the correlation of users' historical interactions. Finally, we introduce a specialized time-aware attention module to enhance recommendations within specific temporal contexts. Extensive experiments demonstrate the effectiveness and generalizability of our framework. We make our codes publicly available.

Knowledge graph completion (KGC) is a widely used method to tackle incompleteness in knowledge graphs (KGs) by making predictions for missing links. Description-based KGC leverages pre-trained language models to learn entity and relation representations with their names or descriptions, which shows promising results. However, the performance of description-based KGC is still limited by the quality of text and the incomplete structure, as it lacks sufficient entity descriptions and relies solely on relation names, leading to sub-optimal results. To address this issue, we propose MPIKGC, a general framework to compensate for the deficiency of contextualized knowledge and improve KGC by querying large language models (LLMs) from various perspectives, which involves leveraging the reasoning, explanation, and summarization capabilities of LLMs to expand entity descriptions, understand relations, and extract structures, respectively. We conducted extensive evaluation of the effectiveness and improvement of our framework based on four description-based KGC models, for both link prediction and triplet classification tasks. All codes and generated data will be publicly available after review.

As a fresh way to improve the user viewing experience, videos of time-sync comments have attracted a lot of interest. Many efforts have been made to explore the effectiveness of time-sync comments for various applications. However, due to the complexity of interactions among users, videos, and comments, it still remains challenging to understand users’ behavior on time-sync comments. Along this line, we study the problem of time-sync comment behavior prediction with considerations of both historical behaviors and multi-modal information of visual frames and textual comments. Specifically, we propose a novel Multi-modal short- and long-Range Temporal Convolutional Network model, namely MRT. Firstly, we design two amplified Temporal Convolutional Networks with different sizes of receptive fields, to capture both short- and long-range surrounding contexts for each frame and time-sync comments. Then, we design a bottle-neck fusion module to obtain the multi-modal enhanced representation. Furthermore, we take the user preferences into consideration to generate the personalized multi-model semantic representation at each timestamp. Finally, we utilize the binary cross-entropy loss to optimize MRT on the basis of users’ historical records. Through comparing with representative baselines, we demonstrate the effectiveness of MRT and qualitatively verify the necessity and utility of short- and long-range contextual and multi-modal information through extensive experiments.

Entity Linking (EL) is the process of associating ambiguous textual mentions to specific entities in a knowledge base.Traditional EL methods heavily rely on large datasets to enhance their performance, a dependency that becomes problematic in the context of few-shot entity linking, where only a limited number of examples are available for training.To address this challenge, we present OneNet, an innovative framework that utilizes the few-shot learning capabilities of Large Language Models (LLMs) without the need for fine-tuning.To the best of our knowledge, this marks a pioneering approach to applying LLMs to few-shot entity linking tasks.OneNet is structured around three key components prompted by LLMs: (1) an entity reduction processor that simplifies inputs by summarizing and filtering out irrelevant entities, (2) a dual-perspective entity linker that combines contextual cues and prior knowledge for precise entity linking, and (3) an entity consensus judger that employs a unique consistency algorithm to alleviate the hallucination in the entity linking reasoning.Comprehensive evaluations across seven benchmark datasets reveal that OneNet outperforms current stateof-the-art entity linking methods.

Chinese Spelling Check (CSC) aims to identify and correct spelling errors in Chinese texts, where enhanced semantic understanding of a sentence can significantly improve correction accuracy.Recently, Large Language Models (LLMs) have demonstrated exceptional mastery of world knowledge and semantic understanding, rendering them more robust against spelling errors.However, the application of LLMs in CSC is a double-edged sword, as they tend to unnecessarily alter sentence length and modify rare but correctly used phrases.In this paper, by leveraging the capabilities of LLMs while mitigating their limitations, we propose a novel plug-and-play Alignment-and-Replacement Module (ARM) that enhances the performance of existing CSC models and without the need for retraining or fine-tuning.Experiment results and analysis on three benchmark datasets demonstrate the effectiveness and competitiveness of the proposed module.

Despite the significant progress of large language models (LLMs) in various tasks, they often produce factual errors due to their limited internal knowledge.Retrieval-Augmented Generation (RAG), which enhances LLMs with external knowledge sources, offers a promising solution.However, these methods can be misled by irrelevant paragraphs in retrieved documents.Due to the inherent uncertainty in LLM generation, inputting the entire document may introduce off-topic information, causing the model to deviate from the central topic and affecting the relevance of the generated content.To address these issues, we propose the Retrieve-Plan-Generation (RPG) framework.RPG generates plan tokens to guide subsequent generation in the plan stage.In the answer stage, the model selects relevant fine-grained paragraphs based on the plan and uses them for further answer generation.This plan-answer process is repeated iteratively until completion, enhancing generation relevance by focusing on specific topics.To implement this framework efficiently, we utilize a simple but effective multi-task prompt-tuning method, enabling the existing LLMs to handle both planning and answering.We comprehensively compare RPG with baselines across 5 knowledge-intensive generation tasks, demonstrating the effectiveness of our approach.1

To protect privacy and meet legal regulations, federated learning (FL) has gained significant attention for training speech-to-text (S2T) systems, including automatic speech recognition (ASR) and speech translation (ST). However, the commonly used FL approach (i.e., FEDAVG) in S2T tasks typically suffers from extensive communication overhead due to multi-round interactions based on the whole model and performance degradation caused by data heterogeneity among clients. To address these issues, we propose a personalized federated S2T framework that introduces FEDLORA, a lightweight LoRA module for client-side tuning and interaction with the server to minimize communication overhead, and FEDMEM, a global model equipped with a k-nearest-neighbor (kNN) classifier that captures client-specific distributional shifts to achieve personalization and overcome data heterogeneity. Extensive experiments based on Conformer and Whisper backbone models on CoVoST and GigaSpeech benchmarks show that our approach significantly reduces the communication overhead on all S2T tasks and effectively personalizes the global model to overcome data heterogeneity.

Data scale has been acknowledged as a crucial factor for enhancing the generalization and effectiveness of pre-training models. While existing methods of multivariate time series pre-training are primarily limited to a single specific dataset, scaling to a larger scenario that includes multiple diverse datasets (e.g., multi-region data) remains a substantial challenge. In this paper, we present a novel Decoupled Spatial-Temporal Representation Learning (DeSTR) framework to serve as the backbone network for investigating the data scaling capability of multivariate time series pre-training architectures. Specifically, DeSTR utilizes two separate encoders to capture both the temporal dynamics within each time series and the spatial correlations among multiple variables. The obtained representations of distinct modalities are then fed into a Spatial-Guided Temporal Transformer to equip the temporal features with spatial discriminative information. Moreover, we employ masked autoencoding as the foundational pre-training framework and introduce spacetime-agnostic augmentation to improve robustness and facilitate implicit spatiotemporal modeling. Finally, we successfully pre-train a unified time series representation learning framework on real-world datasets from three different cities. Extensive experiments are carried out on various downstream tasks to validate the performance of DeSTR, compared with three categories of state-of-the-art baselines: deep sequential models, spatial-temporal graph neural networks, and time series representation learning methods. The results clearly demonstrate the advantages of scaling multivariate time series pre-training to multiple datasets, highlighting the effectiveness of DeSTR as a general spatiotemporal learner.

Micro-expressions (MEs) are fleeting, subtle, and involuntary facial expressions that can reveal genuine emotions of human beings. Although many advanced supervised deep learning efforts have been devoted to ME recognition (MER), they are severely limited by the lack of sufficient well-labeled ME data when learning discriminative ME features. To address this problem, we propose a novel self-supervised ME representation learning method based on Temporal Gaussian Masked Autoencoder, termed TGMAE. Specifically, a Temporal Gaussian Masking strategy is customized to construct a challenging spatiotemporal ME movement reconstruction task, which can effectively assist the model in perceiving ME features from abundant unlabeled ME data. Additionally, to bridge the semantic gap between encoded features for reconstruction and emotion features for recognition, a bridging classifier is introduced for downstream MER. Comprehensive experiments demonstrate the remarkable performance of TGMAE, significantly surpassing the second-best method with a maximum improvement of 4.96% in UF1 and 6.92% in UAR.

Recent years have witnessed the urgent request for multi-modal sarcasm detection in social media platforms. Though large efforts have been made with significant progress, prior arts may fail to fully integrate the commonsense knowledge, and struggle to model the incongruity within implicit meanings of multimodal cues. To address these limitations, we propose a novel Knowledge-Enhanced Multi-perspective Incongruity Perception network, named KEMIP. Specifically, we adopt generative language models to produce captions and commonsense for image and text respectively for comprehensive understanding. Subsequently, to exploit the essential cues from multiple perspectives, we customize an incongruity perception module, which utilizes several fusion networks to capture both literal and implicit inconsistencies. Afterwards, an ensemble weighting gate is employed to integrate the result of individual perspective. Experiments on a public multimodal sarcasm detection benchmark demonstrate the superiority of our proposed KEMIP framework.

The weighted squared loss is a common component in several Collaborative Filtering (CF) algorithms for item recommendation, including the representative implicit Alternating Least Squares (iALS). Despite its widespread use, this loss function lacks a clear connection to ranking objectives such as Discounted Cumulative Gain (DCG), posing a fundamental challenge in explaining the exceptional ranking performance observed in these algorithms. In this work, we make a breakthrough by establishing a connection between squared loss and ranking metrics through a Taylor expansion of the DCG-consistent surrogate loss—softmax loss. We also discover a new surrogate squared loss function, namely Ranking-Generalizable Squared (RG$^2$) loss, and conduct thorough theoretical analyses on the DCG-consistency of the proposed loss function. Later, we present an example of utilizing the RG$^2$ loss with Matrix Factorization (MF), coupled with a generalization upper bound and an ALS optimization algorithm that leverages closed-form solutions over all items. Experimental results over three public datasets demonstrate the effectiveness of the RG$^2$ loss, exhibiting ranking performance on par with, or even surpassing, the softmax loss while achieving faster convergence.

Adaptive Testing System (ATS) is a promising testing mode, extensively utilized in standardized tests like the GRE. It offers personalized ability assessment by dynamically adjusting questions based on individual ability levels. Compared to traditional exams, ATS can improve the accuracy of ability estimates while simultaneously reducing the number of questions required. Despite the diverse testing formats of ATS, tailored to different adaptability requirements in various testing scenarios, there is a notable absence of a unified framework for modeling them. In this paper, we introduce a unified data-driven ATS framework that conceptualizes the various testing formats as a hierarchical test structure search problem. It can learn directly from data to solve for the optimal questions for each student, eliminating the need for manual test design. The proposed solution algorithm comes with theoretical guarantees for estimation error and convergence. Empirical results show that our framework maintains assessment accuracy while reducing question count by 20% on average and improving training stability.

Last year has witnessed the re-flourishment of tag-aware recommender systems supported by the LLM-enriched tags. Unfortunately, though large efforts have been made, current solutions may fail to describe the diversity and uncertainty inherent in user preferences with only tag-driven profiles. Recently, with the development of geometry-based techniques, e.g., box embeddings, the diversity of user preferences now could be fully modeled as the range within a box in high dimension space. However, defect still exists as these approaches are incapable of capturing high-order neighbor signals, i.e., semantic-rich multi-hop relations within the user-tag-item tripartite graph, which severely limits the effectiveness of user modeling. To deal with this challenge, in this paper, we propose a novel framework, called BoxGNN, to perform message aggregation via combinations of logical operations, thereby incorporating high-order signals. Specifically, we first embed users, items, and tags as hyper-boxes rather than simple points in the representation space, and define two logical operations, i.e., union and intersection, to facilitate the subsequent process. Next, we perform the message aggregation mechanism via the combination of logical operations, to obtain the corresponding high-order box representations. Finally, we adopt a volume-based learning objective with Gumbel smoothing techniques to refine the representation of boxes. Extensive experiments on two publicly available datasets and one LLM-enhanced e-commerce dataset have validated the superiority of BoxGNN compared with various state-of-the-art baselines. The code is released online: https://github.com/critical88/BoxGNN.

The sequential recommender (SR) system is a crucial component of modern recommender systems, as it aims to capture the evolving preferences of users. Significant efforts have been made to enhance the capabilities of SR systems. These methods typically follow the model-centric paradigm, which involves developing effective models based on fixed datasets. However, this approach often overlooks potential quality issues and flaws inherent in the data. Driven by the potential of data-centric AI, we propose a novel data-centric paradigm for developing an ideal training dataset using a model-agnostic dataset regeneration framework called DR4SR. This framework enables the regeneration of a dataset with exceptional cross-architecture generalizability. Additionally, we introduce the DR4SR+ framework, which incorporates a model-aware dataset personalizer to tailor the regenerated dataset specifically for a target model. To demonstrate the effectiveness of the data-centric paradigm, we integrate our framework with various model-centric methods and observe significant performance improvements across four widely adopted datasets. Furthermore, we conduct in-depth analyses to explore the potential of the data-centric paradigm and provide valuable insights. The code can be found at https://github.com/USTC-StarTeam/DR4SR.

Recent advancements in Large Language Models~(LLMs) have greatly enhanced the generation capabilities of dialogue systems. However, progress on emotional expression during dialogues might be still limited, especially when capturing and processing the multimodal cues for emotional expression. Therefore, it is urgent to fully adapt the multimodal understanding ability and transferability of LLMs to enhance the emotional-oriented multimodal processing capabilities. To that end, in this paper, we propose a novel Emotion-Guided Multimodal Dialogue model based on LLM, termed ELMD. Specifically, to enhance the emotional expression ability of LLMs, our ELMD customizes an emotional retrieval module, which mainly provides appropriate response demonstration for LLM in understanding emotional context. Subsequently, a two-stage training strategy is proposed, founded on previous demonstration support, to support uncovering nuanced emotions behind multimodal information and constructing natural responses. Comprehensive experiments demonstrate the effectiveness and superiority of ELMD.

Deep Hashing (DH) has emerged as an indispensable technique for fast image search in recent years. To deploy DH on resource-limited devices, the Binary Neural Network (BNN) offers a solution that significantly reduces computations and parameters compared to CNN. Unfortunately, applying BNN directly to DH will lead to huge performance degradation. To tackle this problem, we first conducted extensive experiments and discovered that the center-based method provides a fundamental guarantee for BNN-DH performance. Subsequently, we delved deeper into the impact of BNNs on center-based methods and revealed two key insights. First, we find reducing the distance between hash codes and hash centers is challenging for BNN-DH compared to CNN-based DH. Second, the evolution of hash code aggregation undergoes two stages in BNN-DH, which is different from CNN-based DH. Based on these findings, we designed a strong and general method called One-bit Deep Hashing (ODH). First, ODH incorporates a semantic self-adaptive hash center module to address the problem of hash codes inadequately converging to their hash centers. Then, it employs a novel two-stage training method to consider the evolution of hash code aggregation. Finally, extensive experiments on two datasets demonstrate that ODH can achieve significant superiority over other BNN-DH models.

Multimodal Entity Linking (MEL) aims to address the ambiguity in multimodal mentions and associate them with Multimodal Knowledge Graphs (MMKGs). Existing works primarily focus on designing multimodal interaction and fusion mechanisms to enhance the performance of MEL. However, these methods still overlook two crucial gaps within the MEL task. One is the content discrepancy between mentions and entities, manifested as uneven information density. The other is the knowledge gap, indicating insufficient knowledge extraction and reasoning during the linking process. To bridge these gaps, we propose a novel framework FissFuse, as well as a plug-and-play knowledge-aware re-ranking method KAR. Specifically, FissFuse collaborates with the Fission and Fusion branches, establishing dynamic features for each mention-entity pair and adaptively learning multimodal interactions to alleviate content discrepancy. Meanwhile, KAR is endowed with carefully crafted instruction for intricate knowledge reasoning, serving as re-ranking agents empowered by Large Language Models (LLMs). Extensive experiments on two well-constructed MEL datasets demonstrate outstanding performance of FissFuse compared with various baselines. Comprehensive evaluations and ablation experiments validate the effectiveness and generality of KAR.

Recently, automatic micro-expression (ME) analysis has attracted increasing attention, since ME is a spontaneous facial expression that can truly reflect the emotional state an individual tries to conceal. As a crucial step in ME analysis, Micro- and Macro-expression (MaE) spotting aims to sequentially identify the occurrence intervals of MEs and MaEs within a long video sequence. However, the subtle spatiotemporal movements of MEs and the scarcity of well-labeled data pose great challenges for accurately spotting them. To this end, this paper proposes a novel spotting framework based on Multi-scale Feature Learning Network with Optical Flow Correction. Specifically, we first integrate the pre-trained VideoMAE and customized convolutional layers as a visual feature extraction module to learn the facial motion features in long video sequences. Then, to comprehensively locate and identify the existing ME and MaE segments, we introduce a multi-scale candidate segment generation method based on the ActionFormer. In particular, a multi-start points optical flow filtering method is proposed to improve the precision of expression spotting. Finally, we conduct comprehensive experiments on the MEGC2024 spotting task, and the experimental results demonstrate the effectiveness of our method, which ranks second in this task. The implemented code is also publicly available at https://github.com/zzy188zzy/megc_spotting_code.

Large language models (LLMs) are considered a crucial technology for advancing intelligent education since they exhibit the potential for an in-depth understanding of teaching scenarios and providing students with personalized guidance. Nonetheless, current LLM-based application in personalized teaching predominantly follows a "Question-Answering" paradigm, where students are passively provided with answers and explanations. In this paper, we propose SocraticLM, which achieves a Socratic "Thought-Provoking" teaching paradigm that fulfills the role of a real classroom teacher in actively engaging students in the thought process required for genuine problem-solving mastery. To build SocraticLM, we first propose a novel "Dean-Teacher-Student" multi-agent pipeline to construct a new dataset, SocraTeach, which contains $35$K meticulously crafted Socratic-style multi-round (equivalent to $208$K single-round) teaching dialogues grounded in fundamental mathematical problems. Our dataset simulates authentic teaching scenarios, interacting with six representative types of simulated students with different cognitive states, and strengthening four crucial teaching abilities. SocraticLM is then fine-tuned on SocraTeach with three strategies balancing its teaching and reasoning abilities. Moreover, we contribute a comprehensive evaluation system encompassing five pedagogical dimensions for assessing the teaching quality of LLMs. Extensive experiments verify that SocraticLM achieves significant improvements in the teaching performance, outperforming GPT4 by more than 12\%. Our dataset and code is available at https://github.com/Ljyustc/SocraticLM.

Intelligent education stands as a prominent application of machine learning. Within this domain, cognitive diagnosis (CD) is a key research focus that aims to diagnose students' proficiency levels in specific knowledge concepts. As a crucial task within the field of education, cognitive diagnosis encompasses two fundamental requirements: accuracy and fairness. Existing studies have achieved significant success by primarily utilizing observed historical logs of student-exercise interactions. However, real-world scenarios often present a challenge, where a substantial number of students engage with a limited number of exercises. This data sparsity issue can lead to both inaccurate and unfair diagnoses. To this end, we introduce a monotonic data augmentation framework, CMCD, to tackle the data sparsity issue and thereby achieve accurate and fair CD results. Specifically, CMCD integrates the monotonicity assumption, a fundamental educational principle in CD, to establish two constraints for data augmentation. These constraints are general and can be applied to the majority of CD backbones. Furthermore, we provide theoretical analysis to guarantee the accuracy and convergence speed of CMCD. Finally, extensive experiments on real-world datasets showcase the efficacy of our framework in addressing the data sparsity issue with accurate and fair CD results.

Incomplete spatio-temporal data in real-world has spawned many research. However, existing methods often utilize iterative message-passing across temporal and spatial dimensions, resulting in substantial information loss and high computational cost. We provide a theoretical analysis revealing that such iterative models are not only susceptible to data sparsity but also to graph sparsity, causing unstable performances on different datasets. To overcome these limitations, we introduce a novel method named One-step Propagation and Confidence-based Refinement (OPCR). In the first stage, OPCR leverages inherent spatial and temporal relationships by employing sparse attention mechanism. These modules propagate limited observations directly to the global context through one-step imputation, which are theoretically effected only by data sparsity. Following this, we assign confidence levels to the initial imputations by correlating missing data with valid data. This confidence-based propagation refines the seperate spatial and temporal imputation results through spatio-temporal dependencies. We evaluate the proposed model across various downstream tasks involving highly sparse spatio-temporal data. Empirical results indicate that our model outperforms state-of-the-art imputation methods, demonstrating its superior effectiveness and robustness.

A well-known dilemma in large vision-language models (e.g., GPT-4, LLaVA) is that while increasing the number of vision tokens generally enhances visual understanding, it also significantly raises memory and computational costs, especially in long-term, dense video frame streaming scenarios. Although learnable approaches like Q-Former and Perceiver Resampler have been developed to reduce the vision token burden, they overlook the context causally modeled by LLMs (i.e., key-value cache), potentially leading to missed visual cues when addressing user queries. In this paper, we introduce a novel approach to reduce vision compute by leveraging redundant vision tokens ``skipping layers'' rather than decreasing the number of vision tokens. Our method, VideoLLM-MoD, is inspired by mixture-of-depths LLMs and addresses the challenge of numerous vision tokens in long-term or streaming video. Specifically, for certain transformer layer, we learn to skip the computation for a high proportion (e.g., 80\%) of vision tokens, passing them directly to the next layer. This approach significantly enhances model efficiency, achieving approximately 42% time and 30% memory savings for the entire training. Moreover, our method reduces the computation in the context and avoid decreasing the vision tokens, thus preserving or even improving performance compared to the vanilla model. We conduct extensive experiments to demonstrate the effectiveness of VideoLLM-MoD, showing its state-of-the-art results on multiple benchmarks, including narration, forecasting, and summarization tasks in COIN, Ego4D, and Ego-Exo4D datasets. The code and checkpoints will be made available at github.com/showlab/VideoLLM-online.

Sequential recommendation (SR) aims to predict items that users may be interested in based on their historical behavior sequences. We revisit SR from a novel information-theoretic perspective and find that conventional sequential modeling methods fail to adequately capture the randomness and unpredictability of user behavior. Inspired by fuzzy information processing theory, this paper introduces the DDSR model, which uses fuzzy sets of interaction sequences to overcome the limitations and better capture the evolution of users' real interests. Formally based on diffusion transition processes in discrete state spaces, which is unlike common diffusion models such as DDPM that operate in continuous domains. It is better suited for discrete data, using structured transitions instead of arbitrary noise introduction to avoid information loss. Additionally, to address the inefficiency of matrix transformations due to the vast discrete space, we use semantic labels derived from quantization or RQ-VAE to replace item IDs, enhancing efficiency and improving cold start issues. Testing on three public benchmark datasets shows that DDSR outperforms existing state-of-the-art methods in various settings, demonstrating its potential and effectiveness in handling SR tasks.

Two-stage recommender systems play a crucial role in efficiently identifying relevant items and personalizing recommendations from a vast array of options. This paper, based on an error decomposition framework, analyzes the generalization error for two-stage recommender systems with a tree structure, which consist of an efficient tree-based retriever and a more precise yet time-consuming ranker. We use the Rademacher complexity to establish the generalization upper bound for various tree-based retrievers using beam search, as well as for different ranker models under a shifted training distribution. Both theoretical insights and practical experiments on real-world datasets indicate that increasing the branches in tree-based retrievers and harmonizing distributions across stages can enhance the generalization performance of two-stage recommender systems.

Student knowledge profile is the basis for adaptive learning applications in online learning resulting from modeling the student mastery of knowledge concepts. In recent years, typical works based on knowledge tracing (KT) expect to profile students and have achieved significant success for the next performance prediction. However, in practical online learning scenarios, current methods tend to suffer from the following challenges: 1) Prediction inconsistency: The accuracy of the next performance prediction is inconsistent with the accuracy of student knowledge profile prediction, which is the more required result. 2) Cold start of knowledge: In online learning scenarios, it is often necessary to profile some knowledge concepts without learning records in advance. In this paper, we propose a novel Graph-based Student Knowledge Profile Model (GSKPM), along with a new end-to-end training objective, to tackle these challenges. We first define a new training objective to ensure the model is capable of inferring consistent student knowledge profiles. Then in this model, a two-stage hyper-aggregation process is employed to make full use of the topological relations between knowledge concepts and knowledge domains to provide information during profiling, especially for cold start knowledge concepts. Finally, through extensive experiments on real-world datasets, we will show that GSKPM achieves better prediction performances on student knowledge profiles and well deals with the cold start problem.

Text classification has long time been researched as a fundamental problem in information retrieval. Since text data are frequently connected with graph structures, it poses new possibilities for a more accurate and explainable classification. One common approach of this graph-text integration is to consider text as graph attributes and utilize GNNs to conduct a node classification task. While both text and graph data are modeled, GNNs treat text in a rather coarse-grained way, have limitations in preserving the detailed structures of a graph, and are less robust to graph sparsity. In this paper, we propose to take an alternative perspective instead, viewing graph as the context of texts, as enlightened by retrieval augmented generation. We propose a novel framework called Graph Retrieval Prompt Tuning (GRPT), consisting of a Graph Retrieval Module and a Prompt Tuning Module integrated with graph context. For graph retrieval, two retrieval strategies are designed to retrieve node context and path context, preserving both node proximity and detailed connectivity patterns. Extensive experiments on four real-world datasets show the effectiveness of our framework in both standard supervised and sparse settings.

Formula recognition endeavors to automatically identify mathematical formulas from images. Currently, the Encoder-Decoder model has significantly advanced the translation from image to corresponding formula markups. Nonetheless, previous research primarily concentrated on single-line formula recognition, ignoring the recognition of multi-line formulas, which presents additional challenges such as more stringent grammatical restrictions and two- dimensional positions. In this work, we present GAP (Grammar And Position-Aware formula recognition), a comprehensive framework designed to tackle the challenges in multi-line mathematical formula recognition. First, to overcome the limitations imposed by grammar, we design a novel Grammar Aware Contrastive Learning (GACL) module, integrating complex grammar rules into the transcription model through a contrastive learning mechanism. Furthermore, primitive contrastive learning lacks clear directions for comprehending grammar rules and can lead to unstable convergence or prolonged training cycles. To enhance training efficiency, we propose Rank-Based Sampling (RBS) specialized for multi-line formulas, which guides the learning process by the importance ranking of different grammar errors. Finally, spatial location information is critical considering the two-dimensional nature of multi-line formulas. To aid the model in keeping track of that global information, we introduced a Visual Coverage (VC) mechanism that incorporates historical attention information into the image features via a parameter-free way. To validate the effectiveness of our GAP framework, we construct a new dataset Multi-Line containing 12,002 multi-line formulas and conduct extensive experiments to show the efficacy of our GAP framework in capturing grammatical rules, enhancing recognition accuracy, and enhancing training efficiency. Codes and datasets are available at https://github.com/Sinon02/GAP.

In recommendation systems, users frequently engage in multiple types of behaviors, such as clicking, adding to cart, and purchasing. Multi-behavior sequential recommendation aims to jointly consider multiple behaviors to improve the target behavior's performance. However, with diversified behavior data, user behavior sequences will become very long in the short term, which brings challenges to the efficiency of the sequence recommendation model. Meanwhile, some behavior data will also bring inevitable noise to the modeling of user interests. To address the aforementioned issues, firstly, we develop the Efficient Behavior Sequence Miner (EBM) that efficiently captures intricate patterns in user behavior while maintaining low time complexity and parameter count. Secondly, we design hard and soft denoising modules for different noise types and fully explore the relationship between behaviors and noise. Finally, we introduce a contrastive loss function along with a guided training strategy to contrast the valid information with the noisy signal in the data, and seamlessly integrate the two denoising processes to achieve a high degree of decoupling of the noisy signal. Sufficient experiments on real-world datasets demonstrate the effectiveness and efficiency of our approach in dealing with multi-behavior sequential recommendation.

Unsupervised semantic hashing has emerged as an indispensable technique for fast image search, which aims to convert images into binary hash codes without relying on labels. Recent advancements in the field demonstrate that employing large-scale backbones (e.g., ViT) in unsupervised semantic hashing models can yield substantial improvements. However, the inference delay has become increasingly difficult to overlook. Knowledge distillation provides a means for practical model compression to alleviate this delay. Nevertheless, the prevailing knowledge distillation approaches are not explicitly designed for semantic hashing. They ignore the unique search paradigm of semantic hashing, the inherent necessities of the distillation process, and the property of hash codes. In this paper, we propose an innovative Bit-mask Robust Contrastive knowledge Distillation (BRCD) method, specifically devised for the distillation of semantic hashing models. To ensure the effectiveness of two kinds of search paradigms in the context of semantic hashing, BRCD first aligns the semantic spaces between the teacher and student models through a contrastive knowledge distillation objective. Additionally, to eliminate noisy augmentations and ensure robust optimization, a cluster-based method within the knowledge distillation process is introduced. Furthermore, through a bit-level analysis, we uncover the presence of redundancy bits resulting from the bit independence property. To mitigate these effects, we introduce a bit mask mechanism in our knowledge distillation objective. Finally, extensive experiments not only showcase the noteworthy performance of our BRCD method in comparison to other knowledge distillation methods but also substantiate the generality of our methods across diverse semantic hashing models and backbones. The code for BRCD is available at https://github.com/hly1998/BRCD.

Knowledge-based question answering (KBQA) is a key task in natural language processing research, and also an approach to access the web data and knowledge, which requires exploiting knowledge graphs (KGs) for reasoning. In the literature, one promising solution for KBQA is to incorporate the pretrained language model (LM) with KGs by generating KG-centered pretraining corpus, which has shown its superiority. However, these methods often depend on specific techniques and resources to work, which may not always be available and restrict its application. Moreover, existing methods focus more on improving language understanding with KGs, while neglect the more important human-like complex reasoning. To this end, in this paper, we propose a general K nowledge-I njected C urriculum P retraining framework (KICP) to achieve comprehensive KG learning and exploitation for KBQA tasks, which is composed of knowledge injection (KI), knowledge adaptation (KA) and curriculum reasoning (CR). Specifically, the KI module first injects knowledge into the LM by generating KG-centered pretraining corpus, and generalizes the process into three key steps that could work with different implementations for flexible application. Next, the KA module learns knowledge from the generated corpus with LM equipped with an adapter as well as keeps its original natural language understanding ability to reduce the negative impacts of the difference between the generated and natural corpus. Last, to enable the LM with complex reasoning, the CR module follows human reasoning patterns to construct three corpora with increasing difficulties of reasoning, and further trains the LM from easy to hard in a curriculum manner to promote model learning. We provide an implementation of the general framework, and evaluate the proposed KICP on four real-word datasets. The results demonstrate that our framework can achieve higher performances, and have good generalization ability to other QA tasks.

Cognitive diagnosis models have been widely used in different areas, especially intelligent education, to measure users' proficiency levels on knowledge concepts, based on which users can get personalized instructions. As the measurement is not always reliable due to the weak links of the models and data, the uncertainty of measurement also offers important information for decisions. However, the research on the uncertainty estimation lags behind that on advanced model structures for cognitive diagnosis. Existing approaches have limited efficiency and leave an academic blank for sophisticated models which have interaction function parameters (e.g., deep learning-based models). To address these problems, we propose a unified uncertainty estimation approach for a wide range of cognitive diagnosis models. Specifically, based on the idea of estimating the posterior distributions of cognitive diagnosis model parameters, we first provide a unified objective function for mini-batch based optimization that can be more efficiently applied to a wide range of models and large datasets. Then, we modify the reparameterization approach in order to adapt to parameters defined on different domains. Furthermore, we decompose the uncertainty of diagnostic parameters into data aspect and model aspect, which better explains the source of uncertainty. Extensive experiments demonstrate that our method is effective and can provide useful insights into the uncertainty of cognitive diagnosis.

Abstract Information Extraction (IE) aims to extract structural knowledge from plain natural language texts. Recently, generative Large Language Models (LLMs) have demonstrated remarkable capabilities in text understanding and generation. As a result, numerous works have been proposed to integrate LLMs for IE tasks based on a generative paradigm. To conduct a comprehensive systematic review and exploration of LLM efforts for IE tasks, in this study, we survey the most recent advancements in this field. We first present an extensive overview by categorizing these works in terms of various IE subtasks and techniques, and then we empirically analyze the most advanced methods and discover the emerging trend of IE tasks with LLMs. Based on a thorough review conducted, we identify several insights in technique and promising research directions that deserve further exploration in future studies. We maintain a public repository and consistently update related works and resources on GitHub (LLM4IE repository).

Factorization machines (FMs) are widely used in recommender systems due to their adaptability and ability to learn from sparse data. However, for the ubiquitous non-interactive features in sparse data, existing FMs can only estimate the parameters corresponding to these features via the inner product of their embeddings. Undeniably, they cannot learn the direct interactions of these features, which limits the model's expressive power. To this end, we first present MixFM, inspired by Mixup, to generate auxiliary training data to boost FMs. Unlike existing augmentation strategies that require labor costs and expertise to collect additional information such as position and fields, these augmented data are only by the convex combination of the raw ones without any professional knowledge support. More importantly, if non-interactive features exist in parent samples to be mixed respectively, MixFM will establish their direct interactions. Second, considering that MixFM may generate redundant or even detrimental instances, we further put forward a novel Factorization Machine powered by Saliency-guided Mixup (denoted as SMFM). Guided by the customized saliency, SMFM can generate more informative neighbor data. Through theoretical analysis, we prove that the proposed methods minimize the upper bound of the generalization error, which positively enhances FMs. Finally, extensive experiments on seven datasets confirm that our approaches are superior to baselines. Notably, the results also show that "poisoning" mixed data benefits the FM variants.

One of the most important subconscious reactions, micro-expression (ME), is a spontaneous, subtle, and transient facial expression that reveals human beings' genuine emotion. Therefore, automatically recognizing ME (MER) is becoming increasingly crucial in the field of affective computing, providing essential technical support for lie detection, clinical psychological diagnosis, and public safety. However, the ME data scarcity has severely hindered the development of advanced data-driven MER models. Despite the recent efforts by several spontaneous ME databases to alleviate this problem, there is still a lack of sufficient data. Hence, in this paper, we overcome the ME data scarcity problem by collecting and annotating a dynamic spontaneous ME database with the largest current ME data scale called DFME (Dynamic Facial Micro-expressions). Specifically, the DFME database contains 7,526 well-labeled ME videos spanning multiple high frame rates, elicited by 671 participants and annotated by more than 20 professional annotators over three years. Furthermore, we comprehensively verify the created DFME, including using influential spatiotemporal video feature learning models and MER models as baselines, and conduct emotion classification and ME action unit classification experiments. The experimental results demonstrate that the DFME database can facilitate research in automatic MER, and provide a new benchmark for this field. DFME will be published via https://mea-lab-421.github.io.

Patent litigation is an expensive and time-consuming legal process. To reduce costs, companies can proactively manage patents using predictive analysis to identify potential plaintiffs, defendants, and patents that may lead to litigation. However, there has been limited progress in predicting patent litigation due to the scarcity of lawsuits, the complexities of intentions, and the diversity of litigation characteristics. To this end, in this paper, we summarize the major causes of patent litigation into multiple aspects: the complex relations among plaintiffs, defendants and patents as well as the diverse content information from them. Along this line, we propose a Multi-aspect Neural Tensor Factorization (MANTF) framework for patent litigation prediction. Firstly, a Pair-wise Tensor Factorization (PTF) module is designed to capture the complex relations among plaintiffs, defendants and patents inherent in a three-dimensional tensor, which will produce factorized latent vectors for companies and patents with pair-wise ranking estimators. Then, to better represent the patents and companies as an aid for PTF, we design a Patent Embedding Network (PEN) module and a Mask Company Embedding Network (MCEN) module to generate content-aware embedding for them, where PEN represents patents based on their meta, textual and graphical features, and MCEN represents companies by integrating their intrinsic features and competitions. Next, to integrate these three modules together, we leverage a Gaussian prior on the difference between factorized representations and content-aware embedding, and train MANTF in an end-to-end way. In the end, final predictions for patent litigation, i.e., the potentially litigated plaintiffs, defendants and patents, can be made with the well-trained model. We conduct extensive experiments on two real-world datasets, whose results prove that MANTF not only helps predict potential patent litigation but also shows robustness under various data sparse situations. Our codes and data are released athttps://github.com/USTCwuhan/MANTF-master.

Career mobility analysis aims at discovering the movement patterns of employees across different job positions or grades, which can benefit various human resource-related applications. Indeed, recent studies in this direction mainly focus on modeling individual career trajectories, while the macroguidance for labor market assessment has been largely ignored. To this end, in this article, we propose to study career mobility from a market-driven perspective based on large-scale online professional networks (OPNs). Specifically, we propose an uncertainty-aware graph autoencoders (UnGAEs) framework, which can simultaneously discover potential job title transition patterns and predict job durations. In this phase, we first construct a job title transition graph based on massive career trajectory data from OPNs. Then, considering the inherent uncertainty in career mobility, we introduce a novel uncertainty-aware graph encoder (UnGE) to represent job titles as Gaussian embeddings. Furthermore, we design two task-specific decoders that can preserve the asymmetric relationships between job titles, namely the gravity-inspired decoder (GID) and the energy-inspired decoder (EID), for predicting potential transition patterns and corresponding duration, respectively. In particular, both tasks are modeled through a specially designed multitask learning approach. Finally, extensive experiments on a real-world dataset clearly demonstrate the effectiveness of UnGAE compared with state-of-the-art baselines, as well as some potential applications such as job title benchmarking and career path planning.

With the booming of streaming media platforms, viewers now get used to watching dramas and movies via online platforms with more intelligent services. Usually, character relationships may dynamically evolve with stories promoting in long videos. Therefore, automatic tools to capture the social relation evolution among characters are urgently required to enrich the viewing experience. However, most existing works mainly focus on shorter isolated video clips. Considering the development of the plot, they may fail to effectively summarize relationships as holistic semantic representations for the whole video. To deal with these challenges, in this paper, we propose a novel Dynamic-Evolutionary Graph Attention Network (DE-GAT) framework to generate the evolving social relation graph among characters and capture the characters’ relation evolutionary trajectory throughout the entire video. DE-GAT first integrates the multimodal cues, including visual and textual information in each video clip via the graph attention network (GAT). Expanding the temporal receptive field from clip-level to scenario-level, the most relevant factors of the evolution of social relationships can be explored. Eventually, all the scenario-level social graphs are merged to obtain the evolving global social graph for the entire movie. Extensive evaluations on the real-world MovieGraphs dataset have validated the positive impact of temporal receptive field expansion and multimodal cues on capturing evolving social relations.

Structured pruning is an efficient compression technique that significantly reduces the inference latency and energy consumption of convolutional neural networks (CNNs) by eliminating redundant filters. However, existing works suffer from expensive algorithm costs in multi-hardware deployment scenarios involving several budgets across multiple hardware devices. To tackle this challenge, we propose a novel all-in-one hardware-oriented compression framework (AHC), which integrates structured pruning and data pruning to rapidly generate vast hardware-efficient models with ultra-low pruning and fine-tuning costs. Specifically, AHC develops a unified hardware-aware pruning (UHP), which rapidly generates numerous hardware-efficient models for several budgets across multiple hardware devices in once pruning process, thereby reducing pruning costs in multi-hardware deployment scenarios. Moreover, AHC proposes a progressive data pruning (PDP), which gradually removes samples that have a negligible impact on enhancing the predictive ability of pruned models, thereby accelerating the fine-tuning process with negligible performance loss. Extensive experiments demonstrate the superiority of the AHC over state-of-the-art (SOTA) structured pruning methods in terms of algorithm costs, latency, and accuracy. In particular, compared with SOTA hardware-oriented pruning method, AHC achieves comparable performances while reducing$5.3\times $pruning costs and$2.7\times $fine-tuning costs in multi-hardware deployment scenarios. Code is available athttps://github.com/HXuan-Wang/AHC.

Due to inherent interactivity, time-sync comment of videos have attracted increasing attention and were widely adopted in online video platforms. In addition to enhancing user engagement, time-sync comments provide abundant semantic information that can greatly enhance video understanding, which however is largely overlooked in mainstream video recommender systems. To address this issue, we propose a Hierarchical Multi-modal Attention Network (HMAN) to effectively utilize time-sync comment for recommendation. Specifically, we design a Multi-level Text Condense (MTC) Module to capture the accurate semantics of time-sync comments via text-level and vision-level condense operations. Then we propose a Range Convolution Block (RCB) to capture both visual and textual information from variable-length event segments leveraging the variable respective field. After that, we design a Hierarchical Multi-modal Branch Fusion (HMBF) Module to obtain a comprehensive multi-modal representation of the time-sync comments video. Finally, with the obtained video representation, recommendation scores are obtained through its inner product with user embedding. Extensive experiments demonstrate the effectiveness of the proposed HMAN, and ablation studies on different variants of HMAN further validate the utility of each component and the necessity of the hierarchical multi-modal branch fusion method.

While highly influential in deep learning, especially in natural language processing, the Transformer model has not exhibited competitive performance in unsupervised graph representation learning (UGRL). Conventional approaches, which focus on local substructures on the graph, offer simplicity but often fall short in encapsulating comprehensive structural information of the graph. This deficiency leads to suboptimal generalization performance. To address this, we proposed the GraKerformer model, a variant of the standard Transformer architecture, to mitigate the shortfall in structural information representation and enhance the performance in UGRL. By leveraging the shortest-path graph kernel (SPGK) to weight attention scores and combining graph neural networks, the GraKerformer effectively encodes the nuanced structural information of graphs. We conducted evaluations on the benchmark datasets for graph classification to validate the superior performance of our approach.

The field of education has undergone a significant revolution with the advent of intelligent systems and technology, which aim to personalize the learning experience, catering to the unique needs and abilities of individual learners. In this pursuit, a fundamental challenge is designing proper test for assessing the students’ cognitive status on knowledge and skills accurately and efficiently. One promising approach, referred to as Computerized Adaptive Testing (CAT), is to administrate computer-automated tests that alternately select the next item for each examinee and estimate their cognitive states given their responses to the selected items. Nevertheless, existing CAT systems suffer from inflexibility in item selection and ineffectiveness in cognitive state estimation, respectively. In this article, we propose a Model-Agnostic adaptive testing framework via Meta-leaned Gradient Embeddings, MAMGE for short, improving both item selection and cognitive state estimation simultaneously. For item selection, we design a Gradient Embedding-based Item Selector (GEIS) which incorporates the concept of gradient embeddings to represent items and selects the best ones that are both informative and representative. For cognitive state estimation, we propose a Meta-learned Cognitive State Estimator (MCSE) to automatically control the estimation process by learning to learn a proper initialization and dynamically inferred updates. Both MCSE and GEIS are inherently model-agnostic, and the two modules have an ingenious connection via meta-learned gradient embeddings. Finally, extensive experiments evaluate the effectiveness and flexibility of MAMGE.

The topic of multimodal conversation systems has recently garnered significant attention across various industries, including travel and retail, among others. While pioneering works in this field have shown promising performance, they often focus solely on context information at the utterance level, overlooking the context-aware dependencies of multimodal semantic elements like words and images. Furthermore, the ordinal information of images, which indicates the relevance between visual context and users’ demands, remains underutilized during the integration of visual content. Additionally, the exploration of how to effectively utilize corresponding attributes provided by users when searching for desired products is still largely unexplored. To address these challenges, we propose PMATE, a P osition-aware M ultimodal di A logue system with seman T ic E lements. Specifically, to obtain semantic representations at the element level, we first unfold the multimodal historical utterances and devise a position-aware multimodal element-level encoder. This component considers all images that may be relevant to the current turn and introduces a novel position-aware image selector to choose related images before fusing the information from the two modalities. Finally, we present a knowledge-aware two-stage decoder and an attribute-enhanced image searcher for the tasks of generating textual responses and selecting image responses, respectively. We extensively evaluate our model on two large-scale multimodal dialogue datasets, and the results of our experiments demonstrate that our approach outperforms several baseline methods.

Since social media such as Facebook and X (formerly known as Twitter) have permeated various aspects of daily life, people have strong incentives to influence information dissemination on these platforms and differentiate their content from the fierce competition. Existing dissemination strategies typically employ marketing techniques, such as seeking publicity through renowned actors or targeted advertising placements. Despite their various forms, most simply spread information to strengthen user impressions without conducting formal analyses of specific influence enhancement. And coupled with high costs, most fall short of expectations. To this end, we ingeniously formulate the task of social media dissemination as poisoning attacks, which influence specified content’s dissemination among target users by intervening in some users’ social media behaviors (including retweeting, following, and profile modifying). Correspondingly, we propose a novel poisoning attack, Influence-based Social Media Attack (ISMA) to generate discrete poisoning behaviors, which is difficult to achieve with existing attacks. In ISMA, we first contribute an efficient influence evaluator to quantify the spread influence of poisoning behaviors. Based on the estimated influence, we then present an imperceptible hierarchical selector and a profile modification method ProMix to select influential behaviors to poison. Notably, our attack is driven by custom attack objectives, which allows one to flexibly design different optimization goals to change the information flow, which could solve the blindness of existing influence maximization methods. Besides, behaviors such as retweeting are gentle and simple to implement. These properties make our attack more cost-effective and practical. Extensive experiments on two large-scale real-world datasets demonstrate the superiority of our method as it significantly outperforms baselines, and additionally, the proposed evaluator’s analysis of user influence provides new insights for influence maximization on social media.

By predicting a personalized ranking on a set of items, item recommendation helps users determine the information they need. While optimizing a ranking-focused loss is more in line with the objectives of item recommendation, previous studies have indicated that current sampling-based ranking methods don't always surpass non-sampling ones. This is because it is either inefficient to sample a pool of representative negatives for better generalization or challenging to gauge their contributions to ranking-focused losses accurately. To this end, we propose a novel weighted ranking loss, which weights each negative with the softmax probability based on model's predictive score. Our theoretical analysis suggests that optimizing this loss boosts the normalized discounted cumulative gain. Furthermore, it appears that this loss acts as an approximate analytic solution for adversarial training of personalized ranking. To improve optimization efficiency, we approximate the weighted ranking loss with self-normalized importance sampling and show that the loss has good generalization properties. To improve generalization, we further develop efficient cluster-based negative samplers based on clustering over item vectors, to decrease approximation error caused by the divergence between the proposal and the target distribution. Comprehensive evaluations on real-world datasets show that our methods remarkably outperform leading item recommendation algorithms.

With the explosive growth of scholarly information, researchers spend much time and effort copiously quoting authoritative works to support their ideas or motivations. We aim to alleviate this situation by proposing a citation recommendation strategy that recalls related papers for a rough idea (a piece of text, i.e., abstract, manuscript). However, the perspective of existing citation recommendations can not be well applied to our task for two defects. First, these methods neglect the reasoning of research topics, which makes the recommendation mechanism not meticulous enough and lacks explainability. For instance, they are not able to mine the hidden citing logic for the candidate paper while recommending. We fill the research gap by constructing structural topics consisting of knowledge concepts from the textual content, where reasoning paths between topics are extracted from an external knowledge graph. Second, the citation network is viewed as a crucial structural context to enhance the recommendation performance, but the new target idea does not have links to the citation network as published papers do. To simulate the prospective topological structure, our model, meanwhile, incorporates a contrastive-learning-based alignment paradigm to encourage the consistency of content embeddings and structure-oriented embeddings. We evaluate our proposed model on three real-world datasets and demonstrate that it significantly improves recommendation accuracy while providing high-quality knowledge-aware reasoning. And an interesting visual example illustrates the reasoning process when our model actually judges samples, which supports the feasibility of our topic-view learning paradigm.

Graph learning methods have achieved remarkable performance in various domains such as social recommendation, financial fraud detection, and so on. In real applications, the underlying graph is often dynamically evolving and thus, some recent studies focus on integrating the temporal topology information of graphs into the GNN for learning graph embedding. However, the robustness of training GNNs for dynamic graphs has not been discussed so far. The major reason is how to attack dynamic graph embedding still remains largely untouched, let alone how to defend against the attacks. To enable robust training of GNNs for dynamic graphs, in this paper, we investigate the problem of how to generate attacks and defend against attacks for dynamic graph embedding. Attacking dynamic graph embedding is more challenging than attacking static graph embedding as we need to understand the temporal dynamics of graphs as well as its impact on the embedding and the injected perturbations should be distinguished from the natural evolution. In addition, the defense is very challenging as the perturbations may be hidden within the natural evolution. To tackle these technical challenges, in this paper, we first develop a novel gradient-based attack method from an optimization perspective to generate perturbations to fool dynamic graph learning methods, where a key idea is to use gradient dynamics to attack the natural dynamics of the graph. Further, we borrow the idea of the attack method and integrate it with adversarial training to train a more robust dynamic graph learning method to defend against hand-crafted attacks. Finally, extensive experiments on two real-world datasets demonstrate the effectiveness of the proposed attack and defense method, where our defense method not only achieves comparable performance on clean graphs but also significantly increases the defense performance on attacked graphs.

Automatic analysis of image sentiment has gained considerable attention with the increasing throughput of user-generated visual contents online. Recently, researchers generally tend to design different Convolutional Neural Networks (CNNs) to extract image content features for sentiment analysis. However, they underestimated the importance of image color, which has been proved very crucial for image sentiment expressing by psychology and art theory. Moreover, we further observe that the coordination of content and color is the main form of image sentiment expressing. Different combinations of content and color could express extremely different sentiments. To that end, in this paper, we propose a Color Enhanced Cross Correlation Net (CECCN), a novel architecture for image sentiment analysis that not only leverages contents and colors simultaneously, but also takes their correlations into consideration. Specifically, we first use a pre-trained CNN to extract content features and color moment to collect color features from multiple color spaces. Then, we propose a novel Cross Correlation (CC) method to model the correlations between content features and color features with attention mechanism and sequence convolution, in which sentiment expressing of content and color can be enhanced by each other. Finally, we integrate these two types of information for better image sentiment analysis. Extensive experiments on two popular and well-studied benchmark datasets demonstrate the superiority and rationality of our proposed CECCN.

In recent years, the results of view-based 3D shape recognition methods have saturated, and models with excellent performance cannot be deployed on memory-limited devices due to their huge size of parameters. To address this problem, we introduce a compression method based on knowledge distillation for this field, which largely reduces the number of parameters while preserving model performance as much as possible. Specifically, to enhance the capabilities of smaller models, we design a high-performing large model called Group Multi-view Vision Transformer (GMViT). In GMViT, the view-level ViT first establishes relationships between view-level features. Additionally, to capture deeper features, we employ the grouping module to enhance view-level features into group-level features. Finally, the group-level ViT aggregates group-level features into complete, well-formed 3D shape descriptors. Notably, in both ViTs, we introduce spatial encoding of camera coordinates as innovative position embeddings. Furthermore, we propose two compressed versions based on GMViT, namely GMViT-simple and GMViT-mini. To enhance the training effectiveness of the small models, we introduce a knowledge distillation method throughout the GMViT process, where the key outputs of each GMViT component serve as distillation targets. Extensive experiments demonstrate the efficacy of the proposed method. The large model GMViT achieves excellent 3D classification and retrieval results on the benchmark datasets ModelNet, ShapeNetCore55, and MCB. The smaller models, GMViT-simple and GMViT-mini, reduce the parameter size by 8 and 17.6 times, respectively, and improve shape recognition speed by 1.5 times on average, while preserving at least 90% of the recognition performance.

Automatically solving math word problems (MWPs) is a challenging task for artificial intelligence (AI) and machine learning (ML) research, which aims to answer the problem with a mathematical expression. Many existing solutions simply model the MWP as a sequence of words, which is far from precise solving. To this end, we turn to how humans solve MWPs. Humans read the problem part-by-part and capture dependencies between words for a thorough understanding and infer the expression precisely in a goal-driven manner with knowledge. Moreover, humans can associate different MWPs to help solve the target with related experience. In this article, we present a focused study on an MWP solver by imitating such procedure. Specifically, we first propose a novel hierarchical math solver (HMS) to exploit semantics in one MWP. First, to imitate human reading habits, we propose a novel encoder to learn the semantics guided by dependencies between words following a hierarchical "word-clause-problem" paradigm. Next, we develop a goal-driven tree-based decoder with knowledge application to generate the expression. One step further, to imitate human associating different MWPs for related experience in problem-solving, we extend HMS to the Relation-enHanced Math Solver (RHMS) to utilize the relation between MWPs. First, to capture the structural similarity relation, we develop a meta-structure tool to measure the similarity based on the logical structure of MWPs and construct a graph to associate related MWPs. Then, based on the graph, we learn an improved solver to exploit related experience for higher accuracy and robustness. Finally, we conduct extensive experiments on two large datasets, which demonstrates the effectiveness of the two proposed methods and the superiority of RHMS.

Text classification is one of the fundamental tasks in natural language processing, which requires an agent to determine the most appropriate category for input sentences. Recently, deep neural networks have achieved impressive performance in this area, especially pretrained language models (PLMs). Usually, these methods concentrate on input sentences and corresponding semantic embedding generation. However, for another essential component: labels, most existing works either treat them as meaningless one-hot vectors or use vanilla embedding methods to learn label representations along with model training, underestimating the semantic information and guidance that these labels reveal. To alleviate this problem and better exploit label information, in this article, we employ self-supervised learning (SSL) in model learning process and design a novel self-supervised relation of relation ( [Formula: see text]) classification task for label utilization from a one-hot manner perspective. Then, we propose a novel relation of relation learning network( [Formula: see text]-Net) for text classification, in which text classification and [Formula: see text] classification are treated as optimization targets. Meanwhile, triplet loss is employed to enhance the analysis of differences and connections among labels. Moreover, considering that one-hot usage is still short of exploiting label information, we incorporate external knowledge from WordNet to obtain multiaspect descriptions for label semantic learning and extend [Formula: see text]-Net to a novel description-enhanced label embedding network(DELE) from a label embedding perspective. One step further, since these fine-grained descriptions may introduce unexpected noise, we develop a mutual interaction module to select appropriate parts from input sentences and labels simultaneously based on contrastive learning (CL) for noise mitigation. Extensive experiments on different text classification tasks reveal that [Formula: see text]-Net can effectively improve the classification performance and DELE can make better use of label information and further improve the performance. As a byproduct, we have released the codes to facilitate other research.

The recent development of recommender systems has a focus on collaborative ranking, which provides users with a sorted list rather than rating prediction. The sorted item lists can more directly reflect the preferences for users and usually perform better than rating prediction in practice. While considerable efforts have been made in this direction, the well-known pairwise and listwise approaches have still been limited by various challenges. Specifically, for the pairwise approaches, the assumption of independent pairwise preference is not always held in practice. Also, the listwise approaches cannot efficiently accommodate “ties” and unobserved data due to the precondition of the entire list permutation. To this end, in this article, we propose a novel setwise Bayesian approach for collaborative ranking, namely, SetRank, to inherently accommodate the characteristics of user feedback in recommender systems. SetRank aims to maximize the posterior probability of novel setwise preference structures and three implementations for SetRank are presented. We also theoretically prove that the bound of excess risk in SetRank can be proportional to \(\sqrt {M/N}\) , where M and N are the numbers of items and users, respectively. Finally, extensive experiments on four real-world datasets clearly validate the superiority of SetRank compared with various state-of-the-art baselines.

Career mobility analysis aims at understanding the occupational movement patterns of talents across distinct labor market entities, which enables a wide range of talent-centered applications, such as job recommendation, labor demand forecasting, and company competitive analysis. Existing studies in this field mainly focus on a single fixed scale, investigating either individual trajectories at the micro-level or crowd flows among market entities at the macro-level. Consequently, the intrinsic cross-scale interactions between talents and the labor market are largely overlooked. To bridge this gap, we propose UniTRep , a novel unified representation learning framework for cross-scale career mobility analysis. Specifically, we first introduce a trajectory hypergraph structure to organize the career mobility patterns in a low-information-loss manner, where market entities and talent trajectories are represented as nodes and hyperedges, respectively. Then, for learning the market-aware talent representations , we attentively propagate the node information to the hyperedges and incorporate the market contextual features into the process of individual trajectory modeling. For learning the trajectory-enhanced market representations , we aggregate the message from hyperedges associated with a specific node to integrate the fine-grained semantics of trajectories into labor market modeling. Moreover, we design two auxiliary tasks to optimize both intra-scale and cross-scale learning with a self-supervised strategy. Extensive experiments on a real-world dataset clearly validate that UniTRep can significantly outperform state-of-the-art baselines for various tasks.

The overwhelming surge of online video platforms has raised an urgent need for social interaction recognition techniques. Compared with simple short-term actions, long-term social interactions in semantic-rich videos could reflect more complicated semantics such as character relationships or emotions, which will better support various downstream applications, e.g., story summarization and fine-grained clip retrieval. However, considering the longer duration of social interactions with severe mutual overlap, involving multiple characters, dynamic scenes, and multi-modal cues, among other factors, traditional solutions for short-term action recognition may probably fail in this task. To address these challenges, in this article, we propose a hierarchical graph-based system, named InteractNet, to recognize social interactions in a multi-modal perspective. Specifically, our approach first generates a semantic graph for each sampled frame with integrating multi-modal cues and then learns the node representations as short-term interaction patterns via an adapted GCN module. Along this line, global interaction representations are accumulated through a sub-clip identification module, effectively filtering out irrelevant information and resolving temporal overlaps between interactions. In the end, the association among simultaneous interactions will be captured and modelled by constructing a global-level character-pair graph to predict the final social interactions. Comprehensive experiments on publicly available datasets demonstrate the effectiveness of our approach compared with state-of-the-art baseline methods.

Text-to-Video Retrieval is a typical cross-modal retrieval task that has been studied extensively under a conventional supervised setting. Recently, some works have sought to extend the problem to a weakly supervised formulation, which can be more consistent with real-life scenarios and more efficient in annotation cost. In this context, a new task called Partially Relevant Video Retrieval (PRVR) is proposed, which aims to retrieve videos that are partially relevant to a given textual query, i.e., the videos containing at least one semantically relevant moment. Formulating the task as a Multiple Instance Learning (MIL) ranking problem, prior arts rely on heuristics algorithms such as a simple greedy search strategy and deal with each query independently. Although these early explorations have achieved decent performance, they may not fully utilize the bag-level label and only consider the local optimum, which could result in suboptimal solutions and inferior final retrieval performance. To address this problem, in this paper, we propose to exploit the relationships between instances to boost retrieval performance. Based on this idea, we creatively put forward: (1) a new matching scheme for pairing queries and their related moments in the video; and (2) a new loss function to facilitate cross-modal alignment between two views of an instance. Extensive validations on three publicly available datasets have demonstrated the effectiveness of our solution and verified our hypothesis that modeling instance-level relationships is beneficial in the MIL ranking setting. Our code will be publicly available at https://github.com/xjtupanda/BGM-Net .

Unsupervised domain adaptation aims to transfer rich knowledge from the annotated source domain to the unlabeled target domain with the same label space. One prevalent genre is the bi-discriminator domain adversarial network, which concurrently considers the decision boundaries of both domains during the domain alignment. While effective, methods within this genre still contend with accuracy agnosticism toward the target domain and overconfident estimation in the source domain. Consequently, these two limitations hinder the effectiveness of both decision boundaries. To address the aforementioned challenges, we propose a novel bi-discriminator domain adversarial neural network, denoted as BACG. Specifically, for accuracy awareness of the target domain, we initially devise an optimizable nearest neighbor algorithm for acquiring pseudo-labels of samples, followed by class-level gradient alignment between two domains. This approach explicitly incorporates the accuracy signal from the target domain. To alleviate overconfident estimation in the source domain, we adopt evidential learning theory and develop a multinomial Dirichlet hierarchical model to infer both the classification probability and uncertainty. This approach not only ensures optimal assumptions for the source domain but also guarantees high-quality domain alignment. Additionally, to lower the time overhead caused by pseudo-label assignment, we introduce a memory bank-based variant, namely, fast-BACG, which effectively accelerates the training process at the expense of a slight reduction in accuracy. Extensive experiments and thorough analysis on four benchmark data sets validate the effectiveness and robustness of our algorithm.

The prosperity of knowledge graphs, as well as related downstream applications, has raised the urgent need for knowledge graph completion techniques that fully support knowledge graph reasoning tasks, especially under the circumstance of training data scarcity. Although large efforts have been made on solving this challenge via few-shot learning tools, they mainly focus on simply aggregating entity neighbors to represent few-shot references, whereas the enhancement from latent semantic correlation within neighbors has been largely ignored. To that end, in this article, we propose a novel few-shot learning solution named SIM, a S emantic I nteraction M atching network that applies a Transformer framework to enhance the entity representation with capturing semantic interaction between entity neighbors. Specifically, we first design an entity-relation fusion module to adaptively encode neighbors with incorporating relation representation. Along this line, Transformer layers are integrated to capture latent correlation within neighbors, as well as the semantic diversification of the support set. Finally, a similarity score is attentively estimated with the attention mechanism. Extensive experiments on two public benchmark datasets demonstrate that our model outperforms a variety of state-of-the-art methods by a significant margin.

Abstract The advent of large language models marks a revolutionary breakthrough in artificial intelligence. With the unprecedented scale of training and model parameters, the capability of large language models has been dramatically improved, leading to human-like performances in understanding, language synthesizing, common-sense reasoning, etc. Such a major leap forward in general AI capacity will fundamentally change the pattern of how personalization is conducted. For one thing, it will reform the way of interaction between humans and personalization systems. Instead of being a passive medium of information filtering, like conventional recommender systems and search engines, large language models present the foundation for active user engagement. On top of such a new foundation, users’ requests can be proactively explored, and users’ required information can be delivered in a natural, interactable, and explainable way. For another thing, it will also considerably expand the scope of personalization, making it grow from the sole function of collecting personalized information to the compound function of providing personalized services. By leveraging large language models as a general-purpose interface, the personalization systems may compile user’s requests into plans, calls the functions of external tools (e.g., search engines, calculators, service APIs, etc.) to execute the plans, and integrate the tools’ outputs to complete the end-to-end personalization tasks. Today, large language models are still being rapidly developed, whereas the application in personalization is largely unexplored. Therefore, we consider it to be right the time to review the challenges in personalization and the opportunities to address them with large language models. In particular, we dedicate this perspective paper to the discussion of the following aspects: the development and challenges for the existing personalization system, the newly emerged capabilities of large language models, and the potential ways of making use of large language models for personalization.

Personalized learning is a promising educational approach that aims to provide high-quality personalized services for each student with minimum demands for practice data. The key to achieving that lies in the cognitive diagnosis task, which estimates the cognitive state of the student through his/her logged data of doing practice quizzes. Nevertheless, in the personalized learning scenario, existing cognitive diagnosis models suffer from the inability to (1) quickly adapt to new students using a small amount of data, and (2) measure the reliability of the diagnosis result to avoid improper services that mismatch the student's actual state. In this paper, we propose a general Bayesian mETA-learned Cognitive Diagnosis framework (BETA-CD), which addresses the two challenges by prior knowledge exploitation and model uncertainty quantification, respectively. Specifically, we firstly introduce Bayesian hierarchical modeling to associate each student's cognitive state with a shared prior distribution encoding prior knowledge and a personal posterior distribution indicating model uncertainty. Furthermore, we formulate a meta-learning objective to automatically exploit prior knowledge from historical students, and efficiently solve it with a gradient-based variational inference method. The code will be publicly available at https://github.com/AyiStar/pyat.

Understanding mathematical questions effectively is a crucial task, which can benefit many applications, such as difficulty estimation. Researchers have drawn much attention to designing pre-training models for question representations due to the scarcity of human annotations (e.g., labeling difficulty). However, unlike general free-format texts (e.g., user comments), mathematical questions are generally designed with explicit purposes and mathematical logic, and usually consist of more complex content, such as formulas, and related mathematical knowledge (e.g., Function). Therefore, the problem of holistically representing mathematical questions remains underexplored. To this end, in this paper, we propose a novel contrastive pre-training approach for mathematical question representations, namely QuesCo, which attempts to bring questions with more similar purposes closer. Specifically, we first design two-level question augmentations, including content-level and structure-level, which generate literally diverse question pairs with similar purposes. Then, to fully exploit hierarchical information of knowledge concepts, we propose a knowledge hierarchy-aware rank strategy (KHAR), which ranks the similarities between questions in a fine-grained manner. Next, we adopt a ranking contrastive learning task to optimize our model based on the augmented and ranked questions. We conduct extensive experiments on two real-world mathematical datasets. The experimental results demonstrate the effectiveness of our model.

Maximum Inner Product Search (MIPS) plays an essential role in many applications ranging from information retrieval, recommender systems to natural language processing. However, exhaustive MIPS is often expensive and impractical when there are a large number of candidate items. The state-of-the-art quantization method of approximated MIPS is product quantization with a score-aware loss, developed by assuming that queries are uniformly distributed in the unit sphere. However, in real-world datasets, the above assumption about queries does not necessarily hold. To this end, we propose a quantization method based on the distribution of queries combined with sampled softmax. Further, we introduce a general framework encompassing the proposed method and multiple quantization methods, and we develop an effective optimization for the proposed general framework. The proposed method is evaluated on three real-world datasets. The experimental results show that it outperforms the state-of-the-art baselines.

Few-shot Named Entity Recognition (NER) task, which aims to identify and classify entities from different domains with limited training samples, has long been treated as a basic step for knowledge graph (KG) construction. Great efforts have been made on this task with competitive performance, however, they usually treat the two subtasks, namely span detection and type classification, as mutually independent, and the integrity and correlation between subtasks have been largely ignored. Moreover, prior arts may fail to absorb the coarse-grained features of entities, resulting in a semantic-insufficient representation of entity types. To that end, in this paper, we propose a Hierarchical Enhancing ProtoNet (HEProto) based on multi-task learning, which is utilized to jointly learn these two subtasks and model their correlation. Specifically, we adopt contrastive learning to enhance the span boundary information and the type semantic representations in these two subtasks. Then, the hierarchical prototypical network is designed to leverage the coarse-grained information of entities in the type classification stage, which could help the model to better learn the fine-grained semantic representations. Along this line, we construct a similarity margin loss to reduce the similarity between fine-grained entities and other irrelevant coarse-grained prototypes. Finally, extensive experiments on the Few-NERD dataset prove that our solution outperforms competitive baseline methods. The source code of HEProto is available at \hrefhttps://github.com/fanshu6hao/HEProto https://github.com/fanshu6hao/HEProto.

Computerized Adaptive Testing (CAT) arises as a promising personalized test mode in online education, targeting at revealing students' latent knowledge state by selecting test items adaptively. The item selection strategy is the core component of CAT, which searches for the best suitable test item based on students' current estimated ability at each test step. However, existing selection strategies behave in a brute-force manner, which results in the time complexity being linear to the number of items (N) in the item pool, i.e., O(N). Thus, in reality, the search latency becomes the bottleneck for CAT with a large-scale item pool. To this end, we propose a Search-Efficient Computerized Adaptive Testing framework (SECAT), which aims at enhancing CAT with an efficient selection strategy. Specifically, SECAT contains two main phases: item pool indexing and item search. In the item pool indexing phase, we apply a student-aware spatial partition method on the item pool to divide the test items into many sub-spaces, considering the adaptability of test items. In the item search phase, we optimize the traditional single-round search strategy with the asymptotic theory and propose a multi-round search strategy that can further improve the time efficiency. Compared with existing strategies, the time complexity of SECAT decreases from O(N) to O(logN). Across two real-world datasets, SECAT achieves over 200x speed up with negligible accuracy degradation.

The sequential recommendation system has been widely studied for its promising effectiveness in capturing dynamic preferences buried in users' sequential behaviors. Despite the considerable achievements, existing methods usually focus on intra-sequence modeling while overlooking exploiting global collaborative information by inter-sequence modeling, resulting in inferior recommendation performance. Therefore, previous works attempt to tackle this problem with a global collaborative item graph constructed by pre-defined rules. However, these methods neglect two crucial properties when capturing global collaborative information, i.e., adaptiveness and personalization, yielding sub-optimal user representations. To this end, we propose a graph-driven framework, named Adaptive and Personalized Graph Learning for Sequential Recommendation (APGL4SR), that incorporates adaptive and personalized global collaborative information into sequential recommendation systems. Specifically, we first learn an adaptive global graph among all items and capture global collaborative information with it in a self-supervised fashion, whose computational burden can be further alleviated by the proposed SVD-based accelerator. Furthermore, based on the graph, we propose to extract and utilize personalized item correlations in the form of relative positional encoding, which is a highly compatible manner of personalizing the utilization of global collaborative information. Finally, the entire framework is optimized in a multi-task learning paradigm, thus each part of APGL4SR can be mutually reinforced. As a generic framework, APGL4SR can not only outperform other baselines with significant margins, but also exhibit promising versatility, the ability to learn a meaningful global collaborative graph, and the ability to alleviate the dimensional collapse issue of item embeddings.

Researchers have successfully adapted the privacy-preserving Federated Learning (FL) to news recommendation tasks to better protect users' privacy, although typically at the cost of performance degradation due to the data heterogeneity issue. To address this issue, Personalized Federated Learning (PFL) has emerged, among which model interpolation is a promising approach that interpolates the local personalized models with the global model. However, the existing model interpolation method may not work well for news recommendation tasks for some reasons. First, it neglects the fine-grained personalization needs at both the temporal and spatial levels in news recommendation tasks. Second, due to the cold-user problem in real-world news recommendation tasks, the local personalized models may perform poorly, thus limiting the performance gain from model interpolation. To this end, we propose FINDING (Federated News Recommendation with Fine-grained Interpolation and Dynamic Clustering ), a novel personalized federated learning framework based on model interpolation. Specifically, we first propose the fine-grained model interpolation strategy which interpolates the local personalized models with the global model in a time-aware and layer-aware way. Then, to address the cold-user problem in news recommendation tasks, we adopt the group-level personalization approach where users are dynamically clustered into groups and the group-level personalized models are used for interpolation. Extensive experiments on two real-world datasets show that our method can effectively handle the above limitations of the current model interpolation method and alleviate the heterogeneity issue faced by traditional FL.

Federated learning (FL) is important for privacy-preserving services by training models without collecting raw user data. Most FL algorithms assume all data is annotated, which is impractical due to the high cost of labeling data in real applications. To alleviate the reliance on labeled data, semi-supervised federated learning (SSFL) has been proposed to utilize unlabeled data on clients to improve model performance. However, most existing methods either have privacy issues which share models trained on other clients, or generate pseudo-labels for unlabeled local datasets with the global model, which is usually biased towards the global data distribution. The latter may lead to sub-optimal accuracy of pseudo-labels, due to the gap between the local data distribution and the global model, especially in non-IID settings. In this paper, we propose a semi-supervised heterogeneous federated learning method with local knowledge enhancement, called FedLoKe, which aims to train an accurate global model from both labeled and unlabeled local data with non-IID distributions. Specifically, in FedLoKe, the server maintains a global model to capture global data distribution, and each client learns a local model to capture local data distribution. Since the distribution captured by the local model is aligned with the local data distribution, we utilize it to generate high-accuracy pseudo-labels of the unlabeled dataset for global model training. To prevent the local model from severely overfitting the small number of local labeled data, we further use the exponential moving average and apply the global model to generate pseudo-labels for local modeling training. Experiments on four datasets show the effectiveness of FedLoKe. Our code is available at: https://github.com/zcfinal/FedLoKe.

The fundamental task of intelligent educational systems is to offer adaptive learning services to students, such as exercise recommendations and computerized adaptive testing. However, optimizing required models in these systems would always encounter the collection difficulty of high-quality interaction data in practice. Therefore, establishing a student simulator is of great value since it can generate valid interactions to help optimize models. Existing advances have achieved success but generally suffer from exposure bias and overlook long-term intentions. To tackle these problems, we propose a novel Direct-Adversarial Imitation Student Simulator (DAISim) by formulating it as a Markov Decision Process (MDP), which unifies the workflow of the simulator in training and generating to alleviate the exposure bias and single-step optimization problems. To construct the intentions underlying the complex student interactions, we first propose a direct imitation strategy to mimic the interactions with a simple reward function. Then, we propose an adversarial imitation strategy to learn a rational distribution with the reward given by a parameterized discriminator. Furthermore, we optimize the discriminator in adversarial imitation in a pairwise manner, and the theoretical analysis shows that the pairwise discriminator would improve the generation quality. We conduct extensive experiments on real-world datasets, where the results demonstrate that our DAISim can simulate high-quality student interactions whose distribution is close to real distribution and can promote several downstream services.

Biological Knowledge Graphs (BKGs) can help to model complex biological systems in a structural way to support various tasks. Nevertheless, the incompleteness problem may limit the performance of existing BKGs, which still deserves new methods to reveal the missing relations. Though great efforts have been made to knowledge graph completion, existing methods are not easy to be adapted to the multimodal biological information such as molecular structures and textual descriptions. To this end, we propose a novel co-attention-based multimodal embedding framework, named CamE, for the multimodal BKG completion task. Specifically, we design a Triple Co-Attention (TCA) operator to capture and highlight the same semantic features among different modalities. Based on TCA, we further propose two components to handle multimodal fusion and multimodal entity-relation interaction, respectively. One is the multimodal TCA fusion module to achieve a multimodal joint representation for each entity in the BKG. It aims to project different modal information into a common space by capturing the same semantic features and overcoming the modality gap. The other is the relation-aware interactive TCA module to learn interactive representation by modelling the deep interaction between multimodal entities and relations. Extensive experiments on two real-world multimodal BKG datasets demonstrate that our method significantly outperforms several state-of-the-art baselines, including 10.3% and 16.2% improvement w.r.t MRR and Hits@1 metrics over its best competitors on public DRKG-MM dataset.

Collaborative filtering (CF) from implicit datasets has attracted much attention in recent years. The current mainstream pairwise methods optimize the Area Under the Curve (AUC) and are empirically proven to be helpful to exploit implicit feedback, but lead to not addressing the rank-biased scenarios where positive items are supposed to be placed on the top-k positions. Although there exist listwise methods, they have low efficiency and are not particularly adequate for general implicit feedback situations. To that end, in this paper, we propose a new framework, namely Collaborative List-and-Pairwise Filtering (CLAPF), which aims to introduce pairwise thinking into listwise methods. Specifically, we first smooth a well-known rank-biased measure called Mean Average Precision (MAP) as a low-bound version to make it can be optimized. After that, we combined the objective functions of optimizing the MAP with pairwise comparisons. The CLAPF framework is a new hybrid model that provides the idea of utilizing a listwise measure in a pairwise way on implicit feedback.

With the increasing demand for intelligent services of online video platforms, video character search task has attracted wide attention to support downstream applications like fine-grained retrieval and summarization. However, traditional solutions only focus on visual or coarse-grained social information and thus cannot perform well when facing complex scenes, such as changing camera view or character posture. Along this line, we leverage social information and scene context as prior knowledge to solve the problem of character search in complex scenes. Specifically, we propose a scene-prior-enhanced framework, named SoCoSearch. We first integrate multimodal clues for scene context to estimate the prior probability of social relationships, and then capture characters’ co-occurrence to generate an enhanced social context graph. Afterwards, we design a social context-aware GCN framework to achieve feature passing between characters to obtain robust representation for the character search task. Extensive experiments have validated the effectiveness of SoCoSearch in various metrics.

Automatic Micro-Expression (ME) spotting in long videos is a crucial step in ME analysis but also a challenging task due to the short duration and low intensity of MEs. When solving this problem, previous works generally lack in considering the structures of human faces and the correspondence between expressions and relevant facial muscles. To address this issue for better performance of ME spotting, this paper seeks to extract finer spatial features by modeling the relationships between facial Regions of Interest (ROIs). Specifically, we propose a graph convolutional-based network, called Action-Unit-aWare Graph Convolutional Network (AUW-GCN). In addition, to incorporate prior knowledge and address the issue of small datasets, AU-related statistics are encoded into the network. Comprehensive experiments show that our results outperform baseline methods consistently and achieve new SOTA performance in two benchmark datasets, CAS(ME)2and SAMM-LV. Our code is available at https://github.com/xjtupanda/AUW-GCN.

Automatic micro-expression recognition (MER) has essential applications in the psychological field. Graph-based models, due to their advantages in analyzing regionalized faces, have become a powerful method for MER. However, how to construct a graph from ME videos remains to be studied. To solve this problem, we design an adaptive graph attention network with temporal fusion to model the dynamic relationships between facial regions of interest (ROIs). Specifically, we first propose adaptive graph attention to establish learnable spatial graphs from ME videos. Then, we adopt an optical-flow-based feature as the suitable input for the graph network. In addition, an implicit semantic data augmentation algorithm is employed and improved as a data-driven weighted loss for better performance. Extensive experiments on SMIC-HS, CASME II and SAMM datasets have demonstrated the effectiveness of the proposed method, and it achieves to be the first graph-based model where UF1 and UAR both exceed 0.90 for 3-classes MER on CASME II. Code will be available at https://github.com/MEA-LAB-421/ICME2023-Recognition.

In this paper, we investigate the online allocation problem of maximizing the overall revenue subject to both lower and upper bound constraints. Compared to the extensively studied online problems with only resource upper bounds, the two-sided constraints affect the prospects of resource consumption more severely. As a result, only limited violations of constraints or pessimistic competitive bounds could be guaranteed. To tackle the challenge, we define a measure of feasibility $\xi^*$ to evaluate the hardness of this problem, and estimate this measurement by an optimization routine with theoretical guarantees. We propose an online algorithm adopting a constructive framework, where we initialize a threshold price vector using the estimation, then dynamically update the price vector and use it for decision-making at each step. It can be shown that the proposed algorithm is $\big(1-O(\frac{\varepsilon}{\xi^*-\varepsilon})\big)$ or $\big(1-O(\frac{\varepsilon}{\xi^*-\sqrt{\varepsilon}})\big)$ competitive with high probability for $\xi^*$ known or unknown respectively. To the best of our knowledge, this is the first result establishing a nearly optimal competitive algorithm for solving two-sided constrained online allocation problems with a high probability of feasibility.

Recently, Zero-Shot Node Classification (ZNC) has been an emerging and crucial task in graph data analysis. This task aims to predict nodes from unseen classes which are unobserved in the training process. Existing work mainly utilizes Graph Neural Networks (GNNs) to associate features' prototypes and labels' semantics thus enabling knowledge transfer from seen to unseen classes. However, the multi-faceted semantic orientation in the feature-semantic alignment has been neglected by previous work, i.e. the content of a node usually covers diverse topics that are relevant to the semantics of multiple labels. It's necessary to separate and judge the semantic factors that tremendously affect the cognitive ability to improve the generality of models. To this end, we propose a Knowledge-Aware Multi-Faceted framework (KMF) that enhances the richness of label semantics via the extracted KG (Knowledge Graph)-based topics. And then the content of each node is reconstructed to a topic-level representation that offers multi-faceted and fine-grained semantic relevancy to different labels. Due to the particularity of the graph's instance (i.e., node) representation, a novel geometric constraint is developed to alleviate the problem of prototype drift caused by node information aggregation. Finally, we conduct extensive experiments on several public graph datasets and design an application of zero-shot cross-domain recommendation. The quantitative results demonstrate both the effectiveness and generalization of KMF with the comparison of state-of-the-art baselines.

Cognitive Diagnosis aims to quantify the proficiency level of students on specific knowledge concepts. Existing studies merely leverage observed historical students-exercise interaction logs to access proficiency levels. Despite effectiveness, observed interactions usually exhibit a power-law distribution, where the long tail consisting of students with few records lacks supervision signals. This phenomenon leads to inferior diagnosis among few records students. In this paper, we propose the Exercise-aware Informative Response Sampling (EIRS) framework to address the long-tail problem. EIRS is a general framework that explores the partial order between observed and unobserved responses as auxiliary ranking-based training signals to supplement cognitive diagnosis. Considering the abundance and complexity of unobserved responses, we first design an Exercise-aware Candidates Selection module, which helps our framework produce reliable potential responses for effective supplementary training. Then, we develop an Expected Ability Change-weighted Informative Sampling strategy to adaptively sample informative potential responses that contribute greatly to model training. Experiments on real-world datasets demonstrate the supremacy of our framework in long-tailed data.

Vector retrieval focuses on finding the k-nearest neighbors from a bunch of data points, and is widely used in a diverse set of areas such as information retrieval and recommender system. The current state-of-the-art methods represented by HNSW usually generate indexes with a big memory footprint, restricting the scale of data they can handle, except resorting to a hybrid index with external storage. The space-partitioning learned indexes, which only occupy a small memory, have made great breakthroughs in recent years. However, these methods rely on a large amount of labeled data for supervised learning, so model complexity affects the generalization.

Finding multiple temporal relationships among locations can benefit a bunch of urban applications, such as dynamic offline advertising and smart public transport planning. While some efforts have been made on finding static relationships among locations, little attention is focused on studying time-aware location relationships. Indeed, abundant location-based human activities are time-varying and the availability of these data enables a new paradigm for understanding the dynamic relationships in a period among connective locations. To this end, we propose to study a new problem, namely multi-Temporal relationship inference among locations (Trial for short), where the major challenge is how to integrate dynamic and geographical influence under the relationship sparsity constraint. Specifically, we propose a solution to Trial with a graph learning scheme, which includes a spatially evolving graph neural network (SEENet) with two collaborative components: spatially evolving graph convolution module (SEConv) and spatially evolving self-supervised learning strategy (SE-SSL). SEConv performs the intra-time aggregation and inter-time propagation to capture the multifaceted spatially evolving contexts from the view of location message passing. In addition, SE-SSL designs time-aware self-supervised learning tasks in a global-local manner with additional evolving constraint to enhance the location representation learning and further handle the relationship sparsity. Finally, experiments on four real-world datasets demonstrate the superiority of our method over several state-of-the-art approaches.

Math formulas (e.g., "distance = speed X time'') serve as one of the fundamental commonsense knowledge in human cognition, where humans naturally acquire and manipulate them in logical thinking for mathematical reasoning problems. However, existing reasoning models mainly focus on learning heuristic linguistics or patterns to generate answers, but do not pay enough attention on learning with such formula knowledge. Thus, they are not transparent (thus uninterpretable) in terms of understanding and grasping basic mathematical logic. In this paper, to promote a step forward in the domain, we first construct two datasets (Math23K-F and MAWPS-F) with precise annotations of formula usage in each reasoning step for math word problems. Especially, our datasets are refined on the benchmark datasets, and thus ensure the generality and comparability for relevant research. Then, we propose a novel Formula-mastered Solver (FOMAS) with the guidance of mastering formula knowledge to solve the problems. Specifically, we establish FOMAS with two systems drawing insight from the dual process theory, including a Knowledge System and a Reasoning System, to learn and apply formula knowledge, respectively. The Knowledge System accumulates the math formulas, where we propose a novel pretraining manner to mimic how humans grasp the mathematical logic behind them. Then, in the Reasoning System, we develop elaborate formula-guided symbol prediction and goal generation methods that retrieve the necessary formula knowledge from Knowledge System to improve both reasoning accuracy and interpretability. It organically simulates how humans conduct complex reasoning under the explicit instruction of math formulas. Experimental results prove that FOMAS has a stronger reasoning ability and achieves a more interpretable reasoning process, which verifies the necessity of introducing formula knowledge transparently.

Multimodal entity linking (MEL) task, which aims at resolving ambiguous mentions to a multimodal knowledge graph, has attracted wide attention in recent years. Though large efforts have been made to explore the complementary effect among multiple modalities, however, they may fail to fully absorb the comprehensive expression of abbreviated textual context and implicit visual indication. Even worse, the inevitable noisy data may cause inconsistency of different modalities during the learning process, which severely degenerates the performance. To address the above issues, in this paper, we propose a novel Multi-GraIned Multimodal InteraCtion Network (MIMIC) framework for solving the MEL task. Specifically, the unified inputs of mentions and entities are first encoded by textual/visual encoders separately, to extract global descriptive features and local detailed features. Then, to derive the similarity matching score for each mention-entity pair, we device three interaction units to comprehensively explore the intra-modal interaction and inter-modal fusion among features of entities and mentions. In particular, three modules, namely the Text-based Global-Local interaction Unit (TGLU), Vision-based DuaL interaction Unit (VDLU) and Cross-Modal Fusion-based interaction Unit (CMFU) are designed to capture and integrate the fine-grained representation lying in abbreviated text and implicit visual cues. Afterwards, we introduce a unit-consistency objective function via contrastive learning to avoid inconsistency and model degradation. Experimental results on three public benchmark datasets demonstrate that our solution outperforms various state-of-the-art baselines, and ablation studies verify the effectiveness of designed modules.

Zero-Shot Learning (ZSL), which aims at automatically recognizing unseen objects, is a promising learning paradigm to understand new real-world knowledge for machines continuously. Recently, the Knowledge Graph (KG) has been proven as an effective scheme for handling the zero-shot task with large-scale and non-attribute data. Prior studies always embed relationships of seen and unseen objects into visual information from existing knowledge graphs to promote the cognitive ability of the unseen data. Actually, real-world knowledge is naturally formed by multimodal facts. Compared with ordinary structural knowledge from a graph perspective, multimodal KG can provide cognitive systems with fine-grained knowledge. For example, the text description and visual content can depict more critical details of a fact than only depending on knowledge triplets. Unfortunately, this multimodal fine-grained knowledge is largely unexploited due to the bottleneck of feature alignment between different modalities. To that end, we propose a multimodal intensive ZSL framework that matches regions of images with corresponding semantic embeddings via a designed dense attention module and self-calibration loss. It makes the semantic transfer process of our ZSL framework learns more differentiated knowledge between entities. Our model also gets rid of the performance limitation of only using rough global features. We conduct extensive experiments and evaluate our model on large-scale real-world data. The experimental results clearly demonstrate the effectiveness of the proposed model in standard zero-shot classification tasks.

This paper introduces a unified causal lens for understanding representative model interpretation methods. We show that their explanation scores align with the concept of average treatment effect in causal inference, which allows us to evaluate their relative strengths and limitations from a unified causal perspective. Based on our observations, we outline the major challenges in applying causal inference to model interpretation, including identifying common causes that can be generalized across instances and ensuring that explanations provide a complete causal explanation of model predictions. We then present CIMI, a Causality-Inspired Model Interpreter, which addresses these challenges. Our experiments show that CIMI provides more faithful and generalizable explanations with improved sampling efficiency, making it particularly suitable for larger pretrained models.

Exploring how learners' knowledge states evolve during the learning activities is a critical task in online learning systems, which can facilitate personalized services downstream, such as course recommendation. Most of existing methods have devoted great efforts to analyzing learners' knowledge states according to their responses (i.e., right or wrong) to different questions. However, the significant effect of learners' learning behaviors (e.g., answering speed, the number of attempts) is omitted, which can reflect their knowledge acquisition deeper and ensure the reliability of the response. In this paper, we propose a Learning Behavior-oriented Knowledge Tracing (LBKT) model, with the goal of explicitly exploring the learning behavior effects on learners' knowledge states. Specifically, we first analyze and summarize several dominated learning behaviors including Speed, Attempts and Hints in the learning process. As the characteristics of different learning behaviors vary greatly, we separately estimate their various effects on learners' knowledge acquisition in a quantitative manner. Then, considering that different learning behaviors are closely dependent with each other, we assess the fused effect of multiple learning behaviors by capturing their complex dependent patterns. Finally, we integrate the forgetting factor with learners' knowledge acquisition to comprehensively update their changing knowledge states in learning. Extensive experimental results on several public datasets demonstrate that our model generates better performance prediction for learners against existing methods. Moreover, LBKT shows good interpretability in tracking learners' knowledge state by incorporating the learning behavior effects. Our codes are available at https://github.com/xbh0720/LBKT.

Click-Through Rate (CTR) prediction of intelligent marketing systems is of great importance, in which feature interaction selection plays a key role. Most approaches model interactions of features by the same pre-defined operation under expert guidance, among which improper interactions may bring unnecessary noise and complicate the training process. To that end, in this paper, we aim to adaptively evolve the model to select proper operations to interact on feature pairs under task guidance. Inspired by natural evolution, we propose a general Cognitive EvoLutionary Search (CELS) framework, where cognitive ability refers to the malleability of organisms to orientate to the environment. Specifically, we conceptualize interactions as genomes, models as organisms, and tasks as natural environments. Mirroring how genetic malleability develops environmental adaptability, we thus diagnose the fitness of models to simulate the survival rates of organisms for natural selection, thereby an evolution path can be planned and visualized, offering an intuitive interpretation of the mechanisms underlying interaction modeling and selection. Based on the CELS framework, we develop four instantiations including individual-based search and population-based search. We demonstrate how individual mutation and population crossover enable CELS to evolve into diverse models suitable for various tasks and data, providing ready-to-use models. Extensive experiments on real-world datasets demonstrate that CELS significantly outperforms state-of-the-art approaches.

Deep learning models have progressively advanced time series forecasting due to their powerful capacity in capturing sequence dependence. Nevertheless, it is still challenging to make accurate predictions due to the existence of non-stationarity in real-world data, denoting the data distribution rapidly changes over time. To mitigate such a dilemma, several efforts have been conducted by reducing the non-stationarity with normalization operation. However, these methods typically overlook the distribution discrepancy between the input series and the horizon series, and assume that all time points within the same instance share the same statistical properties, which is too ideal and may lead to suboptimal relative improvements. To this end, we propose a novel slice-level adaptive normalization, referred to \textbf{SAN}, which is a novel scheme for empowering time series forecasting with more flexible normalization and denormalization. SAN includes two crucial designs. First, SAN tries to eliminate the non-stationarity of time series in units of a local temporal slice (i.e., sub-series) rather than a global instance. Second, SAN employs a slight network module to independently model the evolving trends of statistical properties of raw time series. Consequently, SAN could serve as a general model-agnostic plugin and better alleviate the impact of the non-stationary nature of time series data. We instantiate the proposed SAN on four widely used forecasting models and test their prediction results on benchmark datasets to evaluate its effectiveness. Also, we report some insightful findings to deeply analyze and understand our proposed SAN. We make our codes publicly available.

Lightweight compressed models are prevalent in Approximate Nearest Neighbor Search (ANNS) and Maximum Inner Product Search (MIPS) owing to their superiority of retrieval efficiency in large-scale datasets. However, results given by compressed methods are less accurate due to the curse of dimension and the limitations of optimization objectives (e.g., lacking interactions between queries and documents). Thus, we are encouraged to design a new learning algorithm for the compressed search index on high dimensions to improve retrieval performance. In this paper, we propose a novel KnowledgeDistillation for high dimensional search index framework (KDindex), with the aim of efficiently learning lightweight indexes by distilling knowledge from high-precision ANNS and MIPS models such as graph-based indexes. Specifically, the student is guided to keep the same ranking order of the top-k relevant results yielded by the teacher model, which acts as the additional supervision signals between queries and documents to learn the similarities between documents. Furthermore, to avoid the trivial solutions that all candidates are partitioned to the same centroid, the reconstruction loss that minimizes the compressed error, and the posting list balance strategy that equally allocates the candidates, are integrated into the learning objective. Experiment results demonstrate that KDindex outperforms existing learnable quantization-based indexes and is 40× lighter than the state-of-the-art non-exhaustive methods while achieving comparable recall quality.

User modeling techniques profile users' latent characteristics (e.g., preference) from their observed behaviors, and play a crucial role in decision-making. Unfortunately, traditional user models may unconsciously capture biases related to sensitive attributes (e.g., gender) from behavior data, even when this sensitive information is not explicitly provided. This can lead to unfair issues and discrimination against certain groups based on these sensitive attributes. Recent studies have been proposed to improve fairness by explicitly decorrelating user modeling results and sensitive attributes. However, most existing approaches assume that fully sensitive attribute labels are available in the training set, which is unrealistic due to collection limitations like privacy concerns, and hence bear the limitation of performance. In this paper, we focus on a practical situation with limited sensitive data and propose a novel FairLISA framework, which can efficiently utilize data with known and unknown sensitive attributes to facilitate fair model training. We first propose a novel theoretical perspective to build the relationship between data with both known and unknown sensitive attributes with the fairness objective. Then, based on this, we provide a general adversarial framework to effectively leverage the whole user data for fair user modeling. We conduct experiments on representative user modeling tasks including recommender system and cognitive diagnosis. The results demonstrate that our FairLISA can effectively improve fairness while retaining high accuracy in scenarios with different ratios of missing sensitive attributes.

Computerized adaptive testing (CAT), as a tool that can efficiently measure student's ability, has been widely used in various standardized tests (e.g., GMAT and GRE). The adaptivity of CAT refers to the selection of the most informative questions for each student, reducing test length. Existing CAT methods do not explicitly target ability estimation accuracy since there is no student's true ability as ground truth; therefore, these methods cannot be guaranteed to make the estimate converge to the true with such limited responses. In this paper, we analyze the statistical properties of estimation and find a theoretical approximation of the true ability: the ability estimated by full responses to question bank. Based on this, a Bounded Ability Estimation framework for CAT (BECAT) is proposed in a data-summary manner, which selects a question subset that closely matches the gradient of the full responses. Thus, we develop an expected gradient difference approximation to design a simple greedy selection algorithm, and show the rigorous theoretical and error upper-bound guarantees of its ability estimate. Experiments on both real-world and synthetic datasets, show that it can reach the same estimation accuracy using 15\% less questions on average, significantly reducing test length.

Crowdsourcing provides a practical approach for obtaining annotated data to train supervised learning models. However, since the crowd annotators may have different expertise domain and cannot always guarantee the high-quality annotations, learning from crowds generally suffers from the problem of unreliable results of introducing some noises, which makes it hard to achieve satisfying performance. In this work, we investigate the reliability of annotations to improve learning from crowds. Specifically, we first project annotator and data instance to factor vectors and model the complex interaction between annotator expertise and instance difficulty to predict annotation reliability. The learned reliability can be used to evaluate the quality of crowdsourced data directly. Then, we construct a new annotation, namely soft annotation, which serves as the gold label during the training. To recognize the different strengths of annotators, we model each annotator's confusion in an end-to-end manner. Extensive experimental results on three real-world datasets demonstrate the effectiveness of our method.

In recent years, pairwise methods, such as Bayesian Personalized Ranking (BPR), have gained significant attention in the field of collaborative filtering for recommendation systems. Group BPR is an extension of BPR that incorporates user groups to relax the strict assumption of independence between two users. However, the reliability of its user groups may be compromised as they only focus on a few behavioral similarities. To address this problem, this paper proposes a new entropy-weighted similarity measure for implicit feedback to quantify the relation between two users and sample like-minded user groups. We first introduce the group preference into several pairwise ranking algorithms and then utilize the entropy-weighted similarity to sample groups to further improve these algorithms. Unlike other approaches that rely solely on common item ratings, our method incorporates global information into the similarity measure, resulting in a more reliable approach to group sampling. We conducted experiments on two real-world datasets and evaluated our method using different metrics. The results show that our method can construct better user groups from sparse data and produce more accurate recommendations. Our approach can be applied to a wide range of recommendation systems, and this can significantly improve the performance of pairwise ranking algorithms, making it an effective tool for pairwise ranking.

A dozen recommendation libraries have recently been developed to accommodate popular recommendation algorithms for reproducibility. However, they are almost simply a collection of algorithms, overlooking the modularization of recommendation algorithms and their usage in practical scenarios. Algorithmic modularization has the following advantages: 1) helps to understand the effectiveness of each algorithm; 2) easily assembles new algorithms with well-performed modules by either drag-and-drop programming or automatic machine learning; 3) enables reinforcement between algorithms since one algorithm may act as a module of another algorithm. To this end, we develop a highly-modularized recommender system -- RecStudio, in which any recommendation algorithm is categorized into either a ranker or a retriever. In the RecStudio library, we implement 90 recommendation algorithms with the pure Pytorch, covering both common algorithms in other libraries and complex algorithms involving multiple recommendation models. RecStudio is featured from several perspectives, such as index-supported efficient recommendation and evaluation, GPU-accelerated negative sampling, hyperparameter learning on the validation, and cooperation between the retriever and ranker. RecStudio is also equipped with a web service, where the recommendation pipeline can be quickly established and visually evaluated on selected datasets, and the evaluation results are automatically archived and visualized in a leaderboard. The project and documents are released at http://recstudio.org.cn.

Automatically generating controllable and diverse mathematical word problems (MWPs) which conform to equations and topics is a crucial task in information retrieval and natural language generation. Recent deep learning models mainly focus on improving the problem readability but overlook the mathematical logic coherence, which tends to generate unsolvable problems. In this paper, we draw inspiration from the human problem-designing process and propose a Mathematical structure Planning and Knowledge enhanced Generation model (MaPKG), following the "plan-then-generate" steps. Specifically, we propose a novel dynamic planning module to make sentence-level equation plans and a dual-attention mechanism for word-level generation, incorporating equation structure representation and external commonsense knowledge. Extensive experiments on two MWP datasets show our model can guarantee more solvable, high-quality, and diverse problems. Our code is available at https://github.com/KenelmQLH/MaPKG.git

Dynamic share recommendation, which aims at recommending a friend who would like to share a particular item at a certain timestamp, has emerged as a novel task for social-oriented e-commerce platforms. Different from traditional graph-based recommendation tasks, with integrating the interconnected social interactions and fine-grained temporal information from historical share records, this novel task may encounter one unique challenge, i.e., how to deal with the dynamic social connections and asymmetric share interactions. Even worse, users may keep inactive during some periods, which results in difficulties in updating personalized profiles. To address the above challenges, in this paper, we propose a dynamic graph share recommendation model called DynShare. Specifically, we first divide each user embedding into two parts, namely the invitation embedding and vote embedding to show the tendencies of sending and receiving items, respectively. Then, temporal graph attention networks (TGATs) based on bi-directional continuous time dynamic graphs (CTDGs) are leveraged to encode temporal neighbor information from different directions. Afterward, to estimate how different users perceive the time intervals after the last interaction, we further design a time-interval aware personalized projection operator on the foundation of temporal point processes (TPPs) to project user embedding for the next-time share prediction. Extensive experiments on a real-world e-commerce share dataset have demonstrated that our proposed DynShare can achieve better results compared with state-of-the-art baseline methods. And our code is available on the project website: https://github.com/meteor-gif/DynShare.

Recent studies have demonstrated the vulnerability of recommender systems to membership inference attacks, which determine whether a user’s historical data was utilized for model training, posing serious privacy leakage issues. Existing works assumed that member and non-member users follow different recommendation modes, and then infer membership based on the difference vector between the user’s historical behaviors and the recommendation list. The previous frameworks are invalid against inductive recommendations, such as sequential recommendations, since the disparities of difference vectors constructed by the recommendations between members and non-members become imperceptible. This motivates us to dig deeper into the target model. In addition, most MIA frameworks assume that they can obtain some in-distribution data from the same distribution of the target data, which is hard to gain in recommender system.

Deep learning-based algorithms, e.g., convolutional networks, have significantly facilitated multivariate time series classification (MTSC) task. Nevertheless, they suffer from the limitation in modeling long-range dependence due to the nature of convolution operations. Recent advancements have shown the potential of transformers to capture long-range dependence. However, it would incur severe issues, such as fixed scale representations, temporal-invariant and quadratic time complexity, with transformers directly applicable to the MTSC task because of the distinct properties of time series data. To tackle these issues, we propose FormerTime, an hierarchical representation model for improving the classification capacity for the MTSC task. In the proposed FormerTime, we employ a hierarchical network architecture to perform multi-scale feature maps. Besides, a novel transformer encoder is further designed, in which an efficient temporal reduction attention layer and a well-informed contextual positional encoding generating strategy are developed. To sum up, FormerTime exhibits three aspects of merits: (1) learning hierarchical multi-scale representations from time series data, (2) inheriting the strength of both transformers and convolutional networks, and (3) tacking the efficiency challenges incurred by the self-attention mechanism. Extensive experiments performed on 10 publicly available datasets from UEA archive verify the superiorities of the FormerTime compared to previous competitive baselines.

The Embarrassingly Shallow Autoencoders (EASE and SLIM) are strong recommendation methods based on implicit feedback, compared to competing methods like iALS and VAE-CF. However, EASE suffers from several major shortcomings. First, the training and inference of EASE can not scale with the increasing number of items since it requires storing and inverting a large dense matrix; Second, though its optimization objective – the square loss– can yield a closed-form solution, it is not consistent with recommendation goal – predicting a personalized ranking on a set of items, so that its performance is far from optimal w.r.t ranking-oriented recommendation metrics. Finally, the regularization coefficients are sensitive w.r.t recommendation accuracy and vary a lot across different datasets, so the fine-tuning of these parameters is important yet time-consuming. To improve training and inference efficiency, we propose a Similarity-Structure Aware Shallow Autoencoder on top of three similarity structures, including Co-Occurrence, KNN and NSW. We then optimize the model with a weighted square loss, which is proven effective for ranking-based recommendation but still capable of deriving closed-form solutions. However, the weight in the loss can not be learned in the training set and is similarly sensitive w.r.t the accuracy to regularization coefficients. To automatically tune the hyperparameters, we design two validation losses on the validation set for guidance, and update the hyperparameters with the gradient of the validation losses. We finally evaluate the proposed method on multiple real-world datasets and show that it outperforms seven competing baselines remarkably, and verify the effectiveness of each part in the proposed method.

In online games, predicting massive battle outcomes is a fundamental task of many applications, such as team optimization and tactical formulation. Existing works do not pay adequate attention to the massive battle. They either seek to evaluate individuals in isolation or mine simple pair-wise interactions between individuals, neither of which effectively captures the intricate interactions between massive units (e.g., individuals). Furthermore, as the team size increases, the phenomenon of diminishing marginal utility of units emerges. Such a diminishing pattern is rarely noticed in previous work, and how to capture it from data remains a challenge. To this end, we propose a novel Massive battle outcome predictor with margiNal Effect modules, namely MassNE, which comprehensively incorporates individual effects, cooperation effects (i.e., intra-team interactions) and suppression effects (i.e., inter-team interactions) for predicting battle outcomes. Specifically, we design marginal effect modules to learn how units’ marginal utility changing respect to their number, where the monotonicity assumption is applied to ensure rationality. In addition, we evaluate the current classical models and provide mathematical proofs that MassNE is able to generalize several earlier works in massive settings. Massive battle datasets generated by StarCraft II APIs are adopted to evaluate the performances of MassNE. Extensive experiments empirically demonstrate the effectiveness of MassNE, and MassNE can reveal reasonable cooperation effects, suppression effects, and marginal utilities of combat units from the data.

Deep recommender systems (DRS) are intensively applied in modern web services. To deal with the massive web contents, DRS employs a two-stage workflow: retrieval and ranking, to generate its recommendation results. The retriever aims to select a small set of relevant candidates from the entire items with high efficiency; while the ranker, usually more precise but time-consuming, is supposed to further refine the best items from the retrieved candidates. Traditionally, the two components are trained either independently or within a simple cascading pipeline, which is prone to poor collaboration effect. Though some latest works suggested to train retriever and ranker jointly, there still exist many severe limitations: item distribution shift between training and inference, false negative, and misalignment of ranking order. As such, it remains to explore effective collaborations between retriever and ranker.

Vector quantization techniques, such as Product Quantization (PQ), play a vital role in approximate nearest neighbor search (ANNs) and maximum inner product search (MIPS) owing to their remarkable search and storage efficiency. However, the indexes in vector quantization cannot be trained together with the inference models since data indexing is not differentiable. To this end, differentiable vector quantization approaches, such as DiffPQ and DeepPQ, have been recently proposed, but existing methods have two drawbacks. First, they do not impose any constraints on codebooks, such that the resultant codebooks lack diversity, leading to limited retrieval performance. Second, since data indexing resorts to operator, differentiability is usually achieved by either relaxation or Straight-Through Estimation (STE), which leads to biased gradient and slow convergence. To address these problems, we propose a Differentiable Optimized Product Quantization method (DOPQ) and beyond in this paper. Particularly, each data is projected into multiple orthogonal spaces, to generate multiple views of data. Thus, each codebook is learned with one view of data, guaranteeing the diversity of codebooks. Moreover, instead of simple differentiable relaxation, DOPQ optimizes the loss based on direct loss minimization, significantly reducing the gradient bias problem. Finally, DOPQ is evaluated with seven datasets of both recommendation and image search tasks. Extensive experimental results show that DOPQ outperforms state-of-the-art baselines by a large margin.

Knowledge Tracing (KT) aims at tracing the evolution of the knowledge states along the learning process of a learner. It has become a crucial task for online learning systems to model the learning process of their users, and further provide their users a personalized learning guidance. However, recent developments in KT based on deep neural networks mostly focus on increasing the accuracy of predicting the next performance of students. We argue that current KT modeling, as well as training paradigm, can lead to models tracing patterns of learner’s learning activities, instead of their evolving knowledge states. In this paper, we propose a new architecture, Diagnostic Transformer (DTransformer), along with a new training paradigm, to tackle this challenge. With DTransformer, we build the architecture from question-level to knowledge-level, explicitly diagnosing learner’s knowledge proficiency from each question mastery states. We also propose a novel training algorithm based on contrastive learning that focuses on maintaining the stability of the knowledge state diagnosis. Through extensive experiments, we will show that with its understanding of knowledge state evolution, DTransformer achieves a better performance prediction accuracy and more stable knowledge state tracing results. We will also show that DTransformer is less sensitive to specific patterns with case study. We open-sourced our code and data at https://github.com/yxonic/DTransformer.

Recent studies have shown that recommender systems are vulnerable, and it is easy for attackers to inject well-designed malicious profiles into the system, resulting in biased recommendations. We cannot deprive these data's injection right and deny their existence's rationality, making it imperative to study recommendation robustness. Despite impressive emerging work, threat assessment of the bi-level poisoning problem and the imperceptibility of poisoning users remain key challenges to be solved. To this end, we propose Infmix, an efficient poisoning attack strategy. Specifically, Infmix consists of an influence-based threat estimator and a user generator, Usermix. First, the influence-based estimator can efficiently evaluate the user's harm to the recommender system without retraining, which is challenging for existing attacks. Second, Usermix, a distribution-agnostic generator, can generate unnoticeable fake data even with a few known users. Under the guidance of the threat estimator, Infmix can select the users with large attacking impacts from the quasi-real candidates generated by Usermix. Extensive experiments demonstrate Infmix's superiority by attacking six recommendation systems with four real datasets. Additionally, we propose a novel defense strategy, adversarial poisoning training (APT). It mimics the poisoning process by injecting fake users (ERM users) committed to minimizing empirical risk to build a robust system. Similar to Infmix, we also utilize the influence function to solve the bi-level optimization challenge of generating ERM users. Although the idea of "fighting fire with fire" in APT seems counterintuitive, we prove its effectiveness in improving recommendation robustness through theoretical analysis and empirical experiments.

In this article, aiming at a Chinese keyword-based book search service, from a technological perspective, we propose to modify a user query sequence carefully to confuse the user query topics and thus protect the user topic privacy on the untrusted server, without compromising the accuracy of each book search service. First, we propose a client-based framework for the privacy protection of book search, and then a privacy model to formulate the constraints in terms of accuracy, efficiency, and security, which the cover queries generated based on a user query sequence should meet. Second, we present a modification algorithm for a user query sequence, based on some heuristic strategies, which can quickly generate a cover query sequence meeting the privacy model by replacing, deleting, and adding keywords for each user query. Finally, both theoretical analysis and experimental evaluation demonstrate the effectiveness of the proposed approach, i.e., which can improve the security of users’ topic privacy on the untrusted server without compromising the efficiency, accuracy, and usability of an existing Chinese keyword book search service, so it has a positive impact for the construction of a privacy-preserving text retrieval platform under an untrusted network environment.

Recent years have witnessed the booming of online social media platforms with embracing the popular service called “Time-Sync Comment”, which supports the viewers to share their time-sync opinions along with video content. In this way, we observe that numerous semantically-altered terms, or “Memes”, were created by niche users to express their unique ideas and emotions, and further attracted a large group of viewers with better activity and enthusiasm. Unfortunately, since the memes were created based on domain-specific knowledge and semantically varied depending on the multimodal context in videos, newcomers may fail to comprehend the semantic connotation of memes, which may severely impair their user-experiences. To deal with this issue, in this article, we propose a novel meme explanation framework, called ProMDE, to automatically capture and comprehend the memes in time-sync comments, which could further benefit the viewers with meme explanation service. Specifically, we first iteratively reconstruct the original time-sync comments compared with visual embedding to detect the semantically-altered terms as meme candidates. Afterward, based on the guides from the domain-specific corpus, visual and textual features will be fused to represent the context-aware multimodal cues. Moreover, to accurately describe the commonly-seen homophones in memes, i.e., they have the same pronunciation but different word-spelling expressions, we integrate the phonetic symbols as an additional modality to enhance the framework. Finally, we utilize a Transformer-based decoder to generate the natural language explanation for captured memes. Extensive experiments on a large real-world dataset prove that our framework could significantly outperform several state-of-the-art baseline methods, demonstrating the efficacy of modeling multimodal context and pronunciation for meme detection and explanation.

Enterprise competition analysis has long been treated as a crucial task of management science, which can reveal pertinent information about market saturation and business opportunities to support the decision-making process of entrepreneurs and investors. Recently, with the development of graph representation techniques, enterprises could be now formulated in a novel graph-oriented perspective to model their cooperation and competition in a directional graph. However, these prior arts mainly treat the enterprises as individual nodes, while the ecosystems, i.e., enterprise groups formed based on their common interests, and consequent interaction effects have been largely ignored. To address this issue, in this article, by adapting the concept of hypergraph to describe enterprise ecosystems, we propose a novel multiview enterprise relation network (MERN) framework for enterprise competition analysis. Specifically, we first put forward a hypertranslating embedding (HTransE) algorithm inspired by translation distance models for the hyperrelation embedding of ecosystems. Meanwhile, a relational graph convolutional network (RGCN) is designed for ordinary relation embedding, e.g., supplier, client, or competitor. Afterward, considering the commercial diversity in ecosystems, which means that different enterprises may focus on different industry fields, we propose an industry trend embedding module to describe industry distribution and development trend factors for each enterprise. Finally, we apply the self-attention mechanism to integrate these three modules and adapt the bilinear model DistMult for competition link analysis. Extensive experiments on a real-world dataset validate that our solution can achieve better performance compared with several competitive baselines.

Saturation information in hazy images is conducive to effective haze removal, However, existing saturation-based dehazing methods just focus on the saturation value of each pixel itself, while the higher-level distribution characteristic between pixels regarding saturation remains to be harnessed. In this paper, we observe that the pixels, which share the same surface reflectance coefficient in the local patches of haze-free images, exhibit a linear relationship between their saturation component and the reciprocal of their brightness component in the corresponding hazy images normalized by atmospheric light. Furthermore, the intercept of the line described by this linear relationship on the saturation axis is exactly the saturation value of these pixels in the haze-free images. Using this characteristic of saturation, termed saturation line prior (SLP), the transmission estimation is translated into the construction of saturation lines. Accordingly, a new dehazing framework using SLP is proposed, which employs the intrinsic relevance between pixels to achieve a reliable saturation line construction for transmission estimation. This approach can recover the fine details and attain realistic colors from hazy scenes, resulting in a remarkable visibility improvement. Extensive experiments in real-world and synthetic hazy images show that the proposed method performs favorably against state-of-the-art dehazing methods. Code is available on https://github.com/LPengYang/Saturation-Line-Prior.

Recently, learning-based multi-exposure fusion (MEF) methods have made significant improvements. However, these methods mainly focus on static scenes and are prone to generate ghosting artifacts when tackling a more common scenario, i.e., the input images include motion, due to the lack of a benchmark dataset and solution for dynamic scenes. In this paper, we fill this gap by creating an MEF dataset of dynamic scenes, which contains multi-exposure image sequences and their corresponding high-quality reference images. To construct such a dataset, we propose a 'static-for-dynamic' strategy to obtain multi-exposure sequences with motions and their corresponding reference images. To the best of our knowledge, this is the first MEF dataset of dynamic scenes. Correspondingly, we propose a deep dynamic MEF (DDMEF) framework to reconstruct a ghost-free high-quality image from only two differently exposed images of a dynamic scene. DDMEF is achieved through two steps: pre-enhancement-based alignment and privilege-information-guided fusion. The former pre-enhances the input images before alignment, which helps to address the misalignments caused by the significant exposure difference. The latter introduces a privilege distillation scheme with an information attention transfer loss, which effectively improves the deghosting ability of the fusion network. Extensive qualitative and quantitative experimental results show that the proposed method outperforms state-of-the-art dynamic MEF methods. The source code and dataset are released at https://github.com/Tx000/Deep_dynamicMEF.

In recent years, due to the explosive growth of patent applications, patent mining has drawn extensive attention and interest. An important issue of patent mining is that of recognizing the technologies contained in patents, which serves as a fundamental preparation for deeper analysis. To this end, in this article, we make a focused study on constructing a technology portrait for each patent, i.e., to recognize technical phrases concerned in it, which can summarize and represent patents from a technical perspective. Along this line, a critical challenge is how to analyze the unique characteristics of technical phrases and illustrate them with definite descriptions. Therefore, we first generate the detailed descriptions about the technical phrases existing in extensive patents based on different criteria, including various previous works, practical experience, and statistical analyses. Then, considering the unique characteristics of technical phrases and the complex structure of patent documents, such as multi-aspect semantics and multi-level relevances, we further propose a novel unsupervised model, namely TechPat, which can not only automatically recognize technical phrases from massive patents but also avoid the need for expensive human labeling. After that, we evaluate the extraction results from various aspects. Specifically, we propose a novel evaluation metric called Information Retrieval Efficiency (IRE) to quantify the performance of extracted technical phrases from a new perspective. Extensive experiments on real-world patent data demonstrate that the TechPat model can effectively discriminate technical phrases in patents and greatly outperform existing methods. We further apply extracted technical phrases to two practical application tasks, namely patent search and patent classification, where the experimental results confirm the wide application prospects of technical phrases. Finally, we discuss the generalization ability of our proposed methods.

Graph Neural Networks (GNNs) such as Graph Convolutional Networks (GCNs) can effectively learn node representations via aggregating neighbors based on the relation graph. However, despite a few exceptions, most of the previous work in this line does not consider the topical semantics underlying the edges, making the node representations less effective and the learned relation between nodes hard to explain. For instance, the current GNNs make us usually don’t know what is the reason for the connection of network nodes, such as the specific research topics cited in this article and the concerns among friends on social platforms. Some methods have begun to explore the extraction of relation semantics in recent related literature, but existing studies generally face two bottlenecks, i.e., either being unable to explain the mined latent relations to ensure their reasonableness and independence, or demanding the textual content of edges which is unavailable in most real-world datasets. Actually, these two issues are both crucial in practical use. In our work, we propose a novel Topic-Disentangled Graph Neural Network (TDG) to address the above two issues at the same time, which explores the relation topics from the perspective of node contents. We design an optimized graph topic module to handle node features to construct independent and explainable semantic subspaces, then the reasonable relation topics that correspond to these subspaces are assigned to each graph relation via a neighborhood routing mechanism. Our proposed model can be easily combined with related graph tasks to form an end-to-end model, to avoid the risk of deviation between node representation space and task space. To evaluate the efficiency of our model, sufficient node-related tasks are conducted on three public datasets in the experimental section. The results show the obvious superiority of TDG compared with the state-of-the-art models.

With the growing popularity of social media, malicious users (spammers) unfairly overpower legitimate users with unwanted or fake content to achieve their illegal purposes, which encourages research on spammer detection. The existing spammer detection methods can be characterized into feature-based detection and propagation-based detection. However, feature-based methods (e.g., GCN) cannot capture the user’s following relations, while propagation-based methods cannot utilize the rich text features. To this end, we consider combining these two methods and propose an Adaptive Reward Markov Random Field (ARMRF) layer. ARMRF layer models three intuitions on user label relations and assign them different learnable rewards. Besides, we learn the reward weights by stacking the ARMRF layer on top of GCN for end-to-end training, and we call the stacked model ARMGCN. To further improve the expressive power of ARMGCN, we propose the Markov-Driven Graph Convolutional Network (MDGCN), which integrates conditional random fields (CRF) and ARMGCN. CRF establishes the label joint probability distribution conditioned features for learning user dependencies, and the distribution can be optimized by a variational EM algorithm. We extensively evaluate the proposed method on two real-world Twitter datasets, and the experimental results demonstrate that MDGCN outperforms the state-of-the-art baselines. In addition, the ARMRF layer is model-independent, so it can be integrated with existing advanced detection methods to improve detection performance further.

Hierarchical multi-label classification (HMC) deals with the problem of assigning each entity to multiple classes with a taxonomic structure (e.g., tree). Within this structure, classes at different levels tend to have dependencies under the hierarchy constraints. However, most prior studies for HMC tasks tend to ignore the class dependencies within the hierarchy. Moreover, most existing methods generate incoherent predictions and do not satisfy the hierarchy constraint. To this end, based on previously developed HARNN, we propose a general framework, HmcNet, for introducing explicit and implicit class hierarchy constraints to generate coherent predictions. We develop an efficient Prune-based Coherent Prediction (PCP) strategy for the optimal paths selection, which produces coherent predictions in a principled way. HmcNet can be well explained from two perspectives. First, it develops the Hierarchical Attention-based Memory (HAM) unit with implicit class hierarchy constraints to capture class dependencies more intuitively; Second, it subsumes explicit class hierarchy constraints during training and inference phases and generates coherent predictions in a consistent manner. Finally, extensive experimental results on six real-world datasets demonstrate the effectiveness and interpretability of the HmcNet frameworks. To facilitate future research, our code has been made publicly available.

Information network embedding is an important way to enable efficient graph analytics. However, it still faces with computational challenges in problems such as link prediction and node recommendation, particularly with the increasing scale of networks. Both hashing and quantization are promising approaches for accelerating these problems by orders of magnitude. In the preliminary work, we have proposed to learn binary codes for information networks, but graph analytics may suffer from large accuracy degradation. To reduce information loss while achieving memory and search efficiency, we further propose to learn quantized codes for information networks. In particular, each node is represented by compositing multiple latent vectors, each of which is optimally selected from a distinct set. Since (generalized) matrix factorization unifies several well-known embedding methods with high-order proximity preserved, we propose a \underline{N}etwork \underline{R}epresentation \underline{L}ightening framework based on \underline{M}atrix \underline{F}actorization (NRL-MF) to learn binary and quantized codes. We also propose an alternating optimization algorithm for efficient parameter learning, even for the generalized matrix factorization case. We finally evaluate NRL-MF on four real-world information network datasets with respect to the tasks of node classification and node recommendation. The results show that NRL-MF significantly outperforms competing baselines in both tasks, and that quantized representations indeed incur much smaller information loss than binarized codes.

A job description shows the responsibilities of the job position and the skill requirements for the job. An effective job description will help employers to identify the right talents for the job, and give a clear understanding to candidates of what their duties and qualifications for a particular position would be. In this paper, we investigate how to automate the process to generate job descriptions with less human intervention. We propose an end-to-end capability-aware neural job description generation framework, namely Cajon, to facilitate the writing of job description. Specifically, we first propose a novel capability-aware neural topic model to distill the various capability information from the larger-scale recruitment data. Also, an encoder-decoder recurrent neural network is designed for enabling the job description generation. In particular, the capability-aware attention and copy mechanisms are proposed to guide the generation process to ensure the generated job descriptions can comprehensively cover relevant and representative capability requirements for the job. Moreover, we propose a capability-aware policy gradient training algorithm to further enhance the rationality of the generated job description. Finally, extensive experiments on real-world recruitment data clearly show our Cajon framework can help to generate more effective job descriptions in an interpretable way

Reading psychology believes text comprehension to involve a complex psychological construction process, with the reader mind being a dynamic associative system that stores an abundance of schemata. For Chinese text, in particular, the unique ideographic writing system allows its lansign to trigger semantic association and schema recalling without the need of phonetics. In contrast to previous research efforts on text classification problems, in this paper we present an interdisciplinary modeling approach that draws inspirations from the cognitive principles of ideography, schema theory and deep learning to study Chinese text classification. Specifically, we first propose a Radical-guided Associative Model (RAM) for preliminary cognitive imitation, which comprises two coupled spaces, namely the Literal Space and Associative Space. Then, taking consideration of the schemata acquired from the mind of a reader which plays a important role in influencing text-dependent information revision, we extend RAM with a systematic Schema-aware Radical-guided Associative Model (SRAM) that embeds label semantics as essential text-independent human knowledge for real-world abstraction. In SRAM, the Schema Space is introduced and a Schema Attention module is proposed with a novel loss paradigm that includes the linkage and interaction between text-dependent prior concepts and text-independent label schemata. Extensive experiments on three real-world datasets demonstrate the effectiveness and rationality of our proposed method.

Cognitive diagnosis is widely applicable in the scenarios where users’ cognitive states need to be assessed, such as games and clinical measurement. Especially in intelligent education, which has become increasingly popular recent decades, cognitive diagnosis serves as a fundamental module for discovering the proficiency level of students on specific knowledge concepts. Existing approaches usually mine linear interactions of student exercising process by manually designed function (e.g., logistic function). However, the cognitive interactions between students and exercises is a complex process, and excessive simplifications would lead to under fitting and thus get inaccurate diagnostic results. Besides, the manually designed interaction functions are relatively inflexible and limits their extensibility. This consequently causes lack of consideration about useful non-numerical information in the cognitive process besides response logs. In this article, we propose a general Neural Cognitive Diagnosis (NeuralCD) framework as well as several implemented models (a basic implementation NeuralCDM and three extensions), where we project students and exercises to factor vectors and incorporates neural networks to learn the complex exercising interactions. To ensure the interpretability of diagnostic results, which is essential for cognitive diagnosis, we apply an monotonicity assumption to our NeuralCD framework. Moreover, NeuralCD is a general framework and has good extensibility. We show the generality of NeuralCD through proving how it can cover some traditional models. Then, we demonstrate the extensibility of NeuralCD, which benefits future developments. On one hand, we demonstrate content-based extensions where we provide examples of exploring the rich contents of exercise texts (CNCD-Q and CNCD-F). On the other hand, we demonstrate a knowledge-association based extension to show that NeuralCD is flexible for structural adjustments so as to solve specific problems. For instance, we improve the diagnostic results on uncovered knowledge concepts of a student by extending NeuralCD with the knowledge associations consideration (KaNCD). Extensive experimental results on real-world datasets show the effectiveness of NeuralCD framework with both accuracy and interpretability.

Network representation learning or network embedding has attracted wide attention due to its effectiveness on various network-oriented applications in recent years. Though large efforts have been made, they usually model the interactions between nodes reflected by network structure and attributes in a coupled way. To this end, in this article, we comprehensively study the problem of learning attributed network embedding, which focuses on characterizing different types of interactions among nodes and alleviating the sparse attribute problem as well. Specifically, we propose a novel DeCoupled Network Embedding (DCNE) model to learn node representations in a unified framework. We first respectively project both nodes and attributes into low-dimensional vectorial space. Then, we introduce a novel ‘`decoupled-fusion’' learning process into each graph layer to iteratively generate the node embeddings. In particular, we propose two adapted graph convolution modules to decouple the learning of network structure and attributes respectively, and a fusion module to adaptively aggregate the information. Next, we adopt a modified mini-batch algorithm to iteratively aggregate the higher-order information of both nodes and attributes within a multi-task learning framework. Extensive experiments on five public datasets demonstrate that DCNE could outperform state-of-the-art methods on multiple benchmark tasks.

Many data mining tasks rely on graphs to model relational structures among individuals (nodes). Since relational data are often sensitive, there is an urgent need to evaluate the privacy risks in graph data. One famous privacy attack against data analysis models is the model inversion attack, which aims to infer sensitive data in the training dataset and leads to great privacy concerns. Despite its success in grid-like domains, directly applying model inversion attacks on non-grid domains such as graph leads to poor attack performance. This is mainly due to the failure to consider the unique properties of graphs. To bridge this gap, we conduct a systematic study on model inversion attacks against Graph Neural Networks (GNNs), one of the state-of-the-art graph analysis tools in this paper. First, in the white-box setting where the attacker has full access to the target GNN model, we present GraphMI to infer the private training graph data. Specifically in GraphMI, a projected gradient module is proposed to tackle the discreteness of graph edges and preserve the sparsity and smoothness of graph features; a graph auto-encoder module is used to efficiently exploit graph topology, node attributes, and target model parameters for edge inference; a random sampling module can finally sample discrete edges. Furthermore, in the hard-label black-box setting where the attacker can only query the GNN API and receive the classification results, we propose two methods based on gradient estimation and reinforcement learning (RL-GraphMI). With the proposed methods, we study the connection between model inversion risk and edge influence and show that edges with greater influence are more likely to be recovered. Extensive experiments over several public datasets demonstrate the effectiveness of our methods. We also evaluate our attacks under two defenses: one is the well-designed differential private training, and the other is graph preprocessing. Our experimental results show that such defenses are not sufficiently effective and call for more advanced defenses against privacy attacks.

Aspect-level sentiment analysis is a granular emotional classification task that refers to identifying sentiment polarities towards aspects in a sentence. Although previous research has reached a great achievement, this task remains very challenging. First, previous approaches only focus on one specific domain, which lacks the capability of transferring to other domains. Moreover, the majority of prior studies ignore the direct relationship between aspects and the corresponding sentiment words. To this end, in this paper, we propose a novel model named Efficient Adaptive Transfer Network (EATN) for aspect-level sentiment analysis which emphasizes the need of incorporating the correlation among multiple domains. The proposed EATN provides a Domain Adaptation Module (DAM) to learn common features from the sufficiently labeled source domain and to guide the classification performance in the target domain. Specifically, DAM comprises two special tasks, with one sentiment classification task aiming to learn sentiment knowledge and the other domain classification task focusing on learning domain-invariant features. Moreover, we design aspect-aware multi-head attention mechanism to capture the direct associations between the aspects and the contextual sentiment words, which is beneficial to learn the aspect-aware semantic knowledge. Extensive experiments demonstrate the effectiveness and universality of our method.

Entity summarization is a novel and efficient way to understand real-world facts and solve the increasing information overload problem in large-scale knowledge graphs (KG). Existing studies mainly rely on ranking independent entity descriptions as a list under a certain scoring standard such as importance. However, they often ignore the relatedness and even semantic overlap between individual descriptions. This may seriously interfere with the contribution judgment of descriptions for entity summarization. Actually, the entity summary is a whole to comprehensively integrate the main aspects of entity descriptions, which could be naturally treated as a set. Unfortunately, the exploration of these set characteristics for entity summarization is still an open issue with great challenges. To that end, we draw inspiration from a set completion perspective and propose an entity summarization method with complementarity and salience (ESCS) to deeply exploit description complementarity and salience in order to form a summary set for the target entity. Specifically, we first generate entity description representations with textual features in the description embedding module. For the purpose of learning complementary relationships within the entire summary set, we devise a bi-directional long short-term memory structure to capture global complementarity for each summary in the summary complementarity learning module. Meanwhile, in order to estimate the salience of individual descriptions, we calculate similarities between semantic embeddings of the target entity and its property-value pairs in the description salience learning module. Next, with a joint learning stage, we can optimize ESCS from a set completion perspective. Finally, a summary generation strategy is designed to infer the entire summary set step-by-step for the target entity. Extensive experiments on a public benchmark have clearly demonstrated the effectiveness of ESCS and revealed the potential of set completion in entity summarization task.

Sentence semantic matching requires an agent to determine the semantic relation between two sentences, which is widely used in various natural language tasks, such as natural language inference (NLI) and paraphrase identification (PI). Much recent progress has been made in this area, especially attention-based methods and pretrained language model-based methods. However, most of these methods focus on all the important parts in sentences in a static way and only emphasize how important the words are to the query, inhibiting the ability of the attention mechanism. In order to overcome this problem and boost the performance of the attention mechanism, we propose a novel dynamic reread (DRr) attention, which can pay close attention to one small region of sentences at each step and reread the important parts for better sentence representations. Based on this attention variation, we develop a novel DRr network (DRr-Net) for sentence semantic matching. Moreover, selecting one small region in DRr attention seems insufficient for sentence semantics, and employing pretrained language models as input encoders will introduce incomplete and fragile representation problems. To this end, we extend DRr-Net to locally aware dynamic reread attention net (LadRa-Net), in which local structure of sentences is employed to alleviate the shortcoming of byte-pair encoding (BPE) in pretrained language models and boost the performance of DRr attention. Extensive experiments on two popular sentence semantic matching tasks demonstrate that DRr-Net can significantly improve the performance of sentence semantic matching. Meanwhile, LadRa-Net is able to achieve better performance by considering the local structures of sentences. In addition, it is exceedingly interesting that some discoveries in our experiments are consistent with some findings of psychological research.

We focus on Maximum Inner Product Search (MIPS), which is an essential problem in many machine learning communities. Given a query, MIPS finds the most similar items with the maximum inner products. Methods for Nearest Neighbor Search (NNS) which is usually defined on metric space do not exhibit the satisfactory performance for MIPS problem since inner product is a non-metric function. However, inner products exhibit many good properties compared with metric functions, such as avoiding vanishing and exploding gradients. As a result, inner product is widely used in many recommendation systems, which makes efficient Maximum Inner Product Search a key for speeding up many recommendation systems. Graph-based methods for NNS problem show the superiorities compared with other class methods. Each data point of the database is mapped to a node of the proximity graph. Nearest neighbor search in the database can be converted to route on the proximity graph to find the nearest neighbor for the query. This technique can be used to solve MIPS problem. Instead of searching the nearest neighbor for the query, we search the item with a maximum inner product with query on the proximity graph. In this article, we propose a reinforcement model to train an agent to search on the proximity graph automatically for MIPS problem if we lack the ground truths of training queries. If we know the ground truths of some training queries, our model can also utilize these ground truths by imitation learning to improve the agent’s searchability. By experiments, we can see that our proposed mode which combines reinforcement learning with imitation learning shows the superiorities over the state-of-the-art methods.

Similar text search aims to find texts relevant to a given query from a database, which is fundamental in many information retrieval applications, such as question search and exercise search. Since millions of texts always exist behind practical search engine systems, a well-developed text search system usually consists of recall and ranking stages. Specifically, the recall stage serves as the basis in the system, where the main purpose is to find a small set of relevant candidates accurately and efficiently. Towards this goal, deep semantic hashing, which projects original texts into compact hash codes, can support good search performance. However, learning desired textual hash codes is extremely difficult due to the following problems. First, compact hash codes (with short length) can improve retrieval efficiency, but the demand for learning compact hash codes cannot guarantee accuracy due to severe information loss. Second, existing methods always learn the unevenly distributed codes in the space from a local perspective, leading to unsatisfactory code-balance results. Third, a large fraction of textual data contains various types of noise in real-world applications, which causes the deviation of semantics in hash codes. To this end, in this paper, we first propose a general unsupervised encoder-decoder semantic hashing framework, namely MASH (short for Memory-bAsed Semantic Hashing), to learn the balanced and compact hash codes for similar text search. Specifically, with a target of retaining semantic information as much as possible, the encoder introduces a novel relevance constraint among informative high-dimensional representations to guide the compact hash code learning. Then, we design an external memory where the hashing learning can be optimized in the global space to ensure the code balance of the learning results, which can promote search efficiency. Besides, to alleviate the performance degradation problem of the model caused by text noise, we propose an improved SMASH (short for denoiSing Memory-bAsed Semantic Hashing) model by incorporating a noise-aware encoder-decoder framework. This framework considers the noise degree for each text from the semantic deviation aspect, ensuring the robustness of hash codes. Finally, we conduct extensive experiments in three real-world datasets. The experimental results clearly demonstrate the effectiveness and efficiency of MASH and SMASH in generating balanced and compact hash codes, as well as the superior denoising ability of SMASH.

News recommender systems have become an effective manner to help users make decisions by suggesting the potential news that users may click and read, which has shown the proliferation nowadays. Many representative algorithms made great efforts to discover users’ preferences from the histories for triggering news recommendations. However, there exist some limitations due to the following two main issues. First, they mainly rely on the sufficient user data, which cannot well capture users’ temporal interests with very limited records. Second, always perceiving users’ histories for recommendation may ignore some important news (e.g., breaking news). In this article, we propose a novel Multi-factors Fusion model for news recommendation by integrating both user-dependent preference effect and user-independent timeliness effect together. First, to track the preference of a certain user, we decompose her reading history into two user-related factors, including the long-term habit and the short-term interest. Specifically, we extract her persistent habit by exploring the category effect of news that she focuses on from her whole records. Then, we characterize her temporary interests by proposing a recurrent neural network of analyzing the homogeneous relations between her latest clicked news and the candidate ones. Second, to describe the user-independent news timeliness effect, we propose a novel survival analysis model to estimate the instantaneous click probability of a certain news as the occurring probability of an event, where much sensational news tends to be picked out. Last, we fuse all effects to determine the probability of a user clicking on a certain news under the independent event assumption. We conduct extensive experiments on two real-world datasets. Experimental results demonstrate that our model can generate better news recommendations on both general scenario and cold-start scenario.

The generation of large amounts of personal data provides data centers with sufficient resources to mine idiosyncrasy from private records. User modeling has long been a fundamental task with the goal of capturing the latent characteristics of users from their behaviors. However, centralized user modeling on collected data has raised concerns about the risk of data misuse and privacy leakage. As a result, federated user modeling has come into favor, since it expects to provide secure multi-client collaboration for user modeling through federated learning. Unfortunately, to the best of our knowledge, existing federated learning methods that ignore the inconsistency among clients cannot be applied directly to practical user modeling scenarios, and moreover, they meet the following critical challenges: 1) Statistical heterogeneity . The distributions of user data in different clients are not always independently identically distributed (IID), which leads to unique clients with needful personalized information; 2) Privacy heterogeneity . User data contains both public and private information, which have different levels of privacy, indicating that we should balance different information shared and protected; 3) Model heterogeneity . The local user models trained with client records are heterogeneous, and thus require a flexible aggregation in the server; 4) Quality heterogeneity . Low-quality information from inconsistent clients poisons the reliability of user models and offsets the benefit from high-quality ones, meaning that we should augment the high-quality information during the process. To address the challenges, in this paper, we first propose a novel client-server architecture framework, namely Hierarchical Personalized Federated Learning (HPFL), with a primary goal of serving federated learning for user modeling in inconsistent clients. More specifically, the client train and deliver the local user model via the hierarchical components containing hierarchical information from privacy heterogeneity to join collaboration in federated learning. Moreover, the client updates the personalized user model with a fine-grained personalized update strategy for statistical heterogeneity. Correspondingly, the server flexibly aggregates hierarchical components from heterogeneous user models in the case of privacy and model heterogeneity with a differentiated component aggregation strategy. In order to augment high-quality information and generate high-quality user models, we expand HPFL to the Augmented-HPFL (AHPFL) framework by incorporating the augmented mechanisms, which filters out low-quality information such as noise, sparse information and redundant information. Specially, we construct two implementations of AHPFL, i.e., AHPFL-SVD and AHPFL-AE, where the augmented mechanisms follow SVD (singular value decomposition) and AE (autoencoder), respectively. Finally, we conduct extensive experiments on real-world datasets, which demonstrate the effectiveness of both HPFL and AHPFL frameworks.

Recent years have witnessed the rapid accumulation of massive electronic medical records, which highly support intelligent medical services such as drug recommendation. However, although there are multiple interaction types between drugs, e.g., synergism and antagonism, which can influence the effect of a drug package significantly, prior arts generally neglect the interaction between drugs or consider only a single type of interaction. Moreover, most existing studies generally formulate the problem of package recommendation as getting a personalized scoring function for users, despite the limits of discriminative models to achieve satisfactory performance in practical applications. To this end, in this article, we propose a novel end-to-end Drug Package Generation (DPG) framework, which develops a new generative model for drug package recommendation that considers the interaction effects between drugs that are affected by patient conditions. Specifically, we propose to formulate the drug package generation as a sequence generation process. Along this line, we first initialize the drug interaction graph based on medical records and domain knowledge. Then, we design a novel message-passing neural network to capture the drug interaction, as well as a drug package generator based on a recurrent neural network. In detail, a mask layer is utilized to capture the impact of patient condition, and the deep reinforcement learning technique is leveraged to reduce the dependence on the drug order. Finally, extensive experiments on a real-world dataset from a first-rate hospital demonstrate the effectiveness of our DPG framework compared with several competitive baseline methods.

As a vivid and linguistic symbol, Emojis have become a prevailing medium interspersed in text-based communication (e.g., social media and chit-chat) to express emotions, attitudes, and situations. Generally speaking, a social-oriented chatbot that can generate appropriate Emoji-embedded responses would be much more competitive, making communications more fun, engaging, and human-like. However, the current Emoji-related research is still in its infancy, leading to an awkward situation of data deficiency. How to develop an Emoji-embedded dialogue system while addressing the lack of data will be interesting and meaningful for the application of future AI. To bridge this gap, we propose a multi-task learning method for persona-aware Emoji-embedded dialogue generation in this article. Specifically, as the benchmark of model training and evaluation, which includes 1.2 million Emoji-embedded tweets and 1.1 million post-response pairs, we first construct a dataset named EmojiTweet to handle the data deficiency problem. Then, a Seq2Seq-based model with multi-task learning is designed to simultaneously learn response generation and Emoji embedding from the constructed non-Emoji dialogue and Emoji-embedded monologue data. Afterward, we incorporate persona factors into our model by adopting persona fusion and personalized bias methods to deliver personalized dialogues with more accurately selected Emojis. Finally, we conduct extensive experiments, where the experimental results and evaluations demonstrate that our model has three key benefits: improved dialogue quality, higher user engagement, and not relying on large-scale Emoji-embedded dialogue data representing specific personas. EmojiTweet will be published publicly via https://mea-lab-421.github.io/EmojiTweet/ .

Maximum Inner Product Search (MIPS) plays an important role in many applications ranging from information retrieval, recommender systems to natural language processing and machine learning. However, exhaustive MIPS is often expensive and impractical when there are a large number of candidate items. The state-of-the-art approximated MIPS is product quantization with a score-aware loss, which weighs more heavily on items with larger inner product scores. However, it is challenging to extend the score-aware loss for additive quantization due to parallel-orthogonal decomposition of residual error. Learning additive quantization with respect to this loss is important since additive quantization can achieve a lower approximation error than product quantization. To this end, we propose a quantization method called Anisotropic Additive Quantization to combine the score-aware anisotropic loss and additive quantization. To efficiently update the codebooks in this algorithm, we develop a new alternating optimization algorithm. The proposed algorithm is extensively evaluated on three real-world datasets. The experimental results show that it outperforms the state-of-the-art baselines with respect to approximate search accuracy while guaranteeing a similar retrieval efficiency.

Abstractive summarization is a classic task in Natural Language Generation (NLG), which aims to produce a concise summary of the original document. Recently, great efforts have been made on sequence-to-sequence neural networks to generate abstractive sum- maries with a high level of fluency. However, prior arts mainly focus on the optimization of token-level likelihood, while the rich semantic information in documents has been largely ignored. In this way, the summarization results could be vulnerable to hallucinations, i.e., the semantic-level inconsistency between a summary and corresponding original document. To deal with this challenge, in this paper, we propose a novel fact-aware abstractive summarization model, named Entity-Relation Pointer Generator Network (ERPGN). Specially, we attempt to formalize the facts in original document as a factual knowledge graph, and then generate the high-quality summary via directly modeling consistency between summary and the factual knowledge graph. To that end, we first leverage two pointer net- work structures to capture the fact in original documents. Then, to enhance the traditional token-level likelihood loss, we design two extra semantic-level losses to measure the disagreement between a summary and facts from its original document. Extensive experi- ments on public datasets demonstrate that our ERPGN framework could outperform both classic abstractive summarization models and the state-of-the-art fact-aware baseline methods, with significant improvement in terms of faithfulness.

Civil Case Judgment Prediction (CCJP) is a fundamental task in the legal intelligence of the civil law system, which aims to automatically predict the judgment results on each plea of the plaintiff. Existing studies mainly focus on making judgment predictions only on a certain civil cause (e.g., the divorce dispute) by utilizing the fact descriptions and pleas of the plaintiff, which still suffer from the various causes and complicated legal essential elements in the real court. Thus, in this paper, we formalize CCJP as a multi-task learning problem and propose a CCJP method centering on the trial mode of essential elements, CPEE, which explores the practical judicial process and analyzes comprehensive legal essential elements to make judgment predictions. Specifically, we first construct three tasks (i.e., the predictions on the civil causes, law articles, and the final judgment on each plea) necessary for CCJP, that follow the judgment process and exploit the results of intermediate subtasks to make judgment predictions. Then we design a logic-enhanced network to predict the results of three tasks and conduct a comprehensive study of civil cases. Finally, owing to the interlinked and dependent relationships among each task, we adopt the cause prediction result to help predict law articles and incorporate them into final judgment prediction through a gate mechanism. Furthermore, since the existing dataset fails to provide sufficient case information, we construct a real-world CCJP dataset that contains various causes and comprehensive legal elements. Extensive experimental results on the dataset validate the effectiveness of our method.

Adversarial learning has achieved remarkable performances for unsupervised domain adaptation (UDA). Existing adversarial UDA methods typically adopt an additional discriminator to play the min-max game with a feature extractor. However, most of these methods failed to effectively leverage the predicted discriminative information, and thus cause mode collapse for generator. In this work, we address this problem from a different perspective and design a simple yet effective adversarial paradigm in the form of a discriminator-free adversarial learning network (DALN), wherein the category classifier is reused as a discriminator, which achieves explicit domain alignment and category distinguishment through a unified objective, enabling the DALN to leverage the predicted discriminative information for sufficient feature alignment. Basically, we introduce a Nuclear-norm Wasserstein discrepancy (NWD) that has definite guidance meaning for performing discrimination. Such NWD can be coupled with the classifier to serve as a discriminator satisfying the K-Lipschitz constraint without the requirements of additional weight clipping or gradient penalty strategy. Without bells and whistles, DALN compares favorably against the existing state-of-the-art (SOTA) methods on a variety of public datasets. Moreover, as a plug-and-play technique, NWD can be directly used as a generic regularizer to benefit existing UDA algorithms. Code is available at https://github.com/xiaoachen98/DALN.

The end-to-end speech translation (E2E-ST) has received increasing attention due to the potential of its less error propagation, lower latency and fewer parameters.However, the effectiveness of neural-based approaches to this task is severely limited by the available training corpus, especially for domain adaptation where in-domain triplet data is scarce or nonexistent.In this paper, we propose a novel nonparametric method that leverages in-domain text translation corpus to achieve domain adaptation for E2E-ST systems.To this end, we first incorporate an additional encoder into the pre-trained E2E-ST model to realize text translation modeling, based on which the decoder's output representations for text and speech translation tasks are unified by reducing the correspondent representation mismatch in available triplet training data.During domain adaptation, a k-nearest-neighbor (kNN) classifier is introduced to produce the final translation distribution using the external datastore built by the domain-specific text translation corpus, while the universal output representation is adopted to perform a similarity search.Experiments on the Europarl-ST benchmark demonstrate that when in-domain text translation data is involved only, our proposed approach significantly improves baseline by 12.82 BLEU on average in all translation directions, even outperforming the strong in-domain fine-tuning strategy.

Text matching is a fundamental research problem in natural language understanding.Interaction-based approaches treat the text pair as a single sequence and encode it through cross encoders, while representation-based models encode the text pair independently with siamese or dual encoders.Interactionbased models require dense computations and thus are impractical in real-world applications.Representation-based models have become the mainstream paradigm for efficient text matching.However, these models suffer from severe performance degradation due to the lack of interactions between the pair of texts.To remedy this, we propose a Virtual InteRacTion mechanism (VIRT) for improving representation-based text matching while maintaining its efficiency.In particular, we introduce an interactive knowledge distillation module that is only applied during training.It enables deep interaction between texts by effectively transferring knowledge from the interaction-based model.A light interaction strategy is designed to fully leverage the learned interactive knowledge.Experimental results on six text matching benchmarks demonstrate the superior performance of our method over several state-of-the-art representationbased models.We further show that VIRT can be integrated into existing methods as plugins to lift their performances.

Intercorporate investment has a large impact in financial performance and long-term development of a corporate. Among all the concerns for a company ’s investment strategy, complementary and substitutable investments are two fundamental factors. However, these two relations are implicit and entangled in the complex corporate network, requiring extra caution before investment. To this end, in this paper, we proposed a novel graph convolutional network called Series-Parallel decomposed Graph Convolutional Network (SPGCN). We first decompose the complementary and substitutable relations as two information propagating directions in company dependency graph, producing multifaceted node features. Then, with an Attentive Aggregation Module, we are able to further measure the impact of both features to the final investment decision making, producing an interpretable analysis for investment strategy. Finally, we conduct experiments on a real-world dataset, to show the effectiveness of decomposing two concerns on investment recommendation task. With visualization and case studies, our method also shows great potential to help understand and conduct complementary and substitutable investment decisions. We open source our code to support future research: https://github.com/lem0n1e/SPGCN.

Deep learning has brought significant breakthroughs in sequential recommendation (SR) for capturing dynamic user interests. A series of recent research revealed that models with more parameters usually achieve optimal performance for SR tasks, inevitably resulting in great challenges for deploying them in real systems. Following the simple assumption that light networks might already suffice for certain users, in this work, we propose CANet, a conceptually simple yet very scalable framework for assigning adaptive network architecture in an input-dependent manner to reduce unnecessary computation. The core idea of CANet is to route the input user behaviors with a light-weighted router module. Specifically, we first construct the routing space with various submodels parameterized in terms of multiple model dimensions such as the number of layers, hidden size and embedding size. To avoid extra storage overhead of the routing space, we employ a weight-slicing schema to maintain all the submodels in exactly one network. Furthermore, we leverage several solutions to solve the discrete optimization issues caused by the router module. Thanks to them, CANet could adaptively adjust its network architecture for each input in an end-to-end manner, in which the user preference can be effectively captured. To evaluate our work, we conduct extensive experiments on benchmark datasets. Experimental results show that CANet reduces computation by 55 ~ 65% while preserving the accuracy of the original model. Our codes are available at https://github.conicantnamemyself/CANet.

Reasoning answers to mathematical problems requires machines to think and operate like a human to learn knowledge from mathematical data, which is one of the fundamental tasks for exploring general artificial intelligence. Most solutions focus on mimicking how humans understand problems, which generate the necessary expressions for answers. However, they are still far from enough since they ignore the core ability of humans to acquire knowledge from experience. In this paper, we propose a Cognitive Solver (CogSolver) that is capable of autonomously learning knowledge from scratch to solve mathematical problems, inspired by two cognitive science theories. Specifically, we draw one insight from the dual process theory to establish an intelligent BRAIN-ARM framework, and refer to another information processing theory to summarize the knowledge learning process into Store-Apply-Update steps. In CogSolver, the BRAIN system stores three types of mathematical knowledge, including semantics knowledge, relation knowledge, and mathematic rule knowledge. Then, the ARM system applies the knowledge in BRAIN to answer the problems. Specifically, we design a knowledge-aware module and a commutative module in ARM to improve its reasoning ability, where the knowledge is organically integrated into answer reasoning process. After solving the problems, BRAIN updates the stored knowledge according to the feedback of ARM, where we develop knowledge filters to eliminate the redundant ones and further form a more reasonable knowledge base. Our CogSolver carries out the above three steps iteratively, which behaves more like a human. We conduct extensive experiments on real-world math word problem datasets. The experimental results demonstrate the improvement in answer reasoning and clearly show how CogSolver gains knowledge from the problems, leading to superior interpretability. Our codes are available at https://github.com/bigdata-ustc/CogSolver.

Facial expression recognition (FER) has made significant progress over the past few years. But how to overcome the problem of high inter-class similarity and large intra-class difference in FER is still challenging. To address this problem, we propose a novel FER framework called AU-assisted Visual Transformer (AVT) by incorporating facial action units (AU) information into Visual Transformer, which mainly consists of three modules: Local Feature Extraction (LFE) module, Global Relationship Modeling (GRM) module and AU Fusion Module (AFM). Specifically, the LFE module aims to extract local facial expression features by using a deep convolutional neural network, the GRM module is a multi-layer Transformer encoder that captures the relation between local facial regions and obtains a global understanding of the face, and in particular, the AFM introduces fine-grained AU feature and fuses it with expression feature for final classification. Extensive experiments are conducted on RAF-DB and FERPlus datasets, and our AVT achieves competitive results compared to previous state-of-the-art methods, demonstrating the effectiveness of our approach.

Knowledge Tracing (KT) focuses on tracing stu-dents' evolving knowledge state in the process of question-answering, which has been widely applied to online learning systems. Existing KT methods leverage students' online question-answering logs to infer their knowledge state and predict their subsequent performance. However, these methods only focus on students' online behaviors without considering the fact that students' knowledge also evolves offline. Specifically, students have plenty of offline learning actions that also make changes in their knowledge state, which is called the offline knowledge evolution effect in this paper. However, these offline actions are not recorded by online learning systems, therefore leading to unreliable knowledge tracing results. To alleviate this problem, we propose a novel model named offline-aware knowledge tracing (OKT) that considers students' offline knowledge evolution for KT. In O KT, we introduce a novel concept named offline trait to represent student's offline learning action preference. Based on the preference, the model evaluates offline knowledge evolution, and apply the evolution to generate a new knowledge state for students' future performance prediction. To the end, we conduct experiments on three real-world benchmark datasets and demonstrate that OKT achieves state-of-the-art performance. Besides, the experimental results indicate that considering offline actions can help interpret the KT results.

Stock movement prediction aims at predicting the future price trends of stocks, which plays an important role in quantitative investing. Existing approaches toward this direction mainly focus on modeling the historical sequential information, while the fine-grained relationships among stocks (e.g., belonging to the same industry or concept) were largely neglected. To tackle this limitation, in this paper, we propose a Multi-Relational Graph Convolution Network (MRGCN) framework for stock movement prediction, which incorporates the fine-grained multiple relationships into stock representation. Specifically, we first extract the temporal and static information for each stock from the historical series and corporation descriptions respectively. Then, we construct two pre-defined graphs based on domain knowledge and a self-adaptive graph to capture both explicit and implicit relationships among stocks. Along this line, the graph convolution network with attention mechanism is adopted on the multi-relational graph to generate the structural representation for each stock, and an embedding reconstruction module is further designed to refine the representation. Finally, we make predictions by integrating both temporal and structural embedding of stocks. Experiments on real-world China A-share market evince the superior performance of MRGCN compared to other baselines.

Text classification is a fundamental and classical problem in natural language processing. Existing methods in this area attach more attention to structure modeling of texts, while largely ignoring the cognitive principles of human reading. Actually, as an important aspect of exploring the characteristics of language comprehension, neuroscience research in recent years has demonstrated the human instinct for abstract thinking, where semantic processing and summarizing play essential roles. To this end, we propose a novel text classification method with Semantic Dependency and Association (SDA). To be specific, SDA comprises two coupled spaces, namely Dependency Space and Association Space, to imitate the real process when readers are inferring text semantics. Firstly, in Dependency Space, we devise a graph modeling structure to thoroughly capture the dependency relations between words. Then in Association Space, we associate the words with prior concepts and adopt attention mechanisms to match words with the most related concepts based on their dependency relations. Later, we synthesize the above two aspects to enhance the semantic representation of words. Afterwards, we design a Semantic Summarizing module to integrate the semantic information learned above. Finally, extensive experiments are conducted on four public English and Chinese text classification datasets, where experimental results not only demonstrate the effectiveness of our model but also reveal the superiority of the perspective from the human cognition for text comprehension.

The booming of multi-modal knowledge graphs (MMKGs) has raised the imperative demand for multi-modal entity alignment techniques, which facilitate the integration of multiple MMKGs from separate data sources. Unfortunately, prior arts harness multi-modal knowledge only via the heuristic merging of uni-modal feature embeddings. Therefore, inter-modal cues concealed in multi-modal knowledge could be largely ignored. To deal with that problem, in this paper, we propose a novel Multi-modal Siamese Network for Entity Alignment (MSNEA) to align entities in different MMKGs, in which multi-modal knowledge could be comprehensively leveraged by the exploitation of inter-modal effect. Specifically, we first devise a multi-modal knowledge embedding module to extract visual, relational, and attribute features of entities to generate holistic entity representations for distinct MMKGs. During this procedure, we employ inter-modal enhancement mechanisms to integrate visual features to guide relational feature learning and adaptively assign attention weights to capture valuable attributes for alignment. Afterwards, we design a multi-modal contrastive learning module to achieve inter-modal enhancement fusion with avoiding the overwhelming impact of weak modalities. Experimental results on two public datasets demonstrate that our proposed MSNEA provides state-of-the-art performance with a large margin compared with competitive baselines.

Cognitive diagnostic assessment is a fundamental task in intelligent education, which aims at quantifying students' cognitive level on knowledge attributes. Since there exists learning dependency among knowledge attributes, it is crucial for cognitive diagnosis models (CDMs) to incorporate attribute hierarchy when assessing students. The attribute hierarchy is only explored by a few CDMs such as Attribute Hierarchy Method, and there are still two significant limitations in these methods. First, the time complexity would be unbearable when the number of attributes is large. Second, the assumption used to model the attribute hierarchy is too strong so that it may lose some information of the hierarchy and is not flexible enough to fit all situations. To address these limitations, we propose a novel Bayesian network-based Hierarchical Cognitive Diagnosis Framework (HierCDF), which enables many traditional diagnostic models to flexibly integrate the attribute hierarchy for better diagnosis. Specifically, we first use an efficient Bayesian network to model the influence of attribute hierarchy on students' cognitive states. Then we design a CDM adaptor to bridge the gap between students' cognitive states and the input features of existing diagnostic models. Finally, we analyze the generality and complexity of HierCDF to show its effectiveness in modeling hierarchy information. The performance of HierCDF is experimentally proved on real-world large-scale datasets.

Cognitive diagnosis, aiming at providing an approach to reveal the proficiency level of learners on knowledge concepts, plays an important role in intelligent education area and has recently received more and more attention. Although a number of works have been proposed in recent years, most of contemporary works acquire the traits parameters of learners and items in a transductive way, which are only suitable for stationary data. However, in the real scenario, the data is collected online, where learners, test items and interactions usually grow continuously, which can rarely meet the stationary condition. To this end, we propose a novel framework, Incremental Cognitive Diagnosis (ICD), to tailor cognitive diagnosis into the online scenario of intelligent education. Specifically, we first design a Deep Trait Network (DTN), which acquires the trait parameters in an inductive way rather than a transductive way. Then, we propose an Incremental Update Algorithm (IUA) to balance the effectiveness and training efficiency. We carry out Turning Point (TP) analysis to reduce update frequency, where we derive the minimum update condition based on the monotonicity theory of cognitive diagnosis. Meanwhile, we use a momentum update strategy on the incremental data to decrease update time without sacrificing effectiveness. Moreover, to keep the trait parameters as stable as possible, we refine the loss function in the incremental updating stage. Last but no least, our ICD is a general framework which can be applied to most of contemporary cognitive diagnosis models. To the best of our knowledge, this is the first attempt to investigate the incremental cognitive diagnosis problem with theoretical results about the update condition and a tailored incremental learning strategy. Extensive experiments demonstrate the effectiveness and robustness of our method.

Online medical consultation, which enables patients to remotely inquire doctors in the form of web chatting, has become an indispensable part of the social health care system. Intuitively, it is a crucial step to recommend suitable doctor candidates for patients, especially with suffering the severe cold-start challenge of patients due to the limited historical records and insufficient description of patient condition. Along this line, in this paper, we propose a novel Dialogue based Doctor Recommendation (DDR) model, which comprehensively integrates three types of information in modeling, including the profile and chief complaint from patients, the historical records of doctors and the patient-doctor dialogue. Accordingly, we propose 1) a patient encoder which represents the patient's condition and medical requirements; 2) a doctor encoder which distills the doctor's expertise and communication skills; 3) a dialogue encoder which extracts textual features from doctor-patient conversation. Specifically, since the patient-doctor dialogue is not available in the testing stage, we propose to simulate the dialogue embedding with patient embedding via a contrastive learning based module. Experimental results on a real-world data set show that the proposed DDR model can outperform state-of-the-art recommendation-based methods. Moreover, considering the accessibility variance of online medical consultation services between the youth and the elderly, we also conduct a fairness study on the proposed DDR model.

Recently, Micro expression~(ME) has achieved remarkable progress in a wide range of applications, since it's an involuntary facial expression that reflects personal psychological state truly. In the procedure of ME analysis, spotting ME is an essential step, and is non trivial to be detected from a long interval video because of the short duration and low intensity issues. To alleviate this problem, in this paper, we propose a novel Micro- and Macro-Expression~(MaE) Spotting framework based on Apex and Boundary Perception Network~(ABPN), which mainly consists of three parts, i.e., video encoding module ~(VEM), probability evaluation module~(PEM), and expression proposal generation module~(EPGM). Firstly, we adopt Main Directional Mean Optical Flow (MDMO) algorithm and calculate optical flow differences to extract facial motion features in VEM, which can alleviate the impact of head movement and other areas of the face on ME spotting. Then, we extract temporal features with one-dimension convolutional layers and introduce PEM to infer the auxiliary probability that each frame belongs to an apex or boundary frame. With these frame-level auxiliary probabilities, the EPGM further combines the frames from different categories to generate expression proposals for the accurate localization. Besides, we conduct comprehensive experiments on MEGC2022 spotting task, and demonstrate that our proposed method achieves significant improvement with the comparison of state-of-the-art baselines on rm CAS(ME)2 and SAMM-LV datasets. The implemented code is also publicly available at https://github.com/wenhaocold/USTC_ME_Spotting.

Understanding the long duration videos' storyline is often considered a major challenge in the field of video understanding. To promote research on understanding longer videos in the community, the deep video understanding (DVU) task is suggested for recognizing interactions at the scene level and relationships at the movie level, as well as answering questions at these two levels. In this work, we propose a unified QA-aware knowledge graph generation approach, which consists of the relation-centric graph and interaction-centric graph and demonstrates the powerful performance of multimodal pre-training models in solving such problems. Extensive validations on the HLVU dataset demonstrate the effectiveness of our proposed method.

Knowledge Graph Completion (KGC), aiming to infer the missing part of Knowledge Graphs (KGs), has long been treated as a crucial task to support downstream applications of KGs, especially for the multimodal KGs (MKGs) which suffer the incomplete relations due to the insufficient accumulation of multimodal corpus. Though a few research attentions have been paid to the completion task of MKGs, there is still a lack of specially designed negative sampling strategies tailored to MKGs. Meanwhile, though effective negative sampling strategies have been widely regarded as a crucial solution for KGC to alleviate the vanishing gradient problem, we realize that, there is a unique challenge for negative sampling in MKGs about how to model the effect of KG relations during learning the complementary semantics among multiple modalities as an extra context. In this case, traditional negative sampling techniques which only consider the structural knowledge may fail to deal with the multimodal KGC task. To that end, in this paper, we propose a MultiModal Relation-enhanced Negative Sampling (MMRNS) framework for multimodal KGC task. Especially, we design a novel knowledge-guided cross-modal attention (KCA) mechanism, which provides bi-directional attention for visual & textual features via integrating relation embedding. Then, an effective contrastive semantic sampler is devised after consolidating the KCA mechanism with contrastive learning. In this way, a more similar representation of semantic features between positive samples, as well as a more diverse representation between negative samples under different relations could be learned. Afterwards, a masked gumbel-softmax optimization mechanism is utilized for solving the non-differentiability of sampling process, which provides effective parameter optimization compared with traditional sample strategies. Extensive experiments on three multimodal KGs demonstrate that our MMRNS framework could significantly outperform the state-of-the-art baseline methods, which validates the effectiveness of relation guides in multimodal KGC task.

As a typical psychological stress reaction, micro-expression (ME) is usually quickly leaked on a human face and can reveal the true feeling and emotional cognition. Therefore,automatic ME analysis (MEA) has essential applications in safety, clinical and other fields. However, the lack of adequate ME data has severely hindered MEA research. To overcome this dilemma and encouraged by current image generation techniques, this paper proposes a fine-grained ME generation method to enhance ME data in terms of data volume and diversity. Specifically, we first estimate non-linear ME motion using thin-plate spline transformation with a dense motion network. Then, the estimated ME motion transformations, including optical flow and occlusion masks, are sent to the generation network to synthesize the target facial micro-expression. In particular, we obtain the relative action units (AUs) of the source ME to the target face as a constraint to encourage the network to ignore expression-irrelevant movements, thereby generating fine-grained MEs. Through comparative experiments on CASME II, SMIC and SAMM datasets, we demonstrate the effectiveness and superiority of our method. Source code is provided in https://github.com/MEA-LAB-421/MEGC2022-Generation.

Recommender retrievers aim to rapidly retrieve a fraction of items from the entire item corpus when a user query requests, with the representative two-tower model trained with the log softmax loss. For efficiently training recommender retrievers on modern hardwares, inbatch sampling, where the items in the mini-batch are shared as negatives to estimate the softmax function, has attained growing interest. However, existing inbatch sampling based strategies just correct the sampling bias of inbatch items with item frequency, being unable to distinguish the user queries within the mini-batch and still incurring significant bias from the softmax. In this paper, we propose a Cache-Augmented Inbatch Importance Resampling (XIR) for training recommender retrievers, which not only offers different negatives to user queries with inbatch items, but also adaptively achieves a more accurate estimation of the softmax distribution. Specifically, XIR resamples items from the given mini-batch training pairs based on certain probabilities, where a cache with more frequently sampled items is adopted to augment the candidate item set, with the purpose of reusing the historical informative samples. XIR enables to sample query-dependent negatives based on inbatch items and to capture dynamic changes of model training, which leads to a better approximation of the softmax and further contributes to better convergence. Finally, we conduct experiments to validate the superior performance of the proposed XIR compared with competitive approaches.

Inspired by their powerful representation ability on graph-structured data, Graph Convolution Networks (GCNs) have been widely applied to recommender systems, and have shown superior performance. Despite their empirical success, there is a lack of theoretical explorations such as generalization properties. In this paper, we take a first step towards establishing a generalization guarantee for GCN-based recommendation models under inductive and transductive learning. We mainly investigate the roles of graph normalization and non-linear activation, providing some theoretical understanding, and construct extensive experiments to further verify these findings empirically. Furthermore, based on the proven generalization bound and the challenge of existing models in discrete data learning, we propose Item Mixture (IMix) to enhance recommendation. It models discrete spaces in a continuous manner by mixing the embeddings of positive-negative item pairs, and its effectiveness can be strictly guaranteed from empirical and theoretical aspects.

Recommender systems (RS) have to select the top-N items from a massive item set. For the sake of efficient recommendation, RS usually represents user and item as latent embeddings, and relies on approximate nearest neighbour search (ANNs) to retrieve the recommendation result. Despite the reduction of running time, the representation learning is independent of ANNs index construction; thus, the two operations can be incompatible, which results in potential loss of recommendation accuracy. To overcome the above problem, we propose the Recommender Forest (a.k.a., RecForest), which jointly learns latent embedding and index for efficient and high-fidelity recommendation. RecForest consists of multiple k-ary trees, each of which is a partition of the item set via hierarchical balanced clustering such that each item is uniquely represented by a path from the root to a leaf. Given such a data structure, an encoder-decoder based routing network is developed: it first encodes the context, i.e., user information, into hidden states; then, leveraging a transformer-based decoder, it identifies the top-N items via beam search. Compared with the existing methods, RecForest brings in the following advantages: 1) the false partition of the boundary items can be effectively alleviated by the use of multiple trees; 2) the routing operation becomes much more accurate thanks to the powerful transformer decoder; 3) the tree parameters are shared across different tree levels, making the index to be extremely memory-efficient. The experimental studies are performed on five popular recommendation datasets: with a significantly simplified training cost, RecForest outperforms competitive baseline approaches in terms of both recommendation accuracy and efficiency.

Rationale extraction can be considered as a straightforward method of improving the model explainability, where rationales are a subsequence of the original inputs, and can be extracted to support the prediction results. Existing methods are mainly cascaded with the selector which extracts the rationale tokens, and the predictor which makes the prediction based on selected tokens. Since previous works fail to fully exploit the original input, where the information of non-selected tokens is ignored, in this paper, we propose a Disentanglement-Augmented Rationale Extraction (DARE) method, which encapsulates more information from the input to extract rationales. Specifically, it first disentangles the input into the rationale representations and the non-rationale ones, and then learns more comprehensive rationale representations for extracting by minimizing the mutual information (MI) between the two disentangled representations. Besides, to improve the performance of MI minimization, we develop a new MI estimator by exploring existing MI estimation methods. Extensive experimental results on three real-world datasets and simulation studies clearly validate the effectiveness of our proposed method. Code is released at https://github.com/yuelinan/DARE.

Due to the widespread presence of implicit feedback, recommendation based on them has been a long-standing research problem in academia and industry. However, it suffers from the extremely-sparse problem, since each user only interacts with a few items. One well-known and good-performing method is to treat each user's all uninteracted items as negative with low confidence. The method intrinsically imposes an implicit regularization to penalize large deviation of each user's preferences for uninteracted items from a constant. However, these methods have to assume a constant-rating prior to uninteracted items, which may be questionable. In this paper, we propose a novel ring-based regularization to penalize significant differences of each user's preferences between each item and some other items. The ring structure, described by an item graph, determines which other items are selected for each item in the regularization. The regularization not only averts the introduction of the prior ratings but also implicitly penalizes the remarkable preference differences for all items according to theoretical analysis. However, optimizing the recommenders with the regularization still suffers from computational challenges, so we develop a scalable alternating least square algorithm by carefully designing gradient computation. Therefore, as long as connecting each item with a sublinear/constant number of other items in the item graph, the overall learning algorithm could be comparably efficient to the existing algorithms. The proposed regularization is extensively evaluated with several public recommendation datasets, where the results show that the regularization could lead to considerable improvements in recommendation performance.

With the development of deep learning techniques, deep recommendation models also achieve remarkable improvements in terms of recommendation accuracy. However, due to the large number of candidate items in practice and the high cost of preference computation, these methods also suffer from low efficiency of recommendation. The recently proposed tree-based deep recommendation models alleviate the problem by directly learning tree structure and representations under the guidance of recommendation objectives. However, such models have two shortcomings. First, the max-heap assumption in the hierarchical tree, in which the preference for a parent node should be the maximum between the preferences for its children, is difficult to satisfy in their binary classification objectives. Second, the learned index only includes a single tree, which is different from the widely-used multiple trees index, providing an opportunity to improve the accuracy of recommendation.

Programming has become an important skill for individuals nowadays. For the demand to improve personal programming skill, tracking programming skill proficiency is getting more and more important. However, few researchers pay attention to measuring the programming skill of learners. Most of existing studies on learner capability portrait only made use of the exercise results, while the rich behavioral information contained in programming exercise process remains unused. Therefore, we propose a model that measures skill proficiency in programming exercise process named Programming Skill Tracing (PST). We designed Code Information Graph (CIG) to represent the feature of learners' solution code, and Code Tracing Graph (CTG) to measure the changes between the adjacent submissions. Furthermore, we divided programming skill into programming knowledge and coding ability to get more fine-grained assessment. Finally, we conducted various experiments to verify the effectiveness and interpretability of our PST model.

Knowledge Tracing (KT), which aims to assess students' dynamic knowledge states when practicing on various questions, is a fundamental research task for offering intelligent services in online learning systems. Researchers have devoted significant efforts to developing KT models with impressive performance. However, in existing KT methods, the related question difficulty level, which directly affects students' knowledge state in learning, has not been effectively explored and employed. In this paper, we focus on exploring the question difficulty effect on learning to improve student's knowledge state assessment and propose the DIfficulty Matching Knowledge Tracing (DIMKT) model. Specifically, we first explicitly incorporate the difficulty level into the question representation. Then, to establish the relation between students' knowledge state and the question difficulty level during the practice process, we accordingly design an adaptive sequential neural network in three stages: (1) measuring students' subjective feelings of the question difficulty before practice; (2) estimating students' personalized knowledge acquisition while answering questions of different difficulty levels; (3) updating students' knowledge state in varying degrees to match the question difficulty level after practice. Finally, we conduct extensive experiments on real-world datasets, and the results demonstrate that DIMKT outperforms state-of-the-art KT models. Moreover, DIMKT shows superior interpretability by exploring the question difficulty effect when making predictions. Our codes are available at https://github.com/shshen-closer/DIMKT.

Network-aware cascade size prediction aims to predict the final reposted number of user-generated information via modeling the propagation process in social networks. Estimating the user's reposting probability by social influence, namely state activation plays an important role in the information diffusion process. Therefore, Graph Neural Networks (GNN), which can simulate the information interaction between nodes, has been proved as an effective scheme to handle this prediction task. However, existing studies including GNN-based models usually neglect a vital factor of user's preference which influences the state activation deeply. To that end, we propose a novel framework to promote cascade size prediction by enhancing the user preference modeling according to three stages, i.e., preference topics generation, preference shift modeling, and social influence activation. Our end-to-end method makes the user activating process of information diffusion more adaptive and accurate. Extensive experiments on two large-scale real-world datasets have clearly demonstrated the effectiveness of our proposed model compared to state-of-the-art baselines.

Click-through rate (CTR) prediction is fundamental in many industrial applications, such as online advertising and recommender systems. With the development of the online platforms, the sequential user behaviors grow rapidly, bringing us great opportunity to better understand user preferences.However, it is extremely challenging for existing sequential models to effectively utilize the entire behavior history of each user. First, there is a lot of noise in such long histories, which can seriously hurt the prediction performance. Second, feeding the long behavior sequence directly results in infeasible inference time and storage cost. In order to tackle these challenges, in this paper we propose a novel framework, which we name as User Behavior Clustering Sampling (UBCS). In UBCS, short sub-sequences will be obtained from the whole user history sequence with two cascaded modules: (i) Behavior Sampling module samples short sequences related to candidate items using a novel sampling method which takes relevance and temporal information into consideration; (ii) Item Clustering module clusters items into a small number of cluster centroids, mitigating the impact of noise and improving efficiency. Then, the sampled short sub-sequences will be fed into the CTR prediction module for efficient prediction. Moreover, we conduct a self-supervised consistency pre-training task to extract user persona preference and optimize the sampling module effectively. Experiments on real-world datasets demonstrate the superiority and efficiency of our proposed framework.

Cross-domain sentiment classification (CDSC) aims to use the transferable semantics learned from the source domain to predict the sentiment of reviews in the unlabeled target domain. Existing studies in this task attach more attention to the sequence modeling of sentences while largely ignoring the rich domain-invariant semantics embedded in graph structures (i.e., the part-of-speech tags and dependency relations). As an important aspect of exploring characteristics of language comprehension, adaptive graph representations have played an essential role in recent years. To this end, in the paper, we aim to explore the possibility of learning invariant semantic features from graph-like structures in CDSC. Specifically, we present Graph Adaptive Semantic Transfer (GAST) model, an adaptive syntactic graph embedding method that is able to learn domain-invariant semantics from both word sequences and syntactic graphs. More specifically, we first raise a POS-Transformer module to extract sequential semantic features from the word sequences as well as the part-of-speech tags. Then, we design a Hybrid Graph Attention (HGAT) module to generate syntax-based semantic features by considering the transferable dependency relations. Finally, we devise an Integrated aDaptive Strategy (IDS) to guide the joint learning process of both modules. Extensive experiments on four public datasets indicate that GAST achieves comparable effectiveness to a range of state-of-the-art models.

Computerized Adaptive Testing (CAT) is a promising testing mode in personalized online education (e.g., GRE), which aims at measuring student's proficiency accurately and reducing test length. The "adaptive" is reflected in its selection algorithm that can retrieve best-suited questions for student based on his/her estimated proficiency at each test step. Although there are many sophisticated selection algorithms for improving CAT's effectiveness, they are restricted and perturbed by the accuracy of current proficiency estimate, thus lacking robustness. To this end, we investigate a general method to enhance the robustness of existing algorithms by leveraging student's "multi-facet" nature during tests. Specifically, we present a generic optimization criterion Robust Adaptive Testing (RAT) for proficiency estimation via fusing multiple estimates at each step, which maintains a multi-facet description of student's potential proficiency. We further provide theoretical analyses of such estimator's desirable statistical properties: asymptotic unbiasedness, efficiency, and consistency. Extensive experiments on perturbed synthetic data and three real-world datasets show that selection algorithms in our RAT framework are robust and yield substantial improvements.

Knowledge tracing is a crucial and fundamental task in online education systems, which can predict students' knowledge state for personalized learning. Unfortunately, existing methods are domain-specific, whereas there are many domains (e.g., subjects, schools) in the real education scene and some domains suffer from the problem of lacking sufficient data. Therefore, how to exploit the knowledge in other domains, to improve the model's performance for target domain remains pretty much open. We term this problem as Domain Adaptation for Knowledge Tracing (DAKT), which aims to transfer knowledge from the source domain to the target one for knowledge tracing. In this paper, we propose a novel adaptable method, namely Adaptable Knowledge Tracing (AdaptKT), which contains three phases to explore this problem. Specifically, phase I is instance selection. Given the question texts of two domains, we train an auto-encoder to select and embed similar instances from both domains. Phase II is distribution discrepancy minimizing. After obtaining the selected instances and their linguistic representations, we train a knowledge tracing model and adopt the Maximum Mean Discrepancy (MMD) to minimize the discrepancy between the distributions of the domain-specific knowledge states. Phase III is fine-tuning of the output layer. We replace the output layer of the model that trained in phase II by a new one to make the knowledge tracing model's output dimension matches the number of knowledge concepts in the target domain. The new output layer is trained while other parameters that before it are frozen. We conduct extensive experiments on two large-scale real-world datasets, where the experimental results clearly demonstrate the effectiveness of AdaptKT for solving DAKT problem. We will public the code on the Github after the acceptance of the paper.

Recommendation is prevalently studied for implicit feedback recently, but it seriously suffers from the lack of negative samples, which has a significant impact on the training of recommendation models. Existing negative sampling is based on the static or adaptive probability distributions. Sampling from the adaptive probability receives more attention, since it tends to generate more hard examples, to make recommender training faster to converge. However, item sampling becomes much more time-consuming particularly for complex recommendation models. In this paper, we propose an Adaptive Sampling method based on Importance Resampling (AdaSIR for short), which is not only almost equally efficient and accurate for any recommender models, but also can robustly accommodate arbitrary proposal distributions. More concretely, AdaSIR maintains a contextualized sample pool of fixed-size with importance resampling, from which items are only uniformly sampled. Such a simple sampling method can be proved to provide approximately accurate adaptive sampling under some conditions. The sample pool plays two extra important roles in (1) reusing historical hard samples with certain probabilities; (2) estimating the rank of positive samples for weighting, such that recommender training can concentrate more on difficult positive samples. Extensive empirical experiments demonstrate that AdaSIR outperforms state-of-the-art methods in terms of sampling efficiency and effectiveness.

Variational AutoEncoder (VAE) has been extended as a representative nonlinear method for collaborative filtering. However, the bottleneck of VAE lies in the softmax computation over all items, such that it takes linear costs in the number of items to compute the loss and gradient for optimization. This hinders the practical use due to millions of items in real-world scenarios. Importance sampling is an effective approximation method, based on which the sampled softmax has been derived. However, existing methods usually exploit the uniform or popularity sampler as proposal distributions, leading to a large bias of gradient estimation. To this end, we propose to decompose the inner-product-based softmax probability based on the inverted multi-index, leading to sublinear-time and highly accurate sampling. Based on the proposed proposals, we develop a fast Variational AutoEncoder (FastVAE) for collaborative filtering. FastVAE can outperform the state-of-the-art baselines in terms of both sampling quality and efficiency according to the experiments on three real-world datasets.

Recent years have witnessed great success in deep learning-based sequential recommendation (SR), which can provide more timely and accurate recommendations. One of the most effective deep SR architectures is to stack high-performance residual blocks, e.g., prevalent self-attentive and convolutional operations, for capturing long- and short-range dependence of sequential behaviors. By carefully revisiting previous models, we observe: 1) simple architecture modification of gating each residual connection can help us train deeper SR models and yield significant improvements; 2) compared with self-attention mechanism, stacking of convolution layers also can cover each item of the whole sequential behaviors and achieve competitive or even superior performance.

Time-series prediction is of high practical value in a wide range of applications such as econometrics and meteorology, where the data are commonly formed by temporal patterns. Most prior works ignore the diversity of dynamic pattern frequency, i.e., different granularities, suffering from insufficient information exploitation. Thus, multi-granularity learning is still under-explored for time-series prediction. In this paper, we propose a Multi-granularity Residual Learning Framework (MRLF) for more effective time series prediction. For a given time series, intuitively, there are more or less semantic overlaps and validity differences among its representations of different granularities. Due to the information redundancy, straightforward methods that leverage multi-granularity data, such as concatenation or ensemble, can easily lead to the model being dominated by the redundant coarse-grained trend information. Therefore, we design a novel residual learning net to model the prior knowledge of the fine-grained data’s distribution through the coarse-grained one. Then, by calculating the residual between multi-granularity data, the redundant information be removed. Furthermore, to alleviate the side effect of validity differences, we introduce a self-supervised objective for confidence estimation, which delivers more effective optimization without the requirement of additional annotation efforts. Extensive experiments on the real-world datasets indicate that multi-granular information significantly improves the time series prediction performance, and our model is superior in capturing such information.

With the prosperity of recommender systems, the biases existing in user behaviors, which may lead to inconsistency between user preference and behavior records, have attracted wide attention. Though large efforts have been made to infer user preference from biased data with learning to debias, unfortunately, they mainly focus on the effect of one specific item attribute, e.g., position or modality which may affect users’ click probability on items. However, the comprehensive description for potential interactions between multiple items with various attributes, namely the context bias between items, may not be fully summarized. To that end, in this paper, we design a novel Context Bias aware Recommendation (CBR) model for describing and debiasing the context bias caused by comprehensive interactions between multiple items. Specifically, we first propose a content encoder and a bias encoder based on multi-head self-attention to embed the latent interactions between items. Then, we calculate the biased representation for users based on an attention network, which will be further utilized to infer the negative preference, i.e., the dislikes of users based on the items the user never clicked. Finally, the real user preference will be captured based on the negative preference to estimate the click prediction score. Extensive experiments on a real-world dataset demonstrate the competitiveness of our CBR framework compared with state-of-the-art baseline methods.

Emotion Cause Extraction (ECE) aims to reveal the cause clauses behind a given emotion expressed in a text, which has become an emerging topic in broad research communities, such as affective computing and natural language processing. Despite the fact that current methods about the ECE task have made great progress in text semantic understanding from lexicon- and sentence-level, they always ignore the certain causal narratives of emotion text. Significantly, these causal narratives are presented in the form of semantic structure and highly helpful for structure-level emotion cause understanding. Nevertheless, causal narrative is just an abstract narratological concept and its involving semantics is quite different from the common sequential information. Thus, how to properly model and utilize such particular narrative information to boost the ECE performance still remains an unresolved challenge. To this end, in this paper, we propose a novel Causal Narrative Comprehension Model (CNCM) for emotion cause extraction, which learns and leverages causal narrative information smartly to address the above problem. Specifically, we develop a Narrative-aware Causal Association (NCA) unit, which mines the narrative cue about emotional results and uses the semantic correlation between causes and results to model causal narratives of documents. Besides, we design a Result-aware Emotion Attention (REA) unit to make full use of the known result of causal narrative for multiple understanding about emotional causal associations. Through the ingenious combination and collaborative utilization of these two units, we could better identify the emotion cause in the text with causal narrative comprehension. Extensive experiments on the public English and Chinese benchmark datasets of ECE task have validated the effectiveness of CNCM with significant margin by comparing with the state-of-the-art baselines, which demonstrates the potential of narrative information in long text understanding.

In recent years, overlapping community detection algorithms have been paid more and more attention, which not only reveal the real social relations, but also expose the possible communication channels between communities. Those individuals (or people) in the overlapping area are very important to the communities that can promote communication between two or more communities. On the other hand, from the privacy perspective, some people may not want to be found out in the overlapping areas. With this in mind, we raise a question “Can individuals modify their relationships to avoid the community discovery algorithms locating them into overlapping areas?” If this problem could be solved, these people may not need to worry about being disturbed. In particular, we first give three heuristic hiding strategies, i.e., Random Hiding(RH), Based Degree Hiding(DH) and Betweenness Hiding(BH), as comparison, utilizing the randomly the node, information of node degree and node betweenness centrality, respectively. And then, we propose a novel hiding algorithm by exploiting the importance degree of nodes in communities based on which the corresponding social connections are added or deleted called nameBIH. Through extensive experiments, we show the effectiveness of the proposed algorithm in moving out a target node from overlapped areas.

In this big data era, knowledge becomes increasingly linked, along with the rapid growth in data volume. Connected knowledge is naturally represented and stored as knowledge graphs, which are of more and more importance for many frontier research areas such as machine intelligence. Effectively finding relations between entities in a large knowledge graph plays a key role in many knowledge graph applications, as the most valuable part of a knowledge graph is its rich connectedness, which captures rich information about the objects in the real world. However, due to the intrinsic complexity of real-world knowledge, finding semantically close relations by navigation in a large knowledge graph is very challenging. Canonical graph exploration methods inevitably result in combinatorial explosion especially when the paths connecting two entities are long: the search space is$O(d^l)$, where$d$is the average graph node degree and$l$is the length of the path. In this paper, we will systematically study the semantic navigation problem for large knowledge graphs. Inspired by AlphaGo, which was overwhelmingly successful in the game Go, we designed an efficient semantic navigation method based on a well-tailored Monte Carlo Tree Search algorithm with the unique characteristics of knowledge graphs considered. Extensive experiments on different real-life knowledge bases show that our method is not only effective but also very efficient.

Recent years have witnessed a successful rise of the time synchronizedgossiping comment, or so-called danmu combined with online videos. This new business mode has enriched communication among users by sending users’ feelings through danmus and sharing these danmus on time synchronized videos. Can danmu communication be helpful for better user behavior modeling or video analyzing? To this question, in this article, preliminary attempts are made on analysis of users and videos by introducing aDanmudataset which is collected from a real-world danmu-enabled video sharing platform. The dataset contains 1.7 TB of videos and danmus in total across eight video categories. With a focus on the 7.9 million danmus records and 4.8 million video frames, we first perform the basic statistic analysis and high-level semantic analysis. After that, we show some of the previous work on this area, including user behavior modeling, fine-grained video understanding and labeling, video plot generation and image-enhanced semantic understanding. For each application, we also propose its possible future directions. We hope this new dataset will inspire new ideas in areas among language, multimedia, and user understanding.

Short-term road traffic speed prediction is a long-standing topic in the area of Intelligent Transportation System. Apparently, effective prediction of the traffic speed on the road can not only provide timely details for the navigation system concerned and help the drivers to make better path selection, but also greatly improve the road supervision efficiency of the traffic department. At present, some researches on speed prediction based on GPS data, by adding weather and other auxiliary information, using graph convolutional neural network to capture the temporal and spatial characteristics, have achieved excellent results. In this paper, the problem of short-term traffic speed prediction based on GPS positioning data is further studied. For the processing of time series, we innovatively introduce Dynamic Time Warping algorithm into the problem and propose a Long Short-Term Memory with Dynamic Time Warping (D-LSTM) model. D-LSTM model, which integrates Dynamic Time Warping algorithm, can fine-tune the time feature, thus adjusting the current data distribution to be close to the historical data. More importantly, the fine-tuned data can still get a distinct improvement without special treatment of holidays. Meanwhile, considering that the data under different feature distributions have different effects on the prediction results, attention mechanism is also introduced in the model. Our experiments show that our proposed model D-LSTM performs better than other basic models in many kinds of traffic speed prediction problems with different time intervals, and especially significant in the traffic speed prediction on weekends.

One-class collaborative filtering (OCCF) problems are ubiquitous in real-world recommendation systems, such as news recommendation, but suffer from data sparsity and lack of negative items. To address the challenge, the state-of-the-art algorithm assigns uninteracted items with smaller weights of being negative and performs low-rank approximation over the user-item interaction matrix. However, the prior ratings are usually suggested to be zero but may not be well-defined. To avert the direct utilization of prior ratings for uninteracted items, we propose a novel ranking-based implicit regularizer by hypothesizing that users’ preference scores for uninteracted items should not deviate a lot from each other. The regularizer is then used in a ranking-based OCCF framework to penalize large differences of preference scores between uninteracted items. To efficiently optimize model parameters in this framework, we develop the scalable alternating least square algorithm and coordinate descent algorithm, whose time complexity is linearly proportional to the data size. Finally, we extensively evaluate the proposed algorithms on six public real-world datasets. The results show that the proposed regularizer significantly improves the recommendation quality of ranking-based OCCF algorithms, such as BPRMF and RankALS. Moreover, the ranking-based framework with the proposed regularizer outperforms the state-of-the-art recommendation algorithms for implicit feedback.

The implicit feedback based collaborative filtering (CF) has attracted much attention in recent years, mainly because users implicitly express their preferences in many real-world scenarios. The current mainstream pairwise methods optimize the Area Under the Curve (AUC) and are empirically proved to be helpful to exploit binary relevance data, but lead to either not address the ranking problem, or not specifically focus on top-$k$recommendation. Although there exists the listwise method maximizes the Mean Reciprocal Rank (MRR), it has low efficiency and is not particularly adequate for general implicit feedback situations. To that end, in this paper, we propose a new framework, namelyCollaborative List-and-Pairwise Filtering (CLAPF), which aims to introduce pairwise thinking into listwise methods. Specifically, we smooth another well-known rank-biased measure called Mean Average Precision (MAP), and respectively combine two rank-biased metrics (MAP, MRR) with the pairwise objective function to capture the performance of top-$k$recommendation. Furthermore, the sampling scheme for CLAPF is discussed to accelerate the convergence speed. Our CLAPF framework is a new hybrid model that provides an idea of utilizing rank-biased measures in a pairwise way on implicit feedback. Empirical studies demonstrated CLAPF outperforms state-of-the-art approaches on real-world datasets.

The development of intelligent traffic light control systems is essential for smart transportation management. While some efforts have been made to optimize the use of individual traffic lights in an isolated way, related studies have largely ignored the fact that the use of multi-intersection traffic lights is spatially influenced, as well as the temporal dependency of historical traffic status for current traffic light control. To that end, in this article, we propose a novel Spatio-Temporal Multi-Agent Reinforcement Learning (STMARL) framework for effectively capturing the spatio-temporal dependency of multiple related traffic lights and control these traffic lights in a coordinating way. Specifically, we first construct the traffic light adjacency graph based on the spatial structure among traffic lights. Then, historical traffic records will be integrated with current traffic status via Recurrent Neural Network structure. Moreover, based on the temporally-dependent traffic information, we design a Graph Neural Network based model to represent relationships among multiple traffic lights, and the decision for each traffic light will be made in a distributed way by the deep Q-learning method. Finally, the experimental results on both synthetic and real-world data have demonstrated the effectiveness of our STMARL framework, which also provides an insightful understanding of the influence mechanism among multi-intersection traffic lights.

Most of the recent image segmentation methods have tried to achieve the utmost segmentation results using large-scale pixel-level annotated data sets. However, obtaining these pixel-level annotated training data is usually tedious and expensive. In this work, we address the task of semisupervised semantic segmentation, which reduces the need for large numbers of pixel-level annotated images. We propose a method for semisupervised semantic segmentation by improving the confidence of the predicted class probability map via two parts. First, we build an adversarial framework that regards the segmentation network as the generator and uses a fully convolutional network as the discriminator. The adversarial learning makes the prediction class probability closer to 1. Second, the information entropy of the predicted class probability map is computed to represent the unpredictability of the segmentation prediction. Then, we infer the label-error map of the segmentation prediction and minimize the uncertainty on misclassified regions for unlabeled images. In contrast to existing semisupervised and weakly supervised semantic segmentation methods, the proposed method results in more confident predictions by focusing on the misclassified regions, especially the boundary regions. Our experimental results on the PASCAL VOC 2012 and PASCAL-CONTEXT data sets show that the proposed method achieves competitive segmentation performance.

Traffic anomalies, such as traffic accidents and unexpected crowd gathering, may endanger public safety if not handled timely. Detecting traffic anomalies in their early stage can benefit citizens' quality of life and city planning. However, traffic anomaly detection faces two main challenges. First, it is challenging to model traffic dynamics due to the complex spatiotemporal characteristics of traffic data. Second, the criteria of traffic anomalies may vary with locations and times. In this article, we propose a spatiotemporal graph convolutional adversarial network (STGAN) to address these above challenges. More specifically, we devise a spatiotemporal generator to predict the normal traffic dynamics and a spatiotemporal discriminator to determine whether an input sequence is real or not. There are high correlations between neighboring data points in the spatial and temporal dimensions. Therefore, we propose a recent module and leverage graph convolutional gated recurrent unit (GCGRU) to help the generator and discriminator learn the spatiotemporal features of traffic dynamics and traffic anomalies, respectively. After adversarial training, the generator and discriminator can be used as detectors independently, where the generator models the normal traffic dynamics patterns and the discriminator provides detection criteria varying with spatiotemporal features. We then design a novel anomaly score combining the abilities of two detectors, which considers the misleading of unpredictable traffic dynamics to the discriminator. We evaluate our method on two real-world datasets from New York City and California. The experimental results show that the proposed method detects various traffic anomalies effectively and outperforms the state-of-the-art methods. Furthermore, the devised anomaly score achieves more robust detection performances than the general score.

Social recommendation has achieved great success in many domains including e-commerce and location-based social networks. Existing methods usually explore the user-item interactions or user-user connections to predict users’ preference behaviors. However, they usually learn both user and item representations in Euclidean space, which has large limitations for exploring the latent hierarchical property in the data. In this article, we study a novel problem of hyperbolic social recommendation, where we aim to learn the compact but strong representations for both users and items. Meanwhile, this work also addresses two critical domain-issues, which are under-explored. First, users often make trade-offs with multiple underlying aspect factors to make decisions during their interactions with items. Second, users generally build connections with others in terms of different aspects, which produces different influences with aspects in social network. To this end, we propose a novel graph neural network (GNN) framework with multiple aspect learning, namely, HyperSoRec. Specifically, we first embed all users, items, and aspects into hyperbolic space with superior representations to ensure their hierarchical properties. Then, we adapt a GNN with novel multi-aspect message-passing-receiving mechanism to capture different influences among users. Next, to characterize the multi-aspect interactions of users on items, we propose an adaptive hyperbolic metric learning method by introducing learnable interactive relations among different aspects. Finally, we utilize the hyperbolic translational distance to measure the plausibility in each user-item pair for recommendation. Experimental results on two public datasets clearly demonstrate that our HyperSoRec not only achieves significant improvement for recommendation performance but also shows better representation ability in hyperbolic space with strong robustness and reliability.

Person search has long been treated as a crucial and challenging task to support deeper insight in personalized summarization and personality discovery. Traditional methods, e.g., person re-identification and face recognition techniques, which profile video characters based on visual information, are often limited by relatively fixed poses or small variation of viewpoints and suffer from more realistic scenes with high motion complexity (e.g., movies). At the same time, long videos such as movies often have logical story lines and are composed of continuously developmental plots. In this situation, different persons usually meet on a specific occasion, in which informative social cues are performed. We notice that these social cues could semantically profile their personality and benefit person search task in two aspects. First, persons with certain relationships usually co-occur in short intervals; in case one of them is easier to be identified, the social relation cues extracted from their co-occurrences could further benefit the identification for the harder ones. Second, social relations could reveal the association between certain scenes and characters (e.g., classmate relationship may only exist among students), which could narrow down candidates into certain persons with a specific relationship. In this way, high-level social relation cues could improve the effectiveness of person search. Along this line, in this article, we propose a social context-aware framework, which fuses visual and social contexts to profile persons in more semantic perspectives and better deal with person search task in complex scenarios. Specifically, we first segment videos into several independent scene units and abstract out social contexts within these scene units. Then, we construct inner-personal links through a graph formulation operation for each scene unit, in which both visual cues and relation cues are considered. Finally, we perform a relation-aware label propagation to identify characters’ occurrences, combining low-level semantic cues (i.e., visual cues) and high-level semantic cues (i.e., relation cues) to further enhance the accuracy. Experiments on real-world datasets validate that our solution outperforms several competitive baselines.

The job interview is considered as one of the most essential tasks in talent recruitment, which forms a bridge between candidates and employers in fitting the right person for the right job. While substantial efforts have been made on improving the job interview process, it is inevitable to have biased or inconsistent interview assessment due to the subjective nature of the traditional interview process. To this end, in this article, we propose three novel approaches to intelligent job interview by learning the large-scale real-world interview data. Specifically, we first develop a preliminary model, named Joint Learning Model on Interview Assessment (JLMIA), to mine the relationship among job description, candidate resume, and interview assessment. Then, we further design an enhanced model, named Neural-JLMIA, to improve the representative capability by applying neural variance inference. Last, we propose to refine JLMIA with Refined-JLMIA (R-JLMIA) by modeling individual characteristics for each collection, i.e., disentangling the core competences from resume and capturing the evolution of the semantic topics over different interview rounds. As a result, our approaches can effectively learn the representative perspectives of different job interview processes from the successful job interview records in history. In addition, we exploit our approaches for two real-world applications, i.e., person-job fit and skill recommendation for interview assessment. Extensive experiments conducted on real-world data clearly validate the effectiveness of our models, which can lead to substantially less bias in job interviews and provide an interpretable understanding of job interview assessment.

As a major component of strategic talent management, learning and development (L&D) aims at improving the individual and organization performances through planning tailored training for employees to increase and improve their skills and knowledge. While many companies have developed the learning management systems (LMSs) for facilitating the online training of employees, a long-standing important issue is how to achieve personalized training recommendations with the consideration of their needs for future career development. To this end, in this article, we present a focused study on the explainable personalized online course recommender system for enhancing employee training and development. Specifically, we first propose a novel end-to-end hierarchical framework, namely Demand-aware Collaborative Bayesian Variational Network (DCBVN), to jointly model both the employees’ current competencies and their career development preferences in an explainable way. In DCBVN, we first extract the latent interpretable representations of the employees’ competencies from their skill profiles with autoencoding variational inference based topic modeling. Then, we develop an effective demand recognition mechanism for learning the personal demands of career development for employees. In particular, all the above processes are integrated into a unified Bayesian inference view for obtaining both accurate and explainable recommendations. Furthermore, for handling the employees with sparse or missing skill profiles, we develop an improved version of DCBVN, called the Demand-aware Collaborative Competency Attentive Network (DCCAN) framework , by considering the connectivity among employees. In DCCAN, we first build two employee competency graphs from learning and working aspects. Then, we design a graph-attentive network and a multi-head integration mechanism to infer one’s competency information from her neighborhood employees. Finally, we can generate explainable recommendation results based on the competency representations. Extensive experimental results on real-world data clearly demonstrate the effectiveness and the interpretability of both of our frameworks, as well as their robustness on sparse and cold-start scenarios.

Reservoir classification is an essential step for the exploration and production process in the oil and gas industry. An appropriate automatic reservoir classification will not only reduce the manual workloads of experts, but also help petroleum companies to make optimal decisions efficiently, which in turn will dramatically reduce the costs. Existing methods mainly focused on generating reservoir classification in a single geological block but failed to work well on a new oilfield block. Indeed, how to transfer the subsurface characteristics and make accurate reservoir classification across the geological oilfields is a very important but challenging problem. To that end, in this paper, we present a focused study on the cross-oilfield reservoir classification task. Specifically, we first propose a Multi-scale Sensor Extraction (MSE) to extract the multi-scale feature representations of geological characteristics from multivariate well logs. Furthermore, we design an encoder-decoder module, Specific Feature Learning (SFL), to take advantage of specific information of both oilfields. Then, we develop a Knowledge-Attentive Transfer (KAT) module to learn the feature-invariant representation and transfer the geological knowledge from a source oilfield to a target oilfield. Finally, we evaluate our approaches by conducting extensive experiments with real-world industrial datasets. The experimental results clearly demonstrate the effectiveness of our proposed approaches to transfer the geological knowledge and generate the cross-oilfield reservoir classifications.

Automatically solving math word problems is a crucial task for exploring the intelligence levels of machines in the general AI domain. It is highly challenging since it requires not only natural language understanding but also mathematical expression inference. Existing solutions usually explore sequence-to-sequence models to generate expressions, where the problems are simply encoded sequentially. However, such models are generally far from enough for understanding problems as similar to humans and lead to incorrect answers. To this end, in this paper, we propose a novel Hierarchical Math Solver (HMS) to make deep understanding and exploitation of problems. In problem understanding, imitating human reading habits, we propose a hierarchical word-clause-problem encoder. Specifically, we first split each problem into several clauses and learn problem semantics from the local clause level to the global problem level. Then, in clause understanding, we propose a dependency-based module to enhance clause semantics with the dependency structure of the problem. Next, in expression inference, we propose a novel tree-based decoder to generate the mathematical expression for the answer. In the decoder, we apply a hierarchical attention mechanism to enhance the problem semantics with context from different levels, and a pointer-generator network to guide the model to copy existing information and infer extra knowledge. Extensive experimental results on two widely used datasets demonstrate that HMS achieves not only better answers but also more reasonable inference.

Cognitive psychology research shows that humans have the instinct for abstract thinking, where association plays an essential role in language comprehension. Especially for Chinese, its ideographic writing system allows radicals to trigger semantic association without the need of phonetics. In fact, subconsciously using the associative information guided by radicals is a key for readers to ensure the robustness of semantic understanding. Fortunately, many basic and extended concepts related to radicals are systematically included in Chinese language dictionaries, which leaves a handy but unexplored way for improving Chinese text representation and classification. To this end, we draw inspirations from cognitive principles between ideography and human associative behavior to propose a novel Radical-guided Associative Model (RAM) for Chinese text classification. RAM comprises two coupled spaces, namely Literal Space and Associative Space, which imitates the real process in people's mind when understanding a Chinese text. To be specific, we first devise a serialized modeling structure in Literal Space to thoroughly capture the sequential information of Chinese text. Then, based on the authoritative information provided by Chinese language dictionaries, we design an association module and put forward a strategy called Radical-Word Association to use ideographic radicals as the medium to associate prior concept words in Associative Space. Afterwards, we design an attention module to imitate people's matching and decision between Literal Space and Associative Space, which can balance the importance of each associative words under specific contexts. Finally, extensive experiments on two real-world datasets prove the effectiveness and rationality of RAM, with good cognitive insights for future language modeling.

While the stock movement prediction has been intensively studied, existing work suffers from weak generalization because of the uncertainty in both data and modeling. On one hand, training a stock representation on stochastic stock data in an end-to-end manner may lead to excessive modeling, which involves model uncertainty. On the other, the analysis of correlating stock data with its relevant factors involves data uncertainty. To simultaneously address such uncertainty both from data and modeling perspectives, a fundamental yet challenging task is to learn a better stock representation with less uncertainty by considering hierarchical couplings from the macro-level to the sector-and micro-level. Accordingly, we propose a copula-based contrastive predictive coding (Co-CPC) method. Co-CPC first models the dependence between a certain stock sector and relevant macroeconomic variables that are sequential and heterogeneous, e.g., macro-variables are associated with different time intervals, scales, and distributions. Then, by involving a macro-sector context, stock representations are learned in a self-supervised way that can further be used for downstream tasks like stock movement prediction. Extensive experiments on two typical stock datasets verify the effectiveness of our Co-CPC method.

Sentence semantic matching is one of the fundamental tasks in natural language processing, which requires an agent to determine the semantic relation among input sentences. Recently, deep neural networks have achieved impressive performance in this area, especially BERT. Despite the effectiveness of these models, most of them treat output labels as meaningless one-hot vectors, underestimating the semantic information and guidance of relations that these labels reveal, especially for tasks with a small number of labels. To address this problem, we propose a Relation of Relation Learning Network (R2-Net) for sentence semantic matching. Specifically, we first employ BERT to encode the input sentences from a global perspective. Then a CNN-based encoder is designed to capture keywords and phrase information from a local perspective. To fully leverage labels for better relation information extraction, we introduce a self-supervised relation of relation classification task for guiding R2-Net to consider more about labels. Meanwhile, a triplet loss is employed to distinguish the intra-class and inter-class relations in a finer granularity. Empirical experiments on two sentence semantic matching tasks demonstrate the superiority of our proposed model. As a byproduct, we have released the codes to facilitate other researches.

Stock trend prediction plays a crucial role in quantitative investing. Given the prediction task on a certain granularity (e.g., daily trend), a large portion of existing studies merely leverage market data of the same granularity (e.g., daily market data). In financial investment scenarios, however, there exist amounts of finer-grained information (e.g., high-frequency data) that contain more detailed investment signals beyond the original granularity data. This motivates us to investigate how to leverage multi-granularity market data to enhance the accuracy of stock trend prediction. Some straightforward methods, such as concatenating finer-grained data as features or fusing with a model based on finer-grained features, may not lead to more precise stock trend prediction due to some unique challenges. First, the inconsistency of granularity between the target trend and finer-grained data could substantially increase optimization difficulty, such as the relative sparsity of the target trend compared with higher dimensions of finer-grained features. Moreover, the continuously changing financial market state could result in varying efficacy of heterogeneous multi-granularity information, which consequently requires a dynamic approach for proper fusion among them. In this paper, we propose the Contrastive Multi-Granularity Learning Framework (CMLF) to address these challenges. Particularly, we first design two novel contrastive learning objectives at the pre-training stage to address the inconsistency issue by constructing additional self-supervised signals relying on the inherent character of stock data. We also design a gate mechanism based on market-aware technical indicators to fuse the multi-granularity features at each time step adaptively. Extensive experiments on three real-world datasets show significant improvements of our approach over the state-of-the-art baselines on stock trend prediction and profitability in real investing scenarios.

Recent studies in recommender systems have managed to achieve significantly improved performance. However, despite being extensively studied, these methods still suffer from two limitations. First, previous studies either encode the document or extract latent sentiment via neural networks, which are difficult to interpret the sentiment of reviewers intuitively. Second, they neglect the personalized interaction of reviews with user/item, i.e., each review has different contributions when modeling the preference of user/item

This paper proposes a novel architecture, Cross Attention Augmented Transducer (CAAT), for simultaneous translation.The framework aims to jointly optimize the policy and translation models.To effectively consider all possible READ-WRITE simultaneous translation action paths, we adapt the online automatic speech recognition (ASR) model, RNN-T, but remove the strong monotonic constraint, which is critical for the translation task to consider reordering.To make CAAT work, we introduce a novel latency loss whose expectation can be optimized by a forward-backward algorithm.We implement CAAT with Transformer while the general CAAT architecture can also be implemented with other attention-based encoder-decoder frameworks.Experiments on both speech-to-text (S2T) and text-to-text (T2T) simultaneous translation tasks show that CAAT achieves significantly better latency-quality trade-offs compared to the state-of-the-art simultaneous translation approaches. 1

Text to speech (TTS) has been broadly used to synthesize natural and intelligible speech in different scenarios. Deploying TTS in various end devices such as mobile phones or embedded devices requires extremely small memory usage and inference latency. While non-autoregressive TTS models such as FastSpeech have achieved significantly faster inference speed than autoregressive models, their model size and inference latency are still large for the deployment in resource constrained devices. In this paper, we propose LightSpeech, which leverages neural architecture search (NAS) to automatically design more lightweight and efficient models based on FastSpeech. We first profile the components of current Fast-Speech model and carefully design a novel search space containing various lightweight and potentially effective architectures. Then NAS is utilized to automatically discover well performing architectures within the search space. Experiments show that the model discovered by our method achieves 15x model compression ratio and 6.5x inference speedup on CPU with on par voice quality. Audio demos are provided at https://speechresearch.github.io/lightspeech.

Text-to-image synthesis refers to generating an image from a given text description, the key goal of which lies in photo realism and semantic consistency. Previous methods usually generate an initial image with sentence embedding and then refine it with fine-grained word embedding. Despite the significant progress, the ‘aspect’ information (e.g., red eyes) contained in the text, referring to several words rather than a word that depicts ‘a particular part or feature of something’, is often ignored, which is highly helpful for synthesizing image details. How to make better utilization of aspect information in text-to-image synthesis still remains an unresolved challenge. To address this problem, in this paper, we propose a Dynamic Aspect-awarE GAN (DAE-GAN) that represents text information comprehensively from multiple granularities, including sentence-level, word-level, and aspect-level. Moreover, inspired by human learning behaviors, we develop a novel Aspect-aware Dynamic Re-drawer (ADR) for image refinement, in which an Attended Global Refinement (AGR) module and an Aspect-aware Local Refinement (ALR) module are alternately employed. AGR utilizes word-level embedding to globally enhance the previously generated image, while ALR dynamically employs aspect-level embedding to refine image details from a local perspective. Finally, a corresponding matching loss function is designed to ensure the text-image semantic consistency at different levels. Extensive experiments on two well-studied and publicly available datasets (i.e., CUB-200 and COCO) demonstrate the superiority and rationality of our method.

Learning effective user representations from sequential user-item interactions is a fundamental problem for recommender systems (RS). Recently, several unsupervised methods focusing on user representations pre-training have been explored. In general, these methods apply similar learning paradigms by first corrupting the behavior sequence, and then restoring the original input with some item-level prediction loss functions. Despite its effectiveness, we argue that there exist important gaps between such item-level optimization objective and user-level representations, and as a result, the learned user representations may only lead to sub-optimal generalization performance. In this paper, we propose a novel self-supervised pre-training framework, called CLUE, which stands for employing Contrastive Learning for modeling sequence-level User rEpresentation. The core idea of CLUE is to regard each user behavior sequence as a whole and then construct the self-supervision signals by transforming the original user behaviors by data augmentations (DA). Specifically, we employ two Siamese (weight-sharing) networks to learn the user-oriented representations, where the optimization goal is to maximize the similarity of learned representations of the same user by these two encoders. More importantly, we perform careful investigation of the impacts of view generating strategies for user behavior inputs from a more comprehensive perspective, including processing sequential behaviors by explicit DA strategies and employing dropout as implicit DA. To verify the effectiveness of CLUE, we perform extensive experiments on several user-related tasks with different scales and characteristics. Our experimental results show that the user representations learned by CLUE surpass existing item-level baselines under several evaluation protocols.

In intelligent education systems, question difficulty prediction (QDP) is a fundamental task of many applications, such as personalized question recommendation and test paper analysis. Previous work mainly focus on data-driven QDP methods, which are heavily relied on the large-scale labeled dataset of courses. To alleviate the labor intensity, an intuitive method is to introduce domain adaptation into QDP and consider each course as a domain. In educational psychology, there are two factors influencing difficulty common to different courses: the obstacles of comprehending the question and generating a response, namely stimulus and task difficulty. To this end, we propose a novel Stimulus and Task difficulty-based Adversarial Network (STAN) that models question difficulty from the views of stimulus and task. Then, in order to align the difficulty distribution of the source domain and the target domain, we utilize the conditional adversarial learning with readability-enhanced pseudo-labels. Meanwhile, we proposed a sampling method based on density estimation to implicit alignment. Finally, we conduct experiments on the real questions datasets to evaluate the effectiveness of our QDP model and domain adaptation method. Our method significantly improves accuracy over state-of-the-art methods on real-world question data of multiple courses.

Most cognitive diagnosis research in education has been concentrated on individual assessment, aiming at discovering the latent characteristics of students. However, in many real-world scenarios, group-level assessment is an important and meaningful task, e.g., class assessment in different regions can discover the difference of teaching level in different contexts. In this work, we consider assessing cognitive ability for a group of students, which aims to mine groups’ proficiency on specific knowledge concepts. The significant challenge in this task is the sparsity of group-exercise response data, which seriously affects the assessment performance. Existing works either do not make effective use of additional student-exercise response data or fail to reasonably model the relationship between group ability and individual ability in different learning contexts, resulting in sub-optimal diagnosis results. To this end, we propose a general Multi-Task based Group-Level Cognitive Diagnosis (MGCD) framework, which is featured with three special designs: 1) We jointly model student-exercise responses and group-exercise responses in a multi-task manner to alleviate the sparsity of group-exercise responses; 2) We design a context-aware attention network to model the relationship between student knowledge state and group knowledge state in different contexts; 3) We model an interpretable cognitive layer to obtain student ability, group ability and exercise factors (e.g., difficulty), and then we leverage neural networks to learn complex interaction functions among them. Extensive experiments on real-world datasets demonstrate the generality of MGCD and the effectiveness of our attention design and multi-task learning.

With the accelerated technology development, technological trend forecasting through patent mining has become a hot issue for high-tech companies. In this term, extensive attention has been attracted to forecasting technological knowledge flows (TKF), i.e., predicting the directional flows of knowledge from one technological field to another. However, existing studies either rely on labor intensive empirical analysis or do not consider the intrinsic characteristics inherent in TKF, including the double-faced aspects (i.e., act as both the source and target) of technology nodes, multiple complex relationships among different technologies, and dynamics of the TKF process. To this end, in this paper, we make a further study and propose a data-driven solution, i.e., a Hierarchical Interactive Graph Neural Network (HighTKF), to automatically find the potential flow trends of technologies. Specifically, HighTKF makes final predictions through two kinds of representations of each technology node (a diffusion vector and an absorption vector), which is realized by three components: High-Order Interaction Module (HOI), Hierarchical Delivery Module (HD) and Technology Flow Tracing Module (TFT). For one thing, HOI and HD aim to model high-order network relationships and hierarchical relationships among technologies. For another, TFT is designed for capturing the dynamic feature evolution of technologies with the above relations involved. Also, we design a hybrid loss function and propose a new evaluation metric for better predicting the unprecedented flows between technologies. Finally, we conduct extensive experiments on a real-world patent dataset, the results verify the effectiveness of our approach and reveal some interesting phenomenons on technological knowledge flow trends.

Recent years have witnessed the growing research interests in video highlight detection. Existing studies mainly focus on detecting highlights in user-generated videos with simple topics based on visual content. However, relying solely on visual features limits the ability of conventional methods to capture highlights for videos with more complicated semantics, like movies. Therefore, we propose to mine the emotional information in video sounds to enhance highlight detection. Specifically, we design a novel emotion-enhanced framework with multi-stage fusion to detect highlights for complex videos. Along this line, we first extract multi-grained features from the audio waves. Then, the tailored-designed intra-modal fusion is applied on audio features to obtain emotional representation. Furthermore, the cross-modal fusion is developed to generate comprehensive representation of clip by merging audio emotional representations and visual features. This representation can be leveraged for predicting highlight probability. Finally, extensive experiments on real-world datasets demonstrate the effectiveness of our method.

As one of the most popular generative models, Variational Autoencoder (VAE) approximates the posterior of latent variables based on amortized variational inference. However, when the decoder network is sufficiently expressive, VAE may lead to posterior collapse; that is, uninformative latent representations may be learned. To this end, in this paper, we propose an alternative model, DU-VAE, for learning a more Diverse and less Uncertain latent space, and thus the representation can be learned in a meaningful and compact manner. Specifically, we first theoretically demonstrate that it will result in better latent space with high diversity and low uncertainty awareness by controlling the distribution of posterior’s parameters across the whole data accordingly. Then, without the introduction of new loss terms or modifying training strategies, we propose to exploit Dropout on the variances and Batch-Normalization on the means simultaneously to regularize their distributions implicitly. Furthermore, to evaluate the generalization effect, we also exploit DU-VAE for inverse autoregressive flow based-VAE (VAE-IAF) empirically. Finally, extensive experiments on three benchmark datasets clearly show that our approach can outperform state-of-the-art baselines on both likelihood estimation and underlying classification tasks.

In recommender systems, the cold-start problem is a critical issue. To alleviate this problem, an emerging direction adopts meta-learning frameworks and achieves success. Most existing works aim to learn globally shared prior knowledge across all users so that it can be quickly adapted to a new user with sparse interactions. However, globally shared prior knowledge may be inadequate to discern users’ complicated behaviors and causes poor generalization. Therefore, we argue that prior knowledge should be locally shared by users with similar preferences who can be recognized by social relations. To this end, in this paper, we propose a Preference-Adaptive Meta-Learning approach (PAML) to improve existing meta-learning frameworks with better generalization capacity. Specifically, to address two challenges imposed by social relations, we first identify reliable implicit friends to strengthen a user’s social relations based on our defined palindrome paths. Then, a coarse-fine preference modeling method is proposed to leverage social relations and capture the preference. Afterwards, a novel preference-specific adapter is designed to adapt the globally shared prior knowledge to the preference-specific knowledge so that users who have similar tastes share similar knowledge. We conduct extensive experiments on two publicly available datasets. Experimental results validate the power of social relations and the effectiveness of PAML.

As machine learning becomes more widely used for critical applications, the need to study its implications in privacy becomes urgent. Given access to the target model and auxiliary information, model inversion attack aims to infer sensitive features of the training dataset, which leads to great privacy concerns. Despite its success in the grid domain, directly applying model inversion techniques on non grid domains such as graph achieves poor attack performance due to the difficulty to fully exploit the intrinsic properties of graphs and attributes of graph nodes used in GNN models. To bridge this gap, we present Graph Model Inversion attack, which aims to infer edges of the training graph by inverting Graph Neural Networks, one of the most popular graph analysis tools. Specifically, the projected gradient module in our method can tackle the discreteness of graph edges while preserving the sparsity and smoothness of graph features. Moreover, a well designed graph autoencoder module can efficiently exploit graph topology, node attributes, and target model parameters. With the proposed method, we study the connection between model inversion risk and edge influence and show that edges with greater influence are more likely to be recovered. Extensive experiments over several public datasets demonstrate the effectiveness of our method. We also show that differential privacy in its canonical form can hardly defend our attack while preserving decent utility.

Recent years have witnessed the booming of venture capital industry, which may not only bring huge benefits to investors, startups and emerging industries, but also lead to potential risk due to the difficulty to find proper investment targets, especially for inexperienced investors. Therefore, the investment target recommendation task has long been treated as a valuable issue to help the investors for the next investments based on their historical behavior and the market trend. However, due to the sequential features of historical invest records, which could be severely disturbed by the time-varying factors like market trend or short-term invest preference, it will be extremely difficult to accurately reveal the potential rule of investment. Moreover, the latent relationships between investors and targets (i.e., startups) or brands, which formed a heterogeneous graph to reflect the mutual influence within investors, have been largely ignored. To that end, in this paper, we propose a novel paradigm for investment target recommendation with integrating the sequential records and heterogeneous relation graph. Specifically, we aggregate information at the micro level through a time segmentation method, and then extract trends at the macro level. Meanwhile, we use relational graph convolutional networks to propagate the latent preferences through the various relationships. Based on the validations on the realworld dataset, by fusing above information, the Investment Target Recommendation System (ITRS) we proposed has achieved impressive results compared with competitive baseline methods, which indicates the significance of both sequential and structural information.

Conventional knowledge graph embedding methods mainly assume that all entities at reasoning stage are available in the original training graph. But in real-world application scenarios, newly emerged entities are always inevitable, which results in the severe problem of out-of-knowledge-graph entities. Existing efforts on this issue mostly either utilize additional resources, e.g., entity descriptions, or simply aggregate in-knowledge-graph neighbors to embed these new entities inductively. However, high-quality additional resources are usually hard to obtain and existing neighbors of new entities may be too sparse to provide enough information for modeling these entities. Meanwhile, they may fail to integrate the rich information of ontological concepts, which provide a general figure of instance entities and usually remain unchanged in knowledge graph. To this end, we propose a novel inductive framework namely CatE to solve the sparsity problem with the enhancement from ontological concepts. Specifically, we first adopt the transformer encoder to model the complex contextual structure of the ontological concepts. Then, we further develop a template refinement strategy for generating the target entity embedding, where the concept embedding is used to form a basic skeleton of the target entity and the individual characteristics of the entity will be enriched by its existing neighbors. Finally, extensive experiments on public datasets demonstrate the effectiveness of our proposed model compared with state-of-the-art baseline methods.

Assessing the proficiency of trial lawyers in different legal fields is of significant importance since a qualified lawyer or lawyer team can strive for his clients' best rights while ensuring the fairness of litigations. However, proficiency assessment for lawyers is very challenging due to many technical and domain challenges, such as the lack of unified evaluation standards, and the complex interactions between lawyers and cases in real legal systems. To this end, we propose a novel proficiency assessment network for trial lawyers (LawyerPAN) to quantify lawyer proficiency through online litigation records. Specifically, we first leverage the theories in psychological measurement for mapping the proficiency of lawyers in each field into a unified real number space. Meanwhile, the characteristics of cases (i.e., case difficulty and discrimination) are well modeled to ensure fairness when assessing lawyers in different cases and fields. Then, we model the interactions between lawyers and cases from two perspectives: the anticipatory perspective aims to measure the personal proficiency of anticipated strategy, and the adversarial perspective seeks to depict the gap of lawyers' proficiency between both sides (i.e., plaintiffs and defendants). Finally, we conduct extensive experiments on real-world data, and the results show the effectiveness and interpretability of our approaches on assessing the proficiency of trial lawyers.

Learning informative representations for educational questions is a fundamental problem in online learning systems, which can promote many applications, e.g., difficulty estimation. Most solutions integrate all information of one question together following a supervised manner, where the representation results are unsatisfactory sometimes due to the following issues. First, they cannot ensure the presentation ability due to the scarcity of labeled data. Then, the label-dependent representation results have poor feasibility to be transferred. Moreover, aggregating all information into the unified may introduce some noises in applications since it cannot distinguish the diverse characteristics of questions. In this paper, we aim to learn the disentangled representations of questions. We propose a novel unsupervised model, namely DisenQNet, to divide one question into two parts, i.e., a concept representation that captures its explicit concept meaning and an individual representation that preserves its personal characteristics. We achieve this goal via mutual information estimation by proposing three self-supervised estimators in a large unlabeled question corpus. Then, we propose another enhanced model, DisenQNet+, that transfers the representation knowledge from unlabeled questions to labeled questions in specific applications by maximizing the mutual information between both. Extensive experiments on real-world datasets demonstrate that DisenQNet can generate effective and meaningful disentangled representations for questions, and furthermore, DisenQNet+ can improve the performance of different applications.

Approximate nearest neighbor search (ANNs) plays an important role in many applications ranging from information retrieval, recommender systems to machine translation. Several ANN indexes, such as hashing and quantization, have been designed to update for the evolving database, but there exists a remarkable performance gap between them and retrained indexes on the entire database. To close the gap, we propose an online additive quantization algorithm (online AQ) to dynamically update quantization codebooks with the incoming streaming data. Then we derive the regret bound to theoretically guarantee the performance of the online AQ algorithm. Moreover, to improve the learning efficiency, we develop a randomized block beam search algorithm for assigning each data to the codewords of the codebook. Finally, we extensively evaluate the proposed online AQ algorithm on four real-world datasets, showing that it remarkably outperforms the state-of-the-art baselines.

Knowledge tracing (KT), which aims to trace students' changing knowledge state during their learning process, has improved students' learning efficiency in online learning systems. Recently, KT has attracted much research attention due to its critical significance in education. However, most of the existing KT methods pursue high accuracy of student performance prediction but neglect the consistency of students' changing knowledge state with their learning process. In this paper, we explore a new paradigm for the KT task and propose a novel model named Learning Process-consistent Knowledge Tracing (LPKT), which monitors students' knowledge state through directly modeling their learning process. Specifically, we first formalize the basic learning cell as the tuple exercise---answer time---answer. Then, we deeply measure the learning gain as well as its diversity from the difference of the present and previous learning cells, their interval time, and students' related knowledge state. We also design a learning gate to distinguish students' absorptive capacity of knowledge. Besides, we design a forgetting gate to model the decline of students' knowledge over time, which is based on their previous knowledge state, present learning gains, and the interval time. Extensive experimental results on three public datasets demonstrate that LPKT could obtain more reasonable knowledge state in line with the learning process. Moreover, LPKT also outperforms state-of-the-art KT methods on student performance prediction. Our work indicates a potential future research direction for KT, which is of both high interpretability and accuracy.

As an important means to solve information overload, recommender systems have been widely applied in many fields, such as e-commerce and advertising. However, recent studies have shown that recommender systems are vulnerable to poisoning attacks; that is, injecting a group of carefully designed user profiles into the recommender system can severely affect recommendation quality. Despite the development from shilling attacks to optimization-based attacks, the imperceptibility and harmfulness of the generated data in most attacks are arduous to balance. To this end, we propose a triple adversarial learning for influence based poisoning attack (TrialAttack), a flexible end-to-end poisoning framework to generate non-notable and harmful user profiles. Specifically, given the input noise, TrialAttack directly generates malicious users through triple adversarial learning of the generator, discriminator, and influence module. Besides, to provide reliable influence for TrialAttack training, we explore a new approximation approach for estimating each fake user's influence. Through theoretical analysis, we prove that the distribution characterized by TrialAttack approximates to the rating distribution of real users under the premise of performing an efficient attack. This property allows the injected users to attack in an unremarkable way. Experiments on three real-world datasets show that TrialAttack's attack performance outperforms state-of-the-art attacks, and the generated fake profiles are more difficult to detect compared to baselines.

Job mobility prediction is an emerging research topic that can benefit both organizations and talents in various ways, such as job recommendation, talent recruitment, and career planning. Nevertheless, most existing studies only focus on modeling the individual-level career trajectories of talents, while the impact of macro-level job transition relationships (e.g., talent flow among companies and job positions) has been largely neglected. To this end, in this paper we propose an enhanced approach to job mobility prediction based on a heterogeneous company-position network constructed from the massive career trajectory data. Specifically, we design an Attentive heterogeneous graph embedding for sequential prediction (Ahead) framework to predict the next career move of talents, which contains two components, namely an attentive heterogeneous graph embedding (AHGN) model and a Dual-GRU model for career path mining. In particular, the AHGN model is used to learn the comprehensive representation for company and position on the heterogeneous network, in which two kinds of aggregators are employed to aggregate the information from external and internal neighbors for a node. Afterwards, a novel type-attention mechanism is designed to automatically fuse the information of the two aggregators for updating node representations. Moreover, the Dual-GRU model is devised to model the parallel sequences that appear in pair, which can be used to capture the sequential interactive information between companies and positions. Finally, we conduct extensive experiments on a real-world dataset for evaluating our Ahead framework. The experimental results clearly validate the effectiveness of our approach compared with the state-of-the-art baselines in terms of job mobility prediction.

The contexts and cultures have a direct impact on student learning by affecting student's implicit cognitive states, such as the preference and the proficiency on specific knowledge. Motivated by the success of context-aware modeling in various fields, such as recommender systems, in this paper, we propose to study how to model context-aware features and adapt them for more precisely diagnosing student's knowledge proficiency. Specifically, by analyzing the characteristics of educational contexts, we design a two-stage framework ECD (Educational context-aware Cognitive Diagnosis), where a hierarchical attentive network is first proposed to represent the context impact on students and then an adaptive optimization is used to achieve diagnosis enhancement by aggregating the cognitive states reflected from both educational contexts and students' historical learning records. Moreover, we give three implementations of general ECD framework following the typical cognitive diagnosis solutions. Finally, we conduct extensive experiments on nearly 52 million records of the students sampled by PISA (Programme for International Student Assessment) from 73 countries and regions. The experimental results not only prove that ECD is more effective in student performance prediction since it can well capture the impact from educational contexts to students' cognitive states, but also give some interesting discoveries regarding the difference among different educational contexts in different countries and regions.

Urban crowd density prediction, which predicts the future crowd density in different areas based on the historical data, is playing an increasingly significant role in epidemic prevention and traffic optimization. Most existing methods model the spatial information through a single relationship, i.e., distance, and extract the temporal information only by short time sequences, which limits the model to fully capture the spatiotemporal information. Therefore, in this paper, we propose a Multi-relational Graph Convolutional Gate Recurrent Unit (MGC-GRU) model to represent the spatiotemporal information more comprehensively for better urban crowd density prediction. Specifically, we first construct a multi-relation urban area graph to enrich the spatial relationship between areas. Then a graph representation module based on a multi-relational graph convolution network is proposed to represent spatial information of the area, in which aggregator distinguishes the information of different relationships and propagator equips the self-attention mechanism to refine the representation. Afterwards, we further construct a fine-grained sequence prediction module to enhance the temporal dependency by modeling time sequences in different granularity, i.e., daily and hourly. Finally, extensive experiments on a real-world dataset demonstrate the superior performance of MGC-GRU on urban crowd density prediction task.

Recent years have witnessed the booming of online video platforms. Along this line, a graph to illustrate social relation among characters has been long expected to not only benefit the audiences for better understanding the story, but also support the fine-grained video analysis task in a semantic way. Unfortunately, though we humans could easily infer the social relations among characters, it is still an extremely challenging task for intelligent systems to automatically capture the social relation by absorbing multi-modal cues. Besides, they fail to describe the relations among multiple characters in a graph-generation perspective. To that end, inspired by the human inference ability on social relationship, we propose a novel Hierarchical- Cumulative Graph Convolutional Network (HC-GCN) to generate the social relation graph for multiple characters in the video. Specifically, we first integrate the short-term multi-modal cues, including visual, textual and audio information, to generate the frame-level graphs for part of characters via multimodal graph convolution technique. While dealing with the video-level aggregation task, we design an end-to-end framework to aggregate all frame-level subgraphs along the temporal trajectory, which results in a global video-level social graph with various social relationships among multiple characters. Extensive validations on two real-world large-scale datasets demonstrate the effectiveness of our proposed method compared with SOTA baselines.

Micro-expressions (MEs) are significant and effective clues to reveal the true feelings and emotions of human beings, and thus MEs analysis is widely used in different fields such as medical diagnosis, interrogation and security. However, it is extremely difficult to elicit and label MEs, resulting in a lack of sufficient MEs data for MEs analysis. To address this challenge and inspired by the current face generation technology, in this paper we introduce Generative Adversarial Network based on fine-grained Action Units (AUs) modulation to generate MEs sequence (FAMGAN). Specifically, after comprehensively analyzing the factors that lead to inaccurate AU values detection, we performed fine-grained AUs modulation, which includes carefully eliminating the various noises and dealing with the asymmetry of AUs intensity. Additionally, we incorporate super-resolution into our model to enhance the quality of the generated images. Through experiments, we show that the system achieves very competitive results on the Micro-Expression Grand Challenge (MEGC2021).

One-hot encoder accompanied by a softmax loss has become the default configuration to deal with the multiclass problem, and is also prevalent in deep learning (DL) based recommender systems (RS). The standard learning process of such methods is to fit the model outputs to a one-hot encoding of the ground truth, referred to as the hard target. However, it is known that these hard targets largely ignore the ambiguity of unobserved feedback in RS, and thus may lead to sub-optimal generalization performance. In this work, we propose SoftRec, a new RS optimization framework to enhance item recommendation. The core idea is that we add additional supervisory signals - well-designed soft targets - for each instance so as to better guide the recommender learning. Meanwhile, we carefully investigate the impacts of specific soft target distributions by instantiating the SoftRec with a series of strategies, including item-based, user-based, and model-based. To verify the effectiveness of SoftRec, we conduct extensive experiments on two public recommendation datasets by using various deep recommendation architectures. The experimental results show that our methods achieve superior performance compared with the standard optimization approaches. Moreover, SoftRec could also exhibit strong performance in cold-start scenarios where user-item interaction has higher sparsity.

The factorization-based models have achieved great success in online advertisements and recommender systems due to the capability of efficiently modeling combinational features. These models encode feature interactions by the vector product between feature embedding. Despite the improvement of generalization, the memory consumption of these models grows significantly, because they usually take hundreds to thousands of large categorical features as input. Several existing works try to reduce the memory footprint by hashing, randomized embedding composition, and dimensionality search, but they suffer from either substantial performance degradation or limited memory compression. To this end, in this paper, we propose an extremely memory-efficient Factorization Machine (xLightFM), where each category embedding is composited with latent vectors selected from codebooks. Based on the characteristics of each categorical feature, we further propose to adapt the codebook size with the neural architecture search techniques for compositing the embedding of each categorical feature. This further pushes the limits of memory compression while incurring negligible degradation or even some improvements in prediction performance. We extensively evaluate the proposed algorithm with two real-world datasets. The results demonstrate that xLightFM can outperform the state-of-the-art lightweight factorization-based methods in terms of both prediction quality and memory footprint, and achieve more than 18x and 27x memory compression compared to the vanilla FM on these two datasets, respectively.

Representation learning of examination papers is the cornerstone of the Examination Paper Analysis (EPA) in education area including Paper Difficulty Prediction (PDR) and Finding Similar Papers (FSP). Previous works mainly focus on the representation learning of each test item, but few works notice the hierarchical document structure in examination papers. To this end, in this paper, we propose a novel Examination Organization Encoder (EOE) to learn a robust representation of the examination paper with the hierarchical document structure. Specifically, we first propose a syntax parser to recover the hierarchical document structure and convert an examination paper to an Examination Organization Tree (EOT), where the test items are the leaf nodes and the internal nodes are summarization of their child nodes. Then, we applied a two-layer GRU-based module to obtain the representation of each leaf node. After that, we design a subtree encoder module to aggregate the representation of each leaf node, which is used to calculate an embedding for each layer in the EOT. Finally, we feed all the layer embedding into an output module, the process is over and we get the examination paper representation that can be used for downstream tasks. Extensive experiments on real-world data demonstrate the effectiveness and interpretability of our method.

Recent studies have shown that recommender systems are vulnerable, and it is easy for attackers to inject well-designed malicious profiles into the system, leading to biased recommendations. We cannot deny these data's rationality, making it imperative to establish a robust recommender system. Adversarial training has been extensively studied for robust recommendations. However, traditional adversarial training adds small perturbations to the parameters (inputs), which do not comply with the poisoning mechanism in the recommender system. Thus for the practical models that are very good at learning existing data, it does not perform well. To address the above limitations, we propose adversarial poisoning training (APT). It simulates the poisoning process by injecting fake users (ERM users) who are dedicated to minimizing empirical risk to build a robust system. Besides, to generate ERM users, we explore an approximation approach to estimate each fake user's influence on the empirical risk. Although the strategy of "fighting fire with fire" seems counterintuitive, we theoretically prove that the proposed APT can boost the upper bound of poisoning robustness. Also, we deliver the first theoretical proof that adversarial training holds a positive effect on enhancing recommendation robustness. Through extensive experiments with five poisoning attacks on four real-world datasets, the results show that the robustness improvement of APT significantly outperforms baselines. It is worth mentioning that APT also improves model generalization in most cases.

Knowledge tracing is a fundamental task in intelligent education for tracking the knowledge states of students on necessary concepts. In recent years, Deep Knowledge Tracing (DKT) utilizes recurrent neural networks to model student learning sequences. This approach has achieved significant success and has been widely used in many educational applications. However, in practical scenarios, it tends to suffer from the following critical problems due to data isolation: 1) Data scarcity. Educational data, which is usually distributed across different silos (e.g., schools), is difficult to gather. 2) Different data quality. Students in different silos have different learning schedules, which results in unbalanced learning records, meaning that it is necessary to evaluate the learning data quality independently for different silos. 3) Data incomparability. It is difficult to compare the knowledge states of students with different learning processes from different silos. Inspired by federated learning, in this paper, we propose a novel Federated Deep Knowledge Tracing (FDKT) framework to collectively train high-quality DKT models for multiple silos. In this framework, each client takes charge of training a distributed DKT model and evaluating data quality by leveraging its own local data, while a center server is responsible for aggregating models and updating the parameters for all the clients. In particular, in the client part, we evaluate data quality incorporating different education measurement theories, and we construct two quality-oriented implementations based on FDKT, i.e., FDKTCTT and FDKTIRT-where the means of data quality evaluation follow Classical Test Theory and Item Response Theory, respectively. Moreover, in the server part, we adopt hierarchical model interpolation to uptake local effects for model personalization. Extensive experiments on real-world datasets demonstrate the effectiveness and superiority of the FDKT framework.

In the Click-Through Rate (CTR) prediction scenario, user's sequential behaviors are well utilized to capture the user interest in the recent literature. However, despite being extensively studied, these sequential methods still suffer from three limitations. First, existing methods mostly utilize attention on the behavior of users, which is not always suitable for CTR prediction, because users often click on new products that are irrelevant to any historical behaviors. Second, in the real scenario, there are numerous users that have operations a long time ago, but turn relatively inactive in recent times. Thus, it is hard to precisely capture user's current preferences through early behaviors. Third, multiple representations of user's historical behaviors in different feature subspaces are largely ignored. To remedy these issues, we propose a Multi-Interactive Attention Network (MIAN) to comprehensively extract the latent relationship among all kinds of fine-grained features (e.g., gender, age and occupation in user-profile). Specifically, MIAN contains a Multi-Interactive Layer (MIL) that integrates three local interaction modules to capture multiple representations of user preference through sequential behaviors and simultaneously utilize the fine-grained user-specific as well as context information. In addition, we design a Global Interaction Module (GIM) to learn the high-order interactions and balance the different impacts of multiple features. Finally, Offline experiment results from three datasets, together with an Online A/B test in a large-scale recommendation system, demonstrate the effectiveness of our proposed approach.

User modeling aims to capture the latent characteristics of users from their behaviors, and is widely applied in numerous applications. Usually, centralized user modeling suffers from the risk of privacy leakage. Instead, federated user modeling expects to provide a secure multi-client collaboration for user modeling through federated learning. Existing federated learning methods are mainly designed for consistent clients, which cannot be directly applied to practical scenarios, where different clients usually store inconsistent user data. Therefore, it is a crucial demand to design an appropriate federated solution that can better adapt to user modeling tasks, and however, meets following critical challenges: 1) Statistical heterogeneity. The distributions of user data in different clients are not always independently identically distributed which leads to personalized clients; 2) Privacy heterogeneity. User data contains both public and private information, which have different levels of privacy. It means we should balance different information to be shared and protected; 3) Model heterogeneity. The local user models trained with client records are heterogeneous which need flexible aggregation in the server. In this paper, we propose a novel client-server architecture framework, namely Hierarchical Personalized Federated Learning (HPFL) to serve federated learning in user modeling with inconsistent clients. In the framework, we first define hierarchical information to finely partition the data with privacy heterogeneity. On this basis, the client trains a user model which contains different components designed for hierarchical information. Moreover, client processes a fine-grained personalized update strategy to update personalized user model for statistical heterogeneity. Correspondingly, the server completes a differentiated component aggregation strategy to flexibly aggregate heterogeneous user models in the case of privacy and model heterogeneity. Finally, we conduct extensive experiments on real-world datasets, which demonstrate the effectiveness of the HPFL framework.

Recent years have witnessed the rapid accumulation of massive electronic medical records (EMRs), which highly support the intelligent medical services such as drug recommendation. However, prior arts mainly follow the traditional recommendation strategies like collaborative filtering, which usually treat individual drugs as mutually independent, while the latent interactions among drugs, e.g., synergistic or antagonistic effect, have been largely ignored. To that end, in this paper, we target at developing a new paradigm for drug package recommendation with considering the interaction effect within drugs, in which the interaction effects could be affected by patient conditions. Specifically, we first design a pre-training method based on neural collaborative filtering to get the initial embedding of patients and drugs. Then, the drug interaction graph will be initialized based on medical records and domain knowledge. Along this line, we propose a new Drug Package Recommendation (DPR) framework with two variants, respectively DPR on Weighted Graph (DPR-WG) and DPR on Attributed Graph (DPR-AG) to solve the problem, in which each the interactions will be described as signed weights or attribute vectors. In detail, a mask layer is utilized to capture the impact of patient condition, and graph neural networks (GNNs) are leveraged for the final graph induction task to embed the package. Extensive experiments on a real-world data set from a first-rate hospital demonstrate the effectiveness of our DPR framework compared with several competitive baseline methods, and further support the heuristic study for the drug package generation task with adequate performance.

Considering that the attacks against the industrial control system are mostly organized and premeditated actions, IP traceback is significant for the security of the industrial control system. Based on the infrastructure of the internet, we have developed a novel malicious IP traceback model, ICSTrace, without deploying any new services. The model extracts the function codes and their parameters from the attack data according to the format of the industrial control protocol and employs a short sequence probability method to transform the function codes and their parameters into a vector, which characterizes the attack pattern of malicious IP addresses. Furthermore, a partial seeded K-means algorithm is proposed for the pattern’s clustering, which helps in tracing the attacks back to an organization. ICSTrace is evaluated based on the attack data captured by the large-scale deployed honeypots for the industrial control system, and the results demonstrate that ICSTrace is effective on malicious IP traceback in the industrial control system.

Large-scale pre-trained language models (e.g., BERT) have attracted great attention in recent years. It is straightforward to fine-tune them on natural language understanding tasks such as text classification, however, effectively and efficiently incorporating them into natural language generation tasks such as neural machine translation remains a challenging problem. In this paper, we integrate two pre-trained BERT models from the source and target language domains into a sequence-to-sequence model by introducing light-weight adapter modules. The adapters are inserted between BERT layers and tuned on downstream tasks, while the parameters of BERT models are fixed during fine-tuning. As pre-trained language models are usually very deep, inserting adapters into all layers will result in a considerable scale of new parameters. To deal with this problem, we introduce latent variables to decide whether using adapters or not in each layer, which are learned during fine-tuning. In this way, the model is able to automatically determine which adapters to use, therefore hugely promoting the parameter efficiency and decoding speed. We evaluate the proposed framework on various neural machine translation tasks. Equipped with parallel sequence decoding, our model consistently outperforms autoregressive baselines while reducing the inference latency by half. With automatic adapter selection, the proposed model further achieves 20% speedup while still outperforming autoregressive baselines. When applied to autoregressive decoding, the proposed model can also achieve comparable performance with the state-of-the-art baseline models.

Most existing image denoising methods focus on improving denoising quality. However, when applying denoising methods to practical tasks, in addition to the denoising quality, the number of parameters, and the computational complexity should be fully considered. In this article, we propose a demand-oriented framework (DOF) for image denoising, which can give preference to the number of parameters, the computational complexity, and the denoising quality or balance these three performance metrics. To perform the demand-oriented denoising, we first design a scale encoder to help the denoising model extract fewer but more representative features. Then, the split-flow module is introduced to fully exploit the input features by sharing the information of one network branch with other network branches. Finally, the scale decoder is utilized to reconstruct the final noise map without using any parameters. Through extensive experiments, we demonstrate that the proposed framework can be applied to several existing methods to help them achieve a more competitive denoising performance in terms of the number of parameters, and computational complexity.

To train accurate deep object detectors under the extreme foreground-background imbalance, heuristic sampling methods are always necessary, which either re-sample a subset of all training samples (hard sampling methods, e.g. biased sampling, OHEM), or use all training samples but re-weight them discriminatively (soft sampling methods, e.g. Focal Loss, GHM). In this paper, we challenge the necessity of such hard/soft sampling methods for training accurate deep object detectors. While previous studies have shown that training detectors without heuristic sampling methods would significantly degrade accuracy, we reveal that this degradation comes from an unreasonable classification gradient magnitude caused by the imbalance, rather than a lack of re-sampling/re-weighting. Motivated by our discovery, we propose a simple yet effective Sampling-Free mechanism to achieve a reasonable classification gradient magnitude by initialization and loss scaling. Unlike heuristic sampling methods with multiple hyperparameters, our Sampling-Free mechanism is fully data diagnostic, without laborious hyperparameters searching. We verify the effectiveness of our method in training anchor-based and anchor-free object detectors, where our method always achieves higher detection accuracy than heuristic sampling methods on COCO and PASCAL VOC datasets. Our Sampling-Free mechanism provides a new perspective to address the foreground-background imbalance. Our code is released at https://github.com/ChenJoya/sampling-free.

Mobility prediction plays an important role in a wide range of location-based applications and services. However, there are three problems in the existing literature: (1) explicit high-order interactions of spatio-temporal features are not systemically modeled; (2) most existing algorithms place attention mechanisms on top of recurrent network, so they can not allow for full parallelism and are inferior to self-attention for capturing long-range dependence; (3) most literature does not make good use of long-term historical information and do not effectively model the long-term periodicity of users. To this end, we propose MoveNet and RLMoveNet. MoveNet is a self-attention-based sequential model, predicting each user’s next destination based on her most recent visits and historical trajectory. MoveNet first introduces a cross-based learning framework for modeling feature interactions. With self-attention on both the most recent visits and historical trajectory, MoveNet can use an attention mechanism to capture the user’s long-term regularity in a more efficient way. Based on MoveNet, to model long-term periodicity more effectively, we add the reinforcement learning layer and named RLMoveNet. RLMoveNet regards the human mobility prediction as a reinforcement learning problem, using the reinforcement learning layer as the regularization part to drive the model to pay attention to the behavior with periodic actions, which can help us make the algorithm more effective. We evaluate both of them with three real-world mobility datasets. MoveNet outperforms the state-of-the-art mobility predictor by around 10% in terms of accuracy, and simultaneously achieves faster convergence and over 4x training speedup. Moreover, RLMoveNet achieves higher prediction accuracy than MoveNet, which proves that modeling periodicity explicitly from the perspective of reinforcement learning is more effective.

Binary representation of users and items can dramatically improve efficiency of recommendation and reduce size of recommendation models. However, learning optimal binary codes for them is challenging due to binary constraints, even if squared loss is optimized. In this article, we propose a general framework for discrete matrix factorization based on discrete optimization, which can 1) optimize multiple loss functions; 2) handle both explicit and implicit feedback datasets; and 3) take auxiliary information into account without any hyperparameters. To tackle the challenging discrete optimization problem, we propose block coordinate descent based on semidefinite relaxation of binary quadratic programming. We theoretically show that it is equivalent to discrete coordinate descent when only one coordinate is in each block. We extensively evaluate the proposed algorithms on eight real-world datasets. The results of evaluation show that they outperform the state-of-the-art baselines significantly and that auxiliary information of items improves recommendation performance. For better showing the advantages of binary representation, we further propose a two-stage recommender system, consisting of an item-recalling stage and a subsequent fine-ranking stage. Its extensive evaluation shows hashing can dramatically accelerate item recommendation with little degradation of accuracy.

Because of strict response-time constraints, efficiency of top-k recommendation is crucial for real-world recommender systems. Locality sensitive hashing and index-based methods usually store both index data and item feature vectors in main memory, so they handle a limited number of items. Hashing-based recommendation methods enjoy low memory cost and fast retrieval of items, but suffer from large accuracy degradation. In this paper, we propose product Quantized Collaborative Filtering (pQCF) for better trade-off between efficiency and accuracy. pQCF decomposes a joint latent space of users and items into a Cartesian product of low-dimensional subspaces, and learns clustered representation within each subspace. A latent factor is then represented by a short code, which is composed of subspace cluster indexes. A user's preference for an item can be efficiently calculated via table lookup. We then develop block coordinate descent for efficient optimization and reveal the learning of latent factors is seamlessly integrated with quantization. We further investigate an asymmetric pQCF, dubbed as QCF, where user latent factors are not quantized and shared across different subspaces. The extensive experiments with 6 real-world datasets show that pQCF significantly outperforms the state-of-the-art hashing-based CF and QCF increases recommendation accuracy compared to pQCF.

For offering proactive services (e.g., personalized exercise recommendation) to the students in computer supported intelligent education, one of the fundamental tasks is predicting student performance (e.g., scores) on future exercises, where it is necessary to track the change of each student's knowledge acquisition during her exercising activities. Unfortunately, to the best of our knowledge, existing approaches can only exploit the exercising records of students, and the problem of extracting rich information existed in the materials (e.g., knowledge concepts, exercise content) of exercises to achieve both more precise prediction of student performance and more interpretable analysis of knowledge acquisition remains underexplored. To this end, in this paper, we present a holistic study of student performance prediction. To directly achieve the primary goal of performance prediction, we first propose a general Exercise-Enhanced Recurrent Neural Network (EERNN) framework by exploring both student's exercising records and the text content of corresponding exercises. In EERNN, we simply summarize each student's state into an integrated vector and trace it with a recurrent neural network, where we design a bidirectional LSTM to learn the encoding of each exercise from its content. For making final predictions, we design two implementations on the basis of EERNN with different prediction strategies, i.e., EERNNM with Markov property and EERNNA with Attention mechanism. Then, to explicitly track student's knowledge acquisition on multiple knowledge concepts, we extend EERNN to an explainable Exercise-aware Knowledge Tracing (EKT) framework by incorporating the knowledge concept information, where the student's integrated state vector is now extended to a knowledge state matrix. In EKT, we further develop a memory network for quantifying how much each exercise can affect the mastery of students on multiple knowledge concepts during the exercising process. Finally, we conduct extensive experiments and evaluate both EERNN and EKT frameworks on a large-scale real-world data. The results in both general and cold-start scenarios clearly demonstrate the effectiveness of two frameworks in student performance prediction as well as the superior interpretability of EKT.

Online search engine has been widely regarded as the most convenient approach for information acquisition. Indeed, the intensive information-seeking behaviors of search engine users make it possible to exploit search engine queries as effective “crowd sensors” for event monitoring. While some researchers have investigated the feasibility of using search engine queries for coarse-grained event analysis, the capability of search engine queries for real-time event detection has been largely neglected. To this end, in this article, we introduce a large-scale and systematic study on exploiting real-time search engine queries for outbreak event detection, with a focus on earthquake rapid reporting. In particular, we propose a realistic system of real-time earthquake detection through monitoring millions of queries related to earthquakes from a dominant online search engine in China. Specifically, we first investigate a large set of queries for selecting the representative queries that are highly correlated with the outbreak of earthquakes. Then, based on the real-time streams of selected queries, we design a novel machine learning–enhanced two-stage burst detection approach for detecting earthquake events. Meanwhile, the location of an earthquake epicenter can be accurately estimated based on the spatial-temporal distribution of search engine queries. Finally, through the extensive comparison with earthquake catalogs from China Earthquake Networks Center, 2015, the detection precision of our system can achieve 87.9%, and the accuracy of location estimation (province level) is 95.7%. In particular, 50% of successfully detected results can be found within 62 s after earthquake, and 50% of successful locations can be found within 25.5 km of seismic epicenter. Our system also found more than 23.3% extra earthquakes that were felt by people but not publicly released, 12.1% earthquake-like special outbreaks, and meanwhile, revealed many interesting findings, such as the typical query patterns of earthquake rumor and regular memorial events. Based on these results, our system can timely feed back information to the search engine users according to various cases and accelerate the information release of felt earthquakes.

As a crucial task for video analysis, social relation recognition for characters not only provides semantically rich description of video content but also supports intelligent applications, e.g., video retrieval and visual question answering. Unfortunately, due to the semantic gap between visual and semantic features, traditional solutions may fail to reveal the accurate relations among characters. At the same time, the development of social media platforms has now promoted the emergence of crowdsourced comments, which may enhance the recognition task with semantic and descriptive cues. To that end, in this article, we propose a novel multimodal-based solution to deal with the character relation recognition task. Specifically, we capture the target character pairs via a search module and then design a multistream architecture for jointly embedding the visual and textual information, in which feature fusion and attention mechanism are adapted for better integrating the multimodal inputs. Finally, supervised learning is applied to classify character relations. Experiments on real-world data sets validate that our solution outperforms several competitive baselines.

In the area of community question answering (CQA), answer selection and answer ranking are two tasks which are applied to help users quickly access valuable answers. Existing solutions mainly exploit the syntactic or semantic correlation between a question and its related answers (Q&A), where the multifacet domain effects in CQA are still underexplored. In this paper, we propose a unified model, enhanced attentive recurrent neural network (EARNN), for both answer selection and answer ranking tasks by taking full advantages of both Q&A semantics and multifacet domain effects (i.e., topic effects and timeliness). Specifically, we develop a serialized long short-term memory to learn the unified representations of Q&A, where two attention mechanisms at either sentence level or word level are designed for capturing the deep effects of topics. Meanwhile, the emphasis of Q&A can be automatically distinguished. Furthermore, we design a time-sensitive ranking function to model the timeliness in CQA. To effectively trainEARNN, a question-dependent pairwise learning strategy is also developed. Finally, we conduct extensive experiments on a real-world dataset from Quora. Experimental results validate the effectiveness and interpretability of our proposedEARNNmodel.

Natural language inference (NLI) task requires an agent to determine the semantic relation between a premise sentence ( p) and a hypothesis sentence ( h), which demands sufficient understanding about sentences semantic. Due to the issues, such as polysemy, ambiguity, as well as fuzziness of sentences, intense sentence understanding is very challenging. To this end, in this article, we introduce the corresponding image of sentences as reference information for enhancing sentence semantic understanding and representing. Specifically, we first propose an image-enhanced multilevel sentence representation net (IEMLRN), that utilizes the image features from pretrained models for enhancing the sentence semantic understanding at different scales, i.e., lexical-level, phrase-level, and sentence-level. The proposed model advances the performance on NLI tasks by leveraging the pretrained global features of images. However, as these pretrained image features are optimized on specific image classification datasets, they may not show the best performance on NLI tasks. Therefore, we further propose to design an adaptive image feature generator that extracts fine-grained image labels from the corresponding sentences. After that, we extend the IEMLRN to multilevel image-enhanced sentence representation net (MIESR) by utilizing not only the coarse-grained pretrained features of images, but also the fine-grained adaptive features of images. Therefore, sentence semantic can be evaluated and enhanced more comprehensively and precisely. Extensive experiments on two benchmark datasets (SNLI, SICK) clearly show our proposed IEMLRN significantly outperform the state-of-the-art baselines, and our proposed MIESR model achieves the best performance by considering not only the text but also images in an adaptive multigranularities way.

Non-autoregressive translation (NAT) models remove the dependence on previous target tokens and generate all target tokens in parallel, resulting in significant inference speedup but at the cost of inferior translation accuracy compared to autoregressive translation (AT) models. Considering that AT models have higher accuracy and are easier to train than NAT models, and both of them share the same model configurations, a natural idea to improve the accuracy of NAT models is to transfer a well-trained AT model to an NAT model through fine-tuning. However, since AT and NAT models differ greatly in training strategy, straightforward fine-tuning does not work well. In this work, we introduce curriculum learning into fine-tuning for NAT. Specifically, we design a curriculum in the fine-tuning process to progressively switch the training from autoregressive generation to non-autoregressive generation. Experiments on four benchmark translation datasets show that the proposed method achieves good improvement (more than 1 BLEU score) over previous NAT baselines in terms of translation accuracy, and greatly speed up (more than 10 times) the inference process over AT baselines.

Many events occur in real-world and social networks. Events are related to the past and there are patterns in the evolution of event sequences. Understanding the patterns can help us better predict the type and arriving time of the next event. In the literature, both feature-based approaches and generative approaches are utilized to model the event sequence. Feature-based approaches extract a variety of features, and train a regression or classification model to make a prediction. Yet, their performance is dependent on the experience-based feature exaction. Generative approaches usually assume the evolution of events follow a stochastic point process (e.g., Poisson process or its complexer variants). However, the true distribution of events is never known and the performance depends on the design of stochastic process in practice. To solve the above challenges, in this paper, we present a novel probabilistic generative model for event sequences. The model is termed Variational Event Point Process (VEPP). Our model introduces variational auto-encoder to event sequence modeling that can better use the latent information and capture the distribution over inter-arrival time and types of event sequences. Experiments on real-world datasets prove effectiveness of our proposed model.

Cognitive diagnosis is a fundamental issue in intelligent education, which aims to discover the proficiency level of students on specific knowledge concepts. Existing approaches usually mine linear interactions of student exercising process by manual-designed function (e.g., logistic function), which is not sufficient for capturing complex relations between students and exercises. In this paper, we propose a general Neural Cognitive Diagnosis (NeuralCD) framework, which incorporates neural networks to learn the complex exercising interactions, for getting both accurate and interpretable diagnosis results. Specifically, we project students and exercises to factor vectors and leverage multi neural layers for modeling their interactions, where the monotonicity assumption is applied to ensure the interpretability of both factors. Furthermore, we propose two implementations of NeuralCD by specializing the required concepts of each exercise, i.e., the NeuralCDM with traditional Q-matrix and the improved NeuralCDM+ exploring the rich text content. Extensive experimental results on real-world datasets show the effectiveness of NeuralCD framework with both accuracy and interpretability.

The recent growth of social networking platforms also led to the emergence of social spammers, who overwhelm legitimate users with unwanted content. The existing social spammer detection methods can be characterized into two categories: features based ones and propagation-based ones. Features based methods mainly rely on matrix factorization using tweet text features, and regularization using social graphs is incorporated. However, these methods are fully supervised and can only utilize labeled part of social graphs, which fail to work in a real-world semi-supervised setting. The propagation-based methods primarily employ Markov Random Fields (MRFs) to capture human intuitions in user following relations, which cannot take advantages of rich text features. In this paper, we propose a novel social spammer detection model based on Graph Convolutional Networks (GCNs) that operate on directed social graphs by explicitly considering three types of neighbors. Furthermore, inspired by the propagation-based methods, we propose a MRF layer with refining effects to encapsulate these human insights in social relations, which can be formulated as a RNN through mean-field approximate inference, and stack on top of GCN layers to enable end-to-end training. We evaluate our proposed method on two real-world social network datasets, and the results demonstrate that our method outperforms the state-of-the-art approaches.

Well begun is half done. In the crowdfunding market, the early fundraising performance of the project is a concerned issue for both creators and platforms. However, estimating the early fundraising performance before the project published is very challenging and still under-explored. To that end, in this paper, we present a focused study on this important problem in a market modeling view. Specifically, we propose a Graph-based Market Environment model (GME) for estimating the early fundraising performance of the target project by exploiting the market environment. In addition, we discriminatively model the market competition and market evolution by designing two graph-based neural network architectures and incorporating them into the joint optimization stage. Finally, we conduct extensive experiments on the real-world crowdfunding data collected from Indiegogo.com. The experimental results clearly demonstrate the effectiveness of our proposed model for modeling and estimating the early fundraising performance of the target project.