Vision transformers (ViTs) are widely employed in multimodal large language models (MLLMs) for visual encoding. However, they exhibit inferior performance on tasks regarding fine-grained visual perception. We attribute this to the inner limitations of ViTs in capturing diverse visual semantic levels. To address this, we present Hierarchical window (Hiwin) transformer as a plug-and-play solution for MLLMs, centered around our inverse semantic pyramid (ISP). Hiwin transformer comprises two key modules: (i) a visual detail injection module, which progressively injects low-level visual details into high-level language-aligned semantics features, thereby constructing an ISP, and (ii) a hierarchical window attention module, which leverages cross-scale windows to condense multi-level semantics from the ISP. Notably, our design achieves an average boost of 3.7% across 14 benchmarks compared with the baseline method, 9.3% on DocVQA for instance.

Vision-Language Models (VLMs), exemplified by CLIP, have emerged as foundational for multimodal intelligence. However, their capacity for logical understanding remains significantly underexplored, resulting in critical **''logical blindspots''** that limit their reliability in practical applications. To systematically diagnose this, we introduce **LogicBench**, a comprehensive benchmark with over 50,000 vision-language pairs across 9 logical categories and 4 diverse scenarios: images, videos, anomaly detection, and medical diagnostics. Our evaluation reveals that existing VLMs, even the state-of-the-art ones, fall at over 40 accuracy points below human performance, particularly in challenging tasks like Causality and Conditionality, highlighting their reliance on surface semantics over critical logical structures. To bridge this gap, we propose **LogicCLIP**, a novel training framework designed to boost VLMs' logical sensitivity through advancements in both data generation and optimization objectives. LogicCLIP utilizes logic-aware data generation and a contrastive learning strategy that combines coarse-grained alignment, a fine-grained multiple-choice objective, and a novel logical structure-aware objective. Extensive experiments demonstrate LogicCLIP's substantial improvements in logical comprehension across all LogicBench domains, significantly outperforming baselines. Moreover, LogicCLIP retains, and often surpasses, competitive performance on general vision-language benchmarks, demonstrating that the enhanced logical understanding does not come at the expense of general alignment. We believe LogicBench and LogicCLIP will be important resources for advancing VLM logical capabilities.

Text-to-Image (T2I) diffusion models have made rapid progress on semantic alignment (generating what is described in the prompt), yet users still lack reliable control over aesthetic composition (how visual elements are put together). Prior work often treats aesthetics as a single, preference-driven notion (e.g., “high quality”, “detailed”, “breathtaking”), which does not map cleanly to compositional intent.

Visual Document Retrieval (VDR) requires representations that capture both fine-grained visual details and global document structure to ensure retrieval efficacy while maintaining computational efficiency. Existing VDR models struggle to balance effectiveness and efficiency when processing high-resolution documents: they often either lose fine-grained information or generate an excessive number of visual tokens, resulting in significant indexing overhead and high retrieval latency. In this work, we rethink the visual encoding mechanism and propose a new X-VisEmb paradigm that progresses from multi-resolution sampling and encoding, through cross-granularity feature fusion, to adaptive representation distillation. A preliminary study validates its feasibility and effectiveness in capturing complementary visual cues at varying scales. Building on the insights, we develop MURE , a novel framework that employs VLMs as a hierarchical multi-resolution encoder, integrates resolution-level Matryoshka representation learning (RMRL) for effective feature fusion, and applies a semantic-aware hierarchical clustering mechanism for visual token compression. Experiments on two widely used VDR benchmarks show that our MURE framework consistently beats strong baselines. Furthermore, it significantly outperforms ColPali with only 50% of its visual token budget.

Reasoning in Large Language Models (LLMs) has recently shown strong potential in enhancing generative recommendation through deep understanding of complex user preference. Existing approaches follow a reason-then-recommend paradigm, where LLMs perform step-by-step reasoning before item generation. However, this paradigm inevitably suffers from reasoning degradation (i.e., homogeneous or error-accumulated reasoning) due to the lack of intermediate verification, thus undermining the recommendation. To bridge this gap, we propose a novel reason-verify-recommend paradigm, which interleaves reasoning with verification to provide reliable feedback, guiding the reasoning process toward more faithful user preference understanding. To enable effective verification, we establish two key principles for verifier design: 1) reliability ensures accurate evaluation of reasoning correctness and informative guidance generation; and 2) multi-dimensionality emphasizes comprehensive verification across multi-dimensional user preferences. Accordingly, we propose an effective implementation called VRec. It employs a mixture of verifiers to ensure multi-dimensionality, while leveraging a proxy prediction objective to pursue reliability. Experiments on four real-world datasets demonstrate that VRec substantially enhances recommendation effectiveness and scalability without compromising efficiency.

Deep Research (DR) requires LLM agents to autonomously perform multi-step information seeking, processing, and reasoning to generate comprehensive reports. In contrast to existing studies that mainly focus on unstructured web content, a more challenging DR task should additionally utilize structured knowledge to provide a solid data foundation, facilitate quantitative computation, and lead to in-depth analyses. In this paper, we refer to this novel task as Knowledgeable Deep Research (KDR), which requires DR agents to generate reports with both structured and unstructured knowledge. Furthermore, we propose the Hybrid Knowledge Analysis framework (HKA), a multi-agent architecture that reasons over both kinds of knowledge and integrates the texts, figures, and tables into coherent multimodal reports. The key design is the Structured Knowledge Analyzer, which utilizes both coding and vision-language models to produce figures, tables, and corresponding insights. To support systematic evaluation, we construct KDR-Bench, which covers 9 domains, includes 41 expert-level questions, and incorporates a large number of structured knowledge resources (e.g., 1,252 tables). We further annotate the main conclusions and key points for each question and propose three categories of evaluation metrics including general-purpose, knowledge-centric, and vision-enhanced ones. Experimental results demonstrate that HKA consistently outperforms most existing DR agents on general-purpose and knowledge-centric metrics, and even surpasses the Gemini DR agent on vision-enhanced metrics, highlighting its effectiveness in deep, structure-aware knowledge analysis. Finally, we hope this work can serve as a new foundation for structured knowledge analysis in DR agents and facilitate future multimodal DR studies.

Event forecasting is inherently influenced by multifaceted considerations, including international relations, regional historical dynamics, and cultural contexts. However, existing LLM-based approaches employ single-model architectures that generate predictions along a singular explicit trajectory, constraining their ability to capture diverse geopolitical nuances across complex regional contexts. To address this limitation, we introduce ThinkTank-ME, a novel Think Tank framework for Middle East event forecasting that emulates collaborative expert analysis in real-world strategic decision-making. To facilitate expert specialization and rigorous evaluation, we construct POLECAT-FOR-ME, a Middle East–focused event forecasting benchmark. Experimental results demonstrate the superiority of multi-expert collaboration in handling complex temporal geopolitical forecasting tasks. The code is available at https://github.com/LuminosityX/ThinkTank-ME.

Understanding traffic accident scenes is a long-standing research for vision-based safe driving. It seeks to answer why accidents occur, how near-crash scenes develop, and what the key elements of an accident are. This research is challenging due to the scarcity and fragmentation of accident data, as well as the complex accident environments. To study this, we present a framework of Abductive Driving accident Video understanding (ADVersa), which infers a plausible visual and textual explanation for the absent near-crash scenes. ADVersa underscores three groups of tasks: 1) visual past recovery of near-crash scenes, 2) visual prediction of near-crash scenes, and 3) accident cause involved video synthesis. To support the study, we first contribute MM-AU, a novel dataset for Multi-Modal Accident video Understanding. MM-AU contains 11,727 in-the-wild driving accident videos with temporally aligned text descriptions, 2.23 million well-annotated object boxes, and 58,650 pairs of video-based accident cause texts. We then propose an Abductive CLIP model and a Contrastive Graph Video Pre-training (CGVP) model, which exploit relation-aware cross-modal semantic learning to drive spatially abductive and temporally abductive accident video diffusion. Extensive experiments verify the superiority of ADVersa to the state-of-the-art approaches on different tasks, i.e., historical near-crash video frame recovering, crashing video frame prediction, textual accident cause and category reasoning, normal-to-accident video synthesis, and accident video editing. With these efforts, we hope this research can advance the progress on multimodal accident video understanding.

Large Language Models (LLMs) exhibit remarkable proficiency in understanding and managing text-based tasks.Many works try to transfer these capabilities to the video domain, which are referred to as Video-LLMs. However, current Video-LLMs can only grasp the coarse-grained semantics and are unable to efficiently handle tasks involving the comprehension or localization of specific video segments. To address these challenges, we propose Momentor, a Video-LLM designed to perform fine-grained temporal understanding tasks. To facilitate the training of Momentor, we develop an automatic data generation engine to build Moment-10M, a large-scale video instruction dataset with segment-level instruction data. Building upon the foundation of the previously published Momentor and the Moment-10M dataset, we further extend this work by introducing a Spatio-Temporal Token Consolidation (STTC) method, which can merge redundant visual tokens spatio-temporally in a parameter-free manner, thereby significantly promoting computational efficiency while preserving fine-grained visual details. We integrate STTC with Momentor to develop Momentor++ and validate its performance on various benchmarks. Momentor demonstrates robust capabilities in fine-grained temporal understanding and localization. Further, Momentor++ excels in efficiently processing and analyzing extended videos with complex events, showcasing marked advancements in handling extensive temporal contexts.

Multimodal representation learning seeks to create a unified representation space by integrating diverse data modalities to improve multimodal understanding. Traditional methods often depend on pairwise contrastive learning, which relies on a predefined anchor modality, restricting alignment across all modalities. Recent advances have investigated the simultaneous alignment of multiple modalities, yet several challenges remain, such as limitations imposed by fixed anchor points and instability arising from optimizing the product of singular values. To address the challenges, in this paper, we propose Principled Multimodal Representation Learning (PMRL), a novel framework that achieves simultaneous alignment of multiple modalities without anchor dependency in a more stable manner. Specifically, grounded in the theoretical insight that full alignment corresponds to a rank-1 Gram matrix, PMRL optimizes the dominant singular value of the representation matrix to align modalities along a shared leading direction. We propose a softmax-based loss function that treats singular values as logits to prioritize the largest singular value. Besides, instance-wise contrastive regularization on the leading eigenvectors maintains inter-instance separability and prevents representation collapse. Extensive experiments across diverse tasks demonstrate PMRL's superiority compared to baseline methods.

Recent advances in "track-anything" models have significantly improved fine-grained video understanding by simultaneously handling multiple video segmentation and tracking tasks. However, existing models often struggle with robust and efficient temporal propagation. To address these challenges, we propose the Sparse Spatio-Temporal Propagation (SSTP) method, which achieves robust and efficient unified video segmentation by selectively leveraging key spatio-temporal features in videos. Specifically, we design a dynamic 3D spatio-temporal convolution to aggregate global multi-frame spatio-temporal information into memory frames during memory construction. Additionally, we introduce a spatio-temporal aggregation reading strategy to efficiently aggregate the relevant spatio-temporal features from multiple memory frames during memory retrieval. By combining SSTP with an image segmentation foundation model, such as the segment anything model, our method effectively addresses multiple data-scarce video segmentation tasks. Our experimental results demonstrate state-of-the-art performance on five video segmentation tasks across eleven datasets, outperforming both task-specific and unified methods. Notably, SSTP exhibits strong robustness in handling sparse, low-frame-rate videos, making it well-suited for real-world applications.

Video object segmentation (VOS) is a fundamental task in video analysis, aiming to accurately recognize and segment objects of interest within video sequences. Conventional methods, relying on memory networks to store single-frame appearance features, face challenges in computational efficiency and capturing dynamic visual information effectively. To address these limitations, we present a Video Decoupling Network (VDN) with a per-clip memory updating mechanism. Our approach is inspired by the dual-stream hypothesis of the human visual cortex and decomposes multiple previous video frames into fundamental elements: scene, motion, and instance. We propose the Unified Prior-based Spatio-temporal Decoupler (UPSD) algorithm, which parses multiple frames into basic elements in a unified manner. UPSD continuously stores elements over time, enabling adaptive integration of different cues based on task requirements. This decomposition mechanism facilitates comprehensive spatial-temporal information capture and rapid updating, leading to notable enhancements in overall VOS performance. Extensive experiments conducted on multiple VOS benchmarks validate the state-of-the-art accuracy, efficiency, generalizability, and robustness of our approach. Remarkably, VDN demonstrates a significant performance improvement and a substantial speed-up compared to previous state-of-the-art methods on multiple VOS benchmarks. It also exhibits excellent generalizability under domain shift and robustness against various noise types.

Identifying harmful memes is challenging due to their implicit meanings, which are not always evident from texts and images alone. Existing solutions often lack clear explanations to justify their decisions. To address this gap, we propose an explainable approach,ExplainHM++, which detects harmful memes by reasoning over competing rationales from both harmful and harmless perspectives. First, inspired by the capabilities of Large Multimodal Models (LMMs) in text generation and multimodal reasoning, we developExplainHM, a one-stage multimodal debate in which LMMs generate explanations through contradictory arguments. Second, we fine-tune a small language model to serve as a judge in the debate, improving the integration of harmfulness rationales with the multimodal content of memes. However, we observe that a naive multimodal debate remains vulnerable, as it heavily depends on the inherent reasoning ability of LMMs to understand the memes. Given the evolving and noisy nature of memes, we further introduce a meme sample retrieval mechanism and a retrieval-augmented debate paradigm to strengthen and refine LMM-generated explanations. Extensive experiments on three public meme datasets demonstrate thatExplainHM++not only outperforms state-of-the-art methods but also provides superior, interpretable explanations for harmful meme detection.

Sequential recommendation has become a hot topic, which aims to predict the desired items for each user based on his/her historical actions. The mainstream advancements for this task focus on modeling user behaviors in a pure item ID-based manner, which often fail to provide satisfactory results due to data sparsity and cold-start issues. Recently, several studies that leverage multi-modal information (i.e., multi-modal sequential recommendation) have shed light on alleviating such issues. However, we argue that three limitations are still not well addressed: 1) they usually extract ID modality features of an item with an one-hot encoding, which does not include any semantic information; 2) they fail to effectively mitigate the semantic gap issue and explicitly explore the asynchronous interplay between any two modalities; and 3) during the model prediction stage, they neglect the significance of adaptively fusing multi-modal embeddings for each user. To address such defects, we propose a novel framework for multi-modal sequential recommendation, namely, a multi-way cascade-attention network (MCN). Specifically, we apply a lightweight graph propagation network to derive informative representations of the ID-oriented modality, explicitly encoding collaborative signals in the user-item interaction graph. Next, we develop a multi-way cascade-attention module (CAM) to accomplish user behavior sequence alignment across different modality spaces. Each CAM consists of a cross-attention block followed by a series of self-attention blocks. The former encodes the asynchronous interplay between two modalities, while the latter captures intra-modal temporal dependencies. Finally, we design a modality-aware attentive strategy to dynamically fuse the user's dynamic interests across different modality spaces. Our extensive experiments on four public datasets demonstrate the superiority of MCN over recent state-of-the-art recommenders.

Existing efforts in text-based video question answering (TextVideoQA) are criticized for their opaque decision-making and heavy reliance on scene-text recognition. In this paper, we propose to studyGrounded TextVideoQAby forcing models to answer questions and spatio-temporally localize the relevant scene-text regions, thus decoupling QA from scene-text recognition and promoting research towards interpretable QA. The task has three-fold significance. First, it encourages scene-text evidence versus other short-cuts for answer predictions. Second, it directly accepts scene-text regions as visual answers, thus circumventing the problem of ineffective answer evaluation by stringent string matching. Third, it isolates the challenges inherited in VideoQA and scene-text recognition. This enables the diagnosis of the root causes for failure predictions,e.g., wrong QA or wrong scene-text recognition? To achieve Grounded TextVideoQA, we propose the T2S-QA model that highlights a disentangled temporal-to-spatial contrastive learning strategy for weakly-supervised scene-text grounding and grounded TextVideoQA. To facilitate evaluation, we construct a new datasetViTXT-GQAwhich features 52Kscene-text bounding boxes within 2.2Ktemporal segments related to 2Kquestions and 729 videos. With ViTXT-GQA, we perform extensive experiments and demonstrate the severe limitations of existing techniques in Grounded TextVideoQA. While T2S-QA achieves superior results, the large performance gap with human leaves ample space for improvement. Our further analysis of oracle scene-text inputs posits that the major challenge is scene-text recognition. To advance the research of Grounded TextVideoQA, our dataset and code are athttps://github.com/zhousheng97/ViTXT-GQA.git

Recent efforts have aimed to improve AI models in legal case matching by integrating legal domain knowledge. However, successful legal case matching requires the tacit knowledge of legal practitioners, which is difficult to verbalize and encode into models. This emphasizes the crucial role of involving legal practitioners in high-stakes legal case matching. To address this, we propose a collaborative matching framework called Co-Matching , which encourages both the model and the legal practitioner to participate in the matching process, integrating tacit knowledge. Unlike existing methods that rely solely on the model, Co-Matching allows both the legal practitioner and the model to determine key sentences and then combine them probabilistically. Co-Matching introduces a method called ProtoEM to estimate human decision uncertainty, facilitating the probabilistic combination. Experimental results demonstrate that Co-Matching consistently outperforms existing legal case matching methods, delivering significant performance improvements over human- and model-based matching in isolation (on average, +5.51% and +8.71%, respectively). Further analysis shows that Co-Matching also ensures better human–model collaboration effectiveness. Our study represents an effort in human–model collaboration for the legal case matching task, marking a milestone for future collaborative matching studies.

The challenge of text-to-3D generation lies in accurately and efficiently crafting 3D objects based on natural language descriptions, a capability that promises a substantial reduction in manual design efforts and offers an intuitive interface for user interaction with digital environments. Despite recent advancements, effective recovery of fine-grained details and efficient optimization of high-resolution 3D outputs remain critical hurdles. Drawing inspiration from the efficacious paradigm of progressive learning, we present a novel Multi-Scale Triplane Network (MTN) architecture coupled with a tailored progressive learning strategy. As the name implies, the MTN consists of four triplanes transitioning from low to high resolution. This hierarchical structure allows the low-resolution triplane to serve as an initial shape for the high-resolution counterparts, easing the inherent complexity of the optimization process. Furthermore, we introduce the progressive learning scheme that systematically guides the network to shift its attention from prominent coarse-grained structures to intricate fine-grained patterns. This strategic progression ensures that the focus of the model evolves towards emulating the subtlest aspects of the described 3D object. Our experiment verifies that the proposed method performs favorably against contemporary methods. Even for the complex and nuanced textual descriptions, our method consistently excels, delivering robust and viable 3D shapes where other methods falter.

Large Language Models (LLMs) have impressive capabilities in text understanding and zero-shot reasoning. However, delays in knowledge updates may cause them to reason incorrectly or produce harmful results. Knowledge Graphs (KGs) provide rich and reliable contextual information for the reasoning process of LLMs by structurally organizing and connecting a wide range of entities and relations. Existing KG-based LLM reasoning methods only inject KGs' knowledge into prompts in a textual form, ignoring its structural information. Moreover, they mostly rely on close-source models or open-source models with large parameters, which poses challenges to high resource consumption. To address this, we propose a novel Lightweight and efficient Prompt learning-ReasOning Framework for KGQA (LightPROF), which leverages the full potential of LLMs to tackle complex reasoning tasks in a parameter-efficient manner. Specifically, LightPROF follows a “Retrieve-Embed-Reason” process, first accurately, and stably retrieving the corresponding reasoning graph from the KG through retrieval module. Next, through a Transformer-based Knowledge Adapter, it finely extracts and integrates factual and structural information from the KG, then maps this information to the LLM’s token embedding space, creating an LLM-friendly prompt to be used by the LLM for the final reasoning. Additionally, LightPROF only requires training Knowledge Adapter and can be compatible with any open-source LLM. Extensive experiments on two public KGQA benchmarks demonstrate that LightPROF achieves superior performance with small-scale LLMs. Furthermore, LightPROF shows significant advantages in terms of input token count and reasoning time.

Zero-shot learning (ZSL) endeavors to transfer knowledge from the seen categories to recognize unseen categories, which mostly relies on the semantic-visual interactions between image and attribute tokens. Recently, the prompt learning has emerged in ZSL and demonstrated significant potential as it allows the zero-shot transfer of diverse visual concepts to downstream tasks. However, current methods explore the fixed adaptation of the learnable prompt on the seen domains, which make them over-emphasize the primary visual features observed during training, limiting their generalization capabilities to the unseen domains. In this work, we propose AENet, which endows semantic information into the visual prompt to distill semantic-enhanced prompt for visual representation enrichment, enabling effective knowledge transfer for ZSL. AENet comprises two key steps: 1) exploring the concept-harmonized tokens for the visual and attribute modalities, grounded on the modal-sharing token that represents consistent visual-semantic concepts; and 2) yielding the semantic-enhanced prompt via the visual residual refinement unit with attribute consistency supervision. It is further integrated with primary visual features to attend to semantic-related information for visual enhancement, thus strengthening transferable ability. Experimental results on three benchmarks show that our AENet outperforms existing state-of-the-art ZSL methods.

Deep neural networks have become foundational to advancements in multiple domains, including recommendation systems, natural language processing, and so on. Despite their successes, these models often contain incompatible parameters that can be underutilized or detrimental to model performance, particularly when faced with specific, varying data distributions. Existing research excels in removing such parameters or merging the outputs of multiple different pretrained models. However, the former focuses on efficiency rather than performance, while the latter requires several times more computing and storage resources to support inference. In this paper, we set the goal to explicitly improve these incompatible parameters by leveraging the complementary strengths of different models, thereby directly enhancing the models without any additional parameters. Specifically, we propose Compatibility-aware Knowledge Integration (CKI), which consists of Parameter Compatibility Assessment and Parameter Splicing, which are used to evaluate the knowledge content of multiple models and integrate the knowledge into one model, respectively. The integrated model can be used directly for inference or for further fine-tuning. Extensive experiments on various recommendation and language datasets show that CKI can effectively optimize incompatible parameters under multiple tasks and settings to break through the training limit of the original model without increasing the inference cost.

Recent advancements in multimodal large language models (MLLMs) have shown unprecedented capabilities in advancing various vision-language tasks. However, MLLMs face significant challenges with hallucinations, and misleading outputs that do not align with the input data. While existing efforts are paid to combat MLLM hallucinations, several pivotal challenges are still unsolved. First, while current approaches aggressively focus on addressing errors at the perception level, another important type at the cognition level requiring factual commonsense can be overlooked. In addition, existing methods might fall short in finding a more effective way to represent visual input, which is yet a key bottleneck that triggers visual hallucinations. Moreover, MLLMs can frequently be misled by faulty textual inputs and cause hallucinations, while unfortunately, this type of issue has long been overlooked by existing studies. Inspired by human intuition in handling hallucinations, this paper introduces a novel bottom-up reasoning framework. Our framework systematically addresses potential issues in both visual and textual inputs by verifying and integrating perception-level information with cognition-level commonsense knowledge, ensuring more reliable outputs. Extensive experiments demonstrate significant improvements in multiple hallucination benchmarks after integrating MLLMs with the proposed framework. In-depth analyses reveal the great potential of our methods in addressing perception- and cognition-level hallucinations.

Large Language Models (LLMs) have demonstrated impressive capabilities in complex reasoning tasks. However, they can be easily misled by unfaithful arguments during conversations, even when their original statements are correct. To this end, we investigate the problem of maintaining faithful integrity in LLMs. This involves ensuring that LLMs adhere to their faithful statements in the face of opposing arguments and are able to correct their incorrect statements when presented with faithful arguments. In this work, we propose a novel framework, named Alignment for Faithful Integrity with Confidence Estimation (AFICE), which aims to align the LLM responses with faithful integrity. Specifically, AFICE first designs a Bilateral Confidence Estimation (BCE) approach for estimating the uncertainty of each response generated by the LLM given a specific context, which simultaneously estimate the model's confidence to the question based on the internal states during decoding as well as to the answer based on cumulative probability ratios. With the BCE, we construct a conversational preference dataset composed of context, original statement, and argument, which is adopted for aligning the LLM for faithful integrity using Direct Preference Optimization (DPO). Extensive experimental results on a wide range of benchmarks demonstrate significant improvements in the LLM's ability to maintain faithful responses when encountering opposing arguments, ensuring both the practical utility and trustworthiness of LLMs in complex interactive settings.

Hongzhan Lin, Yang Deng, Yuxuan Gu, Wenxuan Zhang, Jing Ma, See-Kiong Ng, Tat-Seng Chua. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

With the advancement of large language models (LLMs), intelligent models have evolved from mere tools to autonomous agents with their own goals and strategies for cooperating with humans.This evolution has birthed a novel paradigm in NLP, i.e., human-model cooperation, that has yielded remarkable progress in numerous NLP tasks in recent years.In this paper, we take the first step to present a thorough review of human-model cooperation, exploring its principles, formalizations, and open challenges.In particular, we introduce a new taxonomy that provides a unified perspective to summarize existing approaches.Also, we discuss potential frontier areas and their corresponding challenges.We regard our work as an entry point, paving the way for more breakthrough research in this regard.

Yuxuan Gu, Wenjie Wang, Xiaocheng Feng, Weihong Zhong, Kun Zhu, Lei Huang, Ting Liu, Bing Qin, Tat-Seng Chua. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Jinghan He, Kuan Zhu, Haiyun Guo, Junfeng Fang, Zhenglin Hua, Yuheng Jia, Ming Tang, Tat-Seng Chua, Jinqiao Wang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Moxin Li, Yong Zhao, Wenxuan Zhang, Shuaiyi Li, Wenya Xie, See-Kiong Ng, Tat-Seng Chua, Yang Deng. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Yiyan Xu, Jinghao Zhang, Alireza Salemi, Xinting Hu, Wenjie Wang, Fuli Feng, Hamed Zamani, Xiangnan He, Tat-Seng Chua. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Jinghao Zhang, Yuting Liu, Wenjie Wang, Qiang Liu, Shu Wu, Liang Wang, Tat-Seng Chua. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Weixiang Zhao, Yulin Hu, Yang Deng, Jiahe Guo, Xingyu Sui, Xinyang Han, An Zhang, Yanyan Zhao, Bing Qin, Tat-Seng Chua, Ting Liu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Weixiang Zhao, Yulin Hu, Yang Deng, Tongtong Wu, Wenxuan Zhang, Jiahe Guo, An Zhang, Yanyan Zhao, Bing Qin, Tat-Seng Chua, Ting Liu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Recent breakthroughs in generative Artificial Intelligence (AI) have ignited a revolutionary wave across information retrieval and recommender systems. This workshop serves as a premier interdisciplinary platform to explore how generative models, particularly Large Language Models (LLMs) and Large Multimodal Models (LMMs), are transforming multimodal search and recommendation paradigms [3, 6, 9, 10, 12-14]. We aim to convene researchers and practitioners to discuss innovative architectures, methodologies, and evaluation strategies spanning generative document retrieval [5, 8] generative image retrieval [ 7, 16], grounded answer generation [17], generative recommendation [2, 4, 11], and related tasks involving multiple modalities [1,15]. The workshop will facilitate discussions on improving algorithms, generating personalized content, evolving user-system interactions, enhancing trustworthiness, and refining evaluation methodologies for these cutting-edge systems. This timely workshop seeks to identify promising research directions, address key challenges, and foster collaborations towards the development of next-generation intelligent systems.

The emergence of large generative models is transforming the landscape of recommender systems. One of the most fundamental components in building these models is action tokenization, the process of converting human-readable data (e.g., user-item interactions) into machine-readable formats (e.g., discrete token sequences). In this tutorial, we present a comprehensive overview of existing action tokenization techniques, converting actions to (1) item IDs, (2) textual descriptions, and (3) semantic IDs. We then make an in-depth discussion on the challenges and open questions of building large generative recommendation models from the perspective of action tokenization. Materials of this tutorial are available at: https://large-genrec.github.io/.

Traditional feedback learning for hallucination reduction relies on labor-intensive manual labeling or expensive proprietary models. This leaves the community without foundational knowledge about how to build high-quality feedback with open-source MLLMs. In this work, we introduce RLAIF-V, a novel framework that aligns MLLMs in a fully open-source paradigm. RLAIF-V maximally explores open-source MLLMs from two perspectives, including high-quality feedback data generation for preference learning and self-feedback guidance for inference-time scaling. Extensive experiments on six benchmarks in both automatic and human evaluation show that RLAIF-V substantially enhances the trustworthiness of models at both preference learning and inference time. RLAIF-V 7B reduces object hallucination by 80.7% and overall hallucination by 33.7%. Remarkably, RLAIF-V 12B further reveals the self-alignment potential of open-source MLLMs, where the model can learn from feedback of itself to achieve super GPT-4V trustworthiness.

Video Large Language Models (Video-LLMs) have recently shown strong performance in basic video understanding tasks, such as captioning and coarse-grained question answering, but struggle with compositional reasoning that requires multi-step spatio-temporal inference across object relations, interactions, and events. The hurdles to enhancing this capability include extensive manual labor, the lack of spatio-temporal compositionality in existing training data and the absence of explicit reasoning supervision. In this paper, we propose STEP, a novel graph-guided self-training method that enables Video-LLMs to generate reasoning-rich fine-tuning data from any raw videos to improve itself. Specifically, we first induce Spatio-Temporal Scene Graph (STSG) representation of diverse videos to capture fine-grained, multi-granular video semantics. Then, the STSGs guide the derivation of multi-step reasoning Question-Answer (QA) data with Chain-of-Thought (CoT) rationales. Both answers and rationales are integrated as training objective, aiming to enhance model’s reasoning abilities by supervision over explicit reasoning steps. Experimental results demonstrate the effectiveness of STEP across models of varying scales, with a significant 21.3% improvement in tasks requiring three or more reasoning steps. Furthermore, it achieves superior performance with a minimal amount of self-generated rationale-enriched training samples in both compositional reasoning and comprehensive understanding benchmarks, highlighting the broad applicability and vast potential.

Large Multimodal Models (LMMs) have demonstrated impressive capabilities in multimodal understanding and generation, pushing forward advancements in text-to-image generation. However, achieving accurate text-image alignment for LMMs, particularly in compositional scenarios, remains challenging. Existing approaches, such as layout planning for multi-step generation and learning3 from human feedback or AI feedback, depend heavily on prompt engineering, costly human annotations, and continual upgrading, limiting flexibility and scalability. In this work, we introduce a model-agnostic iterative self-improvement framework (SILMM) that can enable LMMs to provide helpful and scalable self-feedback and optimize text-image alignment via Direct Preference Optimization (DPO). DPO can readily applied to LMMs that use discrete visual tokens as intermediate image representations; while it is less suitable for LMMs with continuous visual features, as obtaining generation probabilities is challenging. To adapt SILMM to LMMs with continuous features, we propose a diversity mechanism to obtain diverse representations and a kernel-based continuous DPO for alignment. Extensive experiments on three compositional text-to-image generation benchmarks validate the effectiveness and superiority of SILMM, showing improvements exceeding 30% on T2I-CompBench++ and around 20% on DPG-Bench. The code is available at https://silmm.github.io/.

Scene graph (SG) representations can neatly and efficiently describe scene semantics, which has driven sustained intensive research in SG generation. In the real world, multiple modalities often coexist, with different types, such as images, text, video, and 3D data, expressing distinct characteristics. Unfortunately, current SG research is largely confined to single-modality scene modeling, preventing the full utilization of the complementary strengths of different modality SG representations in depicting holistic scene semantics. To this end, we introduce Universal SG (USG), a novel representation capable of fully characterizing comprehensive semantic scenes from any given combination of modality inputs, encompassing modality-invariant and modality-specific scenes. Further, we tailor a niche-targeting USG parser, USG-Par, which effectively addresses two key bottlenecks of cross-modal object alignment and out-of-domain challenges. We design the USG-Par with modular architecture for end-to-end USG generation, in which we devise an object associator to relieve the modality gap for cross-modal object alignment. Further, we propose a text-centric scene contrasting learning mechanism to mitigate domain imbalances by aligning multimodal objects and relations with textual SGs. Through extensive experiments, we demonstrate that USG offers a stronger capability for expressing scene semantics than standalone SGs, and also that our USG-Par achieves higher efficacy and performance. The project page is https://sqwu.top/USG/.

The latest emerged 4D Panoptic Scene Graph (4D-PSG) provides an advanced-ever representation for comprehensively modeling the dynamic 4D visual real world. Unfortunately, current pioneering 4D-PSG research can primarily suffer from data scarcity issues severely, as well as the resulting out-of-vocabulary problems; also, the pipeline nature of the benchmark generation method can lead to suboptimal performance. To address these challenges, this paper investigates a novel framework for 4D-PSG generation that leverages rich 2D visual scene annotations to enhance 4D scene learning. First, we introduce a 4D Large Language Model (4D-LLM) integrated with a 3D mask decoder for end-to-end generation of 4D-PSG. A chained SG inference mechanism is further designed to exploit LLMs’ open-vocabulary capabilities to infer accurate and comprehensive object and relation labels iteratively. Most importantly, we propose a 2D-to-4D visual scene transfer learning framework, where a spatial-temporal scene transcending strategy effectively transfers dimension-invariant features from abundant 2D SG annotations to 4D scenes, effectively compensating for data scarcity in 4D-PSG. Extensive experiments on the benchmark data demonstrate that we strikingly outperform baseline models by a large margin, highlighting the effectiveness of our method. The project page is https://sqwu.top/PSG-4D-LLM/.

Animatable head avatar generation typically requires extensive data for training. To reduce the data requirements, a natural solution is to leverage existing data-free static avatar generation methods, such as pre-trained diffusion models with score distillation sampling (SDS), which align avatars with pseudo ground-truth outputs from the diffusion model. However, directly distilling 4D avatars from video diffusion often leads to over-smooth results due to spatial and temporal inconsistencies in the generated video. To address this issue, we propose Zero-1-to-A, a robust method that synthesizes a spatial and temporal consistency dataset for 4D avatar reconstruction using the video diffusion model. Specifically, Zero-1-to-A iteratively constructs video datasets and optimizes animatable avatars in a progressive manner, ensuring that avatar quality increases smoothly and consistently throughout the learning process. This progressive learning involves two stages: (1) Spatial Consistency Learning fixes expressions and learns from front-to-side views, and (2) Temporal Consistency Learning fixes views and learns from relaxed to exaggerated expressions, generating 4D avatars in a simple-to-complex manner. Extensive experiments demonstrate that Zero-1-to-A improves fidelity, animation quality, and rendering speed compared to existing diffusion-based methods, providing a solution for lifelike avatar creation. Code is publicly available at: https://github.com/ZhenglinZhou/Zero-1-to-A.

We introduce EgoTextVQA, a novel and rigorously constructed benchmark for egocentric QA assistance involving scene text. EgoTextVQA contains 1.5K ego-view videos and 7K scene-text aware questions that reflect real user needs in outdoor driving and indoor house-keeping activities. The questions are designed to elicit identification and reasoning on scene text in an egocentric and dynamic environment. With EgoTextVQA, we comprehensively evaluate 10 prominent multimodal large language models. Currently, all models struggle, and the best results (Gemini 1.5 Pro) are around 33% accuracy, highlighting the severe deficiency of these techniques in egocentric QA assistance. Our further investigations suggest that precise temporal grounding and multi-frame reasoning, along with high resolution and auxiliary scene-text inputs, are key for better performance. With thorough analyses and heuristic suggestions, we hope EgoTextVQA can serve as a solid testbed for research in egocentric scene-text QA assistance. Our dataset is released at: https://github.com/zhousheng97/EgoTextVQA.

Unlike traditional search engines that present ranked lists of webpages, generative search engines rely solely on in-line citations as the key gateway to original real-world webpages, making it crucial to examine whether LLMgenerated citations have biases-particularly for politically sensitive queries.To investigate this, we first construct AllSides-2024, a new dataset comprising the latest real-world news articles (Jan.2024 -Dec.2024) labeled with left-or right-leaning stances.Through systematic evaluations, we find that LLMs exhibit a consistent tendency to cite left-leaning sources at notably higher rates compared to traditional retrieval systems (e.g., BM25 and dense retrievers).Controlled experiments further reveal that this bias arises from a preference for media outlets identified as left-leaning, rather than for left-oriented content itself.Meanwhile, our findings show that while LLMs struggle to infer political bias from news content alone, they can almost perfectly recognize the political orientation of media outlets based on their names.These insights highlight the risk that, in the era of generative search engines, information exposure may be disproportionately shaped by specific media outlets, potentially shaping public perception and decision-making 1 .

Weixiang Zhao, Jiahe Guo, Yulin Hu, Yang Deng, An Zhang, Xingyu Sui, Xinyang Han, Yanyan Zhao, Bing Qin, Tat-Seng Chua, Ting Liu. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Recent advancements in reward signal usage for Large Language Models (LLMs) are remarkable. However, significant challenges exist when transitioning reward signal to the multimodal domain, including labor-intensive annotations, over-reliance on one-step rewards, and inadequate evaluation. To address these issues, we propose SVIP, a novel approach to train a step-level multi-dimensional Chain-of-Thought (CoT) reward model automatically. It generates code for solving visual tasks and transforms the analysis of code blocks into the evaluation of CoT step as training samples. Then, we train SVIP-Reward model using a multi-head attention mechanism called TriAtt-CoT. The advantages of SVIP-Reward are evident throughout the entire process of MLLM. We also introduce a benchmark for CoT reward model training and testing. Experimental results demonstrate that SVIP-Reward improves MLLM performance across training and inference-time scaling, yielding better results on benchmarks while reducing hallucinations and enhancing reasoning ability.

The rapid advancements in Vision Language Models (VLMs) have prompted the development of multi-modal medical assistant systems. Despite this progress, current models still have inherent probabilistic uncertainties, often producing erroneous or unverified responses-an issue with serious implications in medical applications. Existing methods aim to enhance the performance of Medical Vision Language Model (MedVLM) by adjusting model structure, fine-tuning with high-quality data, or through preference fine-tuning. However, these training-dependent strategies are costly and still lack sufficient alignment with clinical expertise. To address these issues, we propose an expert-in-the-loop framework named Expert-Controlled Classifier-Free Guidance (Expert-CFG) to align MedVLM with clinical expertise without additional training. This framework introduces an uncertainty estimation strategy to identify unreliable outputs. It then retrieves relevant references to assist experts in highlighting key terms and applies classifier-free guidance to refine the token embeddings of MedVLM, ensuring that the adjusted outputs are correct and align with expert highlights. Evaluations across three medical visual question answering benchmarks demonstrate that the proposed Expert-CFG, with 4.2B parameters and limited expert annotations, outperforms state-of-the-art models with 13B parameters. The results demonstrate the feasibility of deploying such a system in resource-limited settings for clinical use.

Large Language Model (LLM)-based generative recommendation has achieved notable success, yet its practical deployment is costly particularly due to excessive inference latency caused by autoregressive decoding. For lossless LLM decoding acceleration, Speculative Decoding (SD) has emerged as a promising solution. However, applying SD to generative recommendation presents unique challenges due to the requirement of generating top-K items (i.e., K distinct token sequences) as a recommendation list by beam search. This leads to more stringent verification in SD, where all the top-K sequences from the target LLM must be successfully drafted by the draft model at each decoding step. To alleviate this, we consider 1) boosting top-K sequence alignment between the draft model and the target LLM, and 2) relaxing the verification strategy to reduce trivial LLM calls. To this end, we propose an alignment framework named AtSpeed, which presents the AtSpeed-S optimization objective for top-K alignment under the strict top-K verification. Moreover, we introduce a relaxed sampling verification strategy that allows high-probability non-top-K drafted sequences to be accepted, significantly reducing LLM calls. Correspondingly, we propose AtSpeed-R for top-K alignment under this relaxed sampling verification. Empirical results on two real-world datasets demonstrate that AtSpeed significantly accelerates LLM-based generative recommendation, e.g., near 2x speedup under strict top-K verification and up to 2.5x speedup under relaxed sampling verification. The codes and datasets are available at~\url{https://github.com/Linxyhaha/AtSpeed}.

Group max-min fairness (MMF) is commonly used in fairness-aware recommender systems (RS) as an optimization objective, as it aims to protect marginalized item groups and ensures a fair competition platform. However, our theoretical analysis indicates that integrating MMF constraint violates the assumption of sample independence during optimization, causing the loss function to deviate from linear additivity. Such nonlinearity property introduces the Jensen gap between the model's convergence point and the optimal point if mini-batch sampling is applied. Both theoretical and empirical studies show that as the mini-batch size decreases and the group size increases, the Jensen gap will widen accordingly. Some methods using heuristic re-weighting or debiasing strategies have the potential to bridge the Jensen gap. However, they either lack theoretical guarantees or suffer from heavy computational costs. To overcome these limitations, we first theoretically demonstrate that the MMF-constrained objective can be essentially reformulated as a group-weighted optimization objective. Then we present an efficient and effective algorithm named FairDual, which utilizes a dual optimization technique to minimize Jensen gap. Our theoretical analysis demonstrates that FairDual can achieve a sub-linear convergence rate to the globally optimal solution and the Jensen gap can be well bounded under a mini-batch sampling strategy with random shuffle. Extensive experiments conducted using six large-scale RS backbone models on three publicly available datasets demonstrate that FairDual outperforms all baselines in terms of both accuracy and fairness.

Advancements in neural networks have significantly enhanced the performance of classification models, achieving remarkable accuracy across diverse datasets. However, these models often lack transparency and do not support interactive reasoning with human users, which are essential attributes for applications that require trust and user engagement. To overcome these limitations, we introduce an innovative framework, Neural Causal Graph (NCG), that integrates causal inference with neural networks to enable interpretable and intervenable reasoning. We then propose an intervention training method to model the intervention probability of the prediction, serving as a contextual prompt to facilitate the fine-grained reasoning and human-AI interaction abilities of NCG. Our experiments show that the proposed framework significantly enhances the performance of traditional classification baselines. Furthermore, NCG achieves nearly 95\% top-1 accuracy on the ImageNet dataset by employing a test-time intervention method. This framework not only supports sophisticated post-hoc interpretation but also enables dynamic human-AI interactions, significantly improving the model's transparency and applicability in real-world scenarios.

The recent advancements in large language models (LLMs) and pre-trained vision models have accelerated the development of vision-language large models (VLLMs), enhancing the interaction between visual and linguistic modalities. Despite their notable success across various domains, VLLMs face challenges in modality alignment, which can lead to issues like hallucinations and unsafe content generation. Current alignment techniques often rely on coarse feedback and external datasets, limiting scalability and performance. In this paper, we propose FiSAO (Fine-Grained Self-Alignment Optimization), a novel self-alignment method that utilizes the model’s own visual encoder as a fine-grained verifier to improve vision-language alignment without the need for additional data. By leveraging token-level feedback from the vision encoder, FiSAO significantly improves vision-language alignment, even surpassing traditional preference tuning methods that require additional data. Through both theoretical analysis and experimental validation, we demonstrate that FiSAO effectively addresses the misalignment problem in VLLMs, marking the first instance of token-level rewards being applied to such models. Our code is avaliable at \url{https://anonymous.4open.science/r/FISAO-57F0/}.

Large language models (LLMs) often exhibit hallucinations, producing incorrect or outdated knowledge. Hence, model editing methods have emerged to enable targeted knowledge updates. To achieve this, a prevailing paradigm is the locating-then-editing approach, which first locates influential parameters and then edits them by introducing a perturbation. While effective, current studies have demonstrated that this perturbation inevitably disrupt the originally preserved knowledge within LLMs, especially in sequential editing scenarios. To address this, we introduce AlphaEdit, a novel solution that projects perturbation onto the null space of the preserved knowledge before applying it to the parameters. We theoretically prove that this projection ensures the output of post-edited LLMs remains unchanged when queried about the preserved knowledge, thereby mitigating the issue of disruption. Extensive experiments on various LLMs, including LLaMA3, GPT2-XL, and GPT-J, show that AlphaEdit boosts the performance of most locating-then-editing methods by an average of 36.7% with a single line of additional code for projection solely.

3D molecule generation is crucial for drug discovery and material design. While prior efforts focus on 3D diffusion models for their benefits in modeling continuous 3D conformers, they overlook the advantages of 1D SELFIES-based Language Models (LMs), which can generate 100\% valid molecules and leverage the billion-scale 1D molecule datasets. To combine these advantages for 3D molecule generation, we propose a foundation model -- NExT-Mol: 3D Diffusion Meets 1D Language Modeling for 3D Molecule Generation. NExT-Mol uses an extensively pretrained molecule LM for 1D molecule generation, and subsequently predicts the generated molecule's 3D conformers with a 3D diffusion model. We enhance NExT-Mol's performance by scaling up the LM's model size, refining the diffusion neural architecture, and applying 1D to 3D transfer learning. Notably, our 1D molecule LM significantly outperforms baselines in distributional similarity while ensuring validity, and our 3D diffusion model achieves leading performances in conformer prediction. Given these improvements in 1D and 3D modeling, NExT-Mol achieves a 26\% relative improvement in 3D FCD for de novo 3D generation on GEOM-DRUGS, and a 13\% average relative gain for conditional 3D generation on QM9-2014. Our codes and pretrained checkpoints are available at https://github.com/acharkq/NExT-Mol.

Recommender systems aim to predict personalized item rankings by modeling user preference distributions derived from historical behavior data. While diffusion models (DMs) have recently gained attention for their ability to model complex distributions, current DM-based recommenders typically rely on traditional objectives such as mean squared error (MSE) or standard recommendation objectives. These approaches are either suboptimal for personalized ranking tasks or fail to exploit the full generative potential of DMs. To address these limitations, we propose \textbf{PreferDiff}, an optimization objective tailored for DM-based recommenders. PreferDiff reformulates the traditional Bayesian Personalized Ranking (BPR) objective into a log-likelihood generative framework, enabling it to effectively capture user preferences by integrating multiple negative samples. To handle the intractability, we employ variational inference, minimizing the variational upper bound. Furthermore, we replace MSE with cosine error to improve alignment with recommendation tasks, and we balance generative learning and preference modeling to enhance the training stability of DMs. PreferDiff devises three appealing properties. First, it is the first personalized ranking loss designed specifically for DM-based recommenders. Second, it improves ranking performance and accelerates convergence by effectively addressing hard negatives. Third, we establish its theoretical connection to Direct Preference Optimization (DPO), demonstrating its potential to align user preferences within a generative modeling framework. Extensive experiments across six benchmarks validate PreferDiff's superior recommendation performance. Our codes are available at \url{https://github.com/lswhim/PreferDiff}.

How humans can effectively and efficiently acquire images has always been a perennial question. A classic solution is *text-to-image retrieval* from an existing database; however, the limited database typically lacks creativity. By contrast, recent breakthroughs in *text-to-image generation* have made it possible to produce attractive and counterfactual visual content, but it faces challenges in synthesizing knowledge-intensive images. In this work, we rethink the relationship between text-to-image generation and retrieval, proposing a *unified* framework for both tasks with one single Large Multimodal Model (LMM). Specifically, we first explore the intrinsic discriminative abilities of LMMs and introduce an efficient generative retrieval method for text-to-image retrieval in a training-free manner. Subsequently, we unify generation and retrieval autoregressively and propose an autonomous decision mechanism to choose the best-matched one between generated and retrieved images as the response to the text prompt. To standardize the evaluation of unified text-to-image generation and retrieval, we construct TIGeR-Bench, a benchmark spanning both creative and knowledge-intensive domains. Extensive experiments on TIGeR-Bench and two retrieval benchmarks, *i.e.*, Flickr30K and MS-COCO, demonstrate the superiority of our proposed framework.

Recent studies empirically indicate that language models (LMs) encode rich world knowledge beyond mere semantics, attracting significant attention across various fields. However, in the recommendation domain, it remains uncertain whether LMs implicitly encode user preference information. Contrary to prevailing understanding that LMs and traditional recommenders learn two distinct representation spaces due to the huge gap in language and behavior modeling objectives, this work re-examines such understanding and explores extracting a recommendation space directly from the language representation space. Surprisingly, our findings demonstrate that item representations, when linearly mapped from advanced LM representations, yield superior recommendation performance. This outcome suggests the possible homomorphism between the advanced language representation space and an effective item representation space for recommendation, implying that collaborative signals may be implicitly encoded within LMs. Motivated by the finding of homomorphism, we explore the possibility of designing advanced collaborative filtering (CF) models purely based on language representations without ID-based embeddings. To be specific, we incorporate several crucial components (i.e., a multilayer perceptron (MLP), graph convolution, and contrastive learning (CL) loss function) to build a simple yet effective model, with the language representations of item textual metadata (i.e., title) as the input. Empirical results show that such a simple model can outperform leading ID-based CF models on multiple datasets, which sheds light on using language representations for better recommendation. Moreover, we systematically analyze this simple model and find several key features for using advanced language representations: a good initialization for item representations, superior zero-shot recommendation abilities in new datasets, and being aware of user intention. Our findings highlight the connection between language modeling and behavior modeling, which can inspire both natural language processing and recommender system communities.

Multimodal Large Language Models (MLLMs) have demonstrated exceptional capabilities in processing vision-language tasks. One of the crux of MLLMs lies in vision tokenization, which involves efficiently transforming input visual signals into feature representations that are most beneficial for LLMs. However, existing vision tokenizers, essential for semantic alignment between vision and language, remain problematic. Existing methods aggressively fragment visual input, corrupting the visual semantic integrity. To address this, this paper proposes a novel dynamic Semantic-Equivalent Vision Tokenizer (SeTok), which groups visual features into semantic units via a dynamic clustering algorithm, flexibly determining the number of tokens based on image complexity. The resulting vision tokens effectively preserve semantic integrity and capture both low-frequency and high-frequency visual features. The proposed MLLM (Setokim) equipped with SeTok significantly demonstrates superior performance across various tasks, as evidenced by our experimental results.

The Multimodal Large Language Model (MLLM) is currently experiencing rapid growth, driven by the advanced capabilities of language-based LLMs. Unlike their specialist predecessors, existing MLLMs are evolving towards a Multimodal Generalist paradigm. Initially limited to understanding multiple modalities, these models have advanced to not only comprehend but also generate across modalities. Their capabilities have expanded from coarse-grained to fine-grained multimodal understanding and from supporting singular modalities to accommodating a wide array of or even arbitrary modalities. To assess the capabilities of various MLLMs, a diverse array of benchmark test sets has been proposed. This leads to a critical question: Can we simply assume that higher performance across tasks indicates a stronger MLLM capability, bringing us closer to human-level AI? We argue that the answer is not as straightforward as it seems. In this project, we introduce an evaluation framework to delineate the capabilities and behaviors of current multimodal generalists. This framework, named General-Level, establishes 5-scale levels of MLLM performance and generality, offering a methodology to compare MLLMs and gauge the progress of existing systems towards more robust multimodal generalists and, ultimately, towards AGI (Artificial General Intelligence). Central to our framework is the use of Synergy as the evaluative criterion, categorizing capabilities based on whether MLLMs preserve synergy across comprehension and generation, as well as across multimodal interactions. To evaluate the comprehensive abilities of various generalists, we present a massive multimodal benchmark, General-Bench, which encompasses a broader spectrum of skills, modalities, formats, and capabilities, including over 700 tasks and 325,800 instances. The evaluation results that involve over 100 existing state-of-the-art MLLMs uncover the capability rankings of generalists, highlighting the challenges in reaching genuine AI. We expect this project to pave the way for future research on next-generation multimodal foundation models, providing a robust infrastructure to accelerate the realization of AGI. Project Page: https://generalist.top/, Leaderboard: https://generalist.top/leaderboard/, Benchmark: https://huggingface.co/General-Level/.

Large language models (LLMs) often produce incorrect or outdated information, necessitating efficient and precise knowledge updates. Current model editing methods, however, struggle with long-form knowledge in diverse formats, such as poetry, code snippets, and mathematical derivations. These limitations arise from their reliance on editing a single token’s hidden state, a limitation we term as ``efficacy barrier''. To solve this, we propose \textbf{AnyEdit}, a new autoregressive editing paradigm. It decomposes long-form knowledge into sequential chunks and iteratively edits the key token in each chunk, ensuring consistent and accurate outputs. Theoretically, we ground AnyEdit in the Chain Rule of Mutual Information, showing its ability to update any knowledge within LLMs. Empirically, it outperforms strong baselines by 21.5\% on benchmarks including UnKEBench, AKEW, and our new \textbf{EditEverything} dataset for long-form diverse-formatted knowledge. Additionally, AnyEdit serves as a plug-and-play framework, enabling current editing methods to update knowledge with arbitrary length and format, significantly advancing the scope and practicality of LLM knowledge editing. Our code is available at: \url{https://github.com/jianghoucheng/AnyEdit}.

Federated prompt learning (FPL) for vision-language models is a powerful approach to collaboratively adapt models across distributed clients while preserving data privacy. However, existing FPL approaches suffer from a trade-off between performance and robustness, particularly in out-of-distribution (OOD) shifts, limiting their reliability in real-world scenarios. The inherent in-distribution (ID) data heterogeneity among different clients makes it more challenging to maintain this trade-off. To fill this gap, we introduce a Federated OOD-aware Context Optimization (FOCoOp) framework, which captures diverse distributions among clients using ID global prompts, local prompts, and OOD prompts. Specifically, FOCoOp leverages three sets of prompts to create both class-level and distribution-level separations, which adapt to OOD shifts through bi-level distributionally robust optimization. Additionally, FOCoOp improves the discrimination consistency among clients, i.e., calibrating global prompts, seemly OOD prompts, and OOD prompts by Semi-unbalanced optimal transport. The extensive experiments on real-world datasets demonstrate that FOCoOp effectively captures decentralized heterogeneous distributions and enhances robustness of different OOD shifts. The project is available at GitHub.

The development of Generalist Virtual Agents (GVAs) has shown significant promise in autonomous task execution. However, current training paradigms face critical limitations, including reliance on outcome supervision and labor-intensive human annotations. To address these challenges, we propose Similar, a step-wise multi-dimensional generalist reward model, which offers fine-grained signals for agent training and can choose better actions for inference-time scaling. Specifically, we begin by systematically defining five dimensions for evaluating agent actions. Building on this framework, we design an MCTS-P algorithm to automatically collect and annotate step-wise, five-dimensional agent execution data. Using this data, we train Similar with our crafted Triple-M strategy. Furthermore, we introduce the first benchmark in the virtual agent domain for step-wise, multi-dimensional reward model training and evaluation, named SRM. This benchmark consists of two components: SRMTrain, which serves as the training set for Similar, and SRMEval, a manually selected test set for evaluating the reward model. Experimental results demonstrate that Similar, through its step-wise, multi-dimensional assessment and synergistic gain, provides GVAs with effective intermediate signals during both training and inference-time scaling. The code is available at https://github.com/antgroup/Similar.

Recent advances in conditional diffusion models have shown promise for generating realistic TalkingFace videos, yet challenges persist in achieving consistent head movement, synchronized facial expressions, and accurate lip synchronization over extended generations. To address these, we introduce the Motion-priors Conditional Diffusion Model (MCDM), which utilizes both archived and current clip motion priors to enhance motion prediction and ensure temporal consistency. The model consists of three key elements: (1) an archived-clip motion-prior that incorporates historical frames and a reference frame to preserve identity and context; (2) a present-clip motion-prior diffusion model that captures multimodal causality for accurate predictions of head movements, lip sync, and expressions; and (3) a memory-efficient temporal attention mechanism that mitigates error accumulation by dynamically storing and updating motion features. We also introduce the TalkingFace-Wild dataset, a multilingual collection of over 200 hours of footage across 10 languages. Experimental results demonstrate the effectiveness of MCDM in maintaining identity and motion continuity for long-term TalkingFace generation.

Text-to-3D generation automates 3D content creation from textual descriptions, which offers transformative potential across various fields. However, existing methods often struggle to align generated content with human preferences, limiting their applicability and flexibility. To address these limitations, in this paper, we propose DreamDPO, an optimization-based framework that integrates human preferences into the 3D generation process, through direct preference optimization. Practically, DreamDPO first constructs pairwise examples, then validates their alignment with human preferences using reward or large multimodal models, and lastly optimizes the 3D representation with a preference-driven loss function. By leveraging relative preferences, DreamDPO reduces reliance on precise quality evaluations while enabling fine-grained controllability through preference-guided optimization. Experiments demonstrate that DreamDPO achieves state-of-the-art results, and provides higher-quality and more controllable 3D content compared to existing methods. The code and models will be open-sourced.

Due to the critical issues of privacy and partial occlusion, license plate information is not always available in vehicle recognition systems. Consequently, researchers have increasingly turned towards vehicle re-identification (reID) techniques to bridge the gap between cross-view camera systems. Despite the growing interest, one major challenge persists: the scarcity of authentic, large-scale training datasets. To address this challenge, this paper introduces a coarse-to-fine generation pipeline designed to synthesize high-fidelity vehicle data, thereby facilitating subsequent vehicle representation learning. Specifically, the proposed approach consists of three stages: Prompt Processing, Diffusion Fine-tuning, and Semantic Filtering. First, we collect detailed prompts from vehicle websites and companies with fine-grained vehicle prototype attributes. Next, we leverage the prior knowledge of these automotive prototypes to fine-tune diffusion models. Finally, to ensure the quality of the synthesized data, we employ pretrained vision-language models to filter out substandard images. Building upon the high-quality data generated by this pipeline, we validate the effectiveness using vanilla models. Extensive experimental evaluations demonstrate that our approach achieves competitive accuracy on public benchmarks such as VeRi-776, VehicleID and CityFlowV2, and is compatible with various model architectures.

Long video understanding with Large Language Models (LLMs) enables the description of objects that are not explicitly present in the training data. However, continuous changes in known objects and the emergence of new ones require up-to-date knowledge of objects and their dynamics for effective understanding of the open world. To alleviate this, we propose an efficient Retrieval-Enhanced Video Understanding method, dubbed REVU, which leverages external knowledge to enhance the performance of open-world learning. First, REVU introduces an extensible external text-object memory with minimal text-visual mapping, involving static and dynamic multimodal information to help LLMs-based models align text and vision features. Second, REVU retrieves object information from external databases and dynamically integrates frame-specific data from videos, enabling effective knowledge aggregation to comprehend the open world. We conducted experiments on multiple benchmark datasets, and our model demonstrates strong adaptability to out-of-domain data without requiring additional fine-tuning or re-training. Experiments on benchmark video understanding datasets reveal that our model achieves state-of-the-art performance and robust generalization.

Large Language Model-based generative recommendation (LLMRec) has achieved notable success, but it suffers from high inference latency due to massive computational overhead and memory pressure of KV Cache. Existing KV Cache reduction methods face critical limitations: cache compression offers marginal acceleration given recommendation tasks' short decoding steps, while prompt compression risks discarding vital interaction history. Through systematic analysis of attention patterns in LLMRec, we uncover two pivotal insights: 1) layer-wise attention sparsity inversion where early layers retain dense informative patterns while later layers exhibit high redundancy, and 2) dual attention sinks phenomenon where attention scores concentrate on both head and tail tokens of input sequences. Motivated by these insights, we propose EARN, an efficient inference framework that leverages the early layers to compress information into register tokens placed at the input sequence boundaries, then focuses solely on these tokens in the subsequent layers. Extensive experiments on three datasets, two LLMRec methods and two LLM architectures demonstrate EARN's superiority, achieving up to 3.79x speedup and 80.8% KV Cache reduction with better accuracy than the general finetuning approach. Our work bridges the efficiency-effectiveness gap in LLMRec, offering practical deployment advantages for industrial scenarios.

Sequential recommendation aims to predict users' future interactions by modeling collaborative filtering (CF) signals from historical behaviors of similar users or items. Traditional sequential recommenders predominantly rely on ID-based embeddings, which capture CF signals through high-order co-occurrence patterns. However, these embeddings depend solely on past interactions, lacking transferable knowledge to generalize to unseen domains. Recent advances in large language models (LLMs) have motivated text-based recommendation approaches that derive item representations from textual descriptions. While these methods enhance generalization, they fail to encode CF signals-i.e., latent item correlations and preference patterns-crucial for effective recommendation. We argue that an ideal embedding model should seamlessly integrate CF signals with rich semantic representations to improve both in-domain and out-of-domain recommendation performance. To this end, we propose LLM2Rec, a novel embedding model tailored for sequential recommendation, integrating the rich semantic understanding of LLMs with CF awareness. Our approach follows a two-stage training framework: (1) Collaborative Supervised Fine-tuning, which adapts LLMs to infer item relationships based on historical interactions, and (2) Item-level Embedding Modeling, which refines these specialized LLMs into structured item embedding models that encode both semantic and collaborative information. Extensive experiments on real-world datasets demonstrate that LLM2Rec effectively improves recommendation quality across both in-domain and out-of-domain settings. Our findings highlight the potential of leveraging LLMs to build more robust, generalizable embedding models for sequential recommendation. Our codes are available at: https://github.com/HappyPointer/LLM2Rec.

Recent advances in product bundling have leveraged multimodal information through sophisticated encoders, but remain constrained by limited semantic understanding and a narrow scope of knowledge. Therefore, some attempts employ In-context Learning (ICL) to explore the potential of large language models (LLMs) for their extensive knowledge and complex reasoning abilities. However, these efforts are inadequate in understanding mulitmodal data and exploiting LLMs' knowledge for product bundling. To bridge the gap, we introduce Bundle-MLLM, a novel framework that fine-tunes LLMs through a hybrid item tokenization approach within a well-designed optimization strategy. Specifically, we integrate textual, media, and relational data into a unified tokenization, introducing a soft separation token to distinguish between textual and non-textual tokens. Additionally, a streamlined yet powerful multimodal fusion module is employed to embed all non-textual features into a single, informative token, significantly boosting efficiency. To tailor product bundling tasks for LLMs, we reformulate the task as a multiple-choice question with candidate items as options. We further propose a progressive optimization strategy that fine-tunes LLMs for disentangled objectives: learning bundle patterns and enhancing multimodal semantic understanding specific to product bundling. Extensive experiments demonstrate that our approach outperforms a range of state-of-the-art (SOTA) methods. Codes are available at https://github.com/Xiaohao-Liu/Bundle-MLLM

Video Question Answering (Video QA) has emerged as a central task in multimodal learning. This tutorial provides a comprehensive overview of VideoQA research and highlights new frontiers. We begin with an introduction to VideoQA preliminaries, tracing how methods have adapted from third-person view short videos to capture egocentric and long-ranged spatial-temporal dynamics. We then focus on the impact of large multimodal models. Next, we expand the scope to spatial understanding beyond videos. Each topic is discussed through the lens of tasks, datasets, methods, and evaluation protocols. Finally, we conclude with future directions, including fine-grained and long-ranged video understanding, robustness and trustworthiness, Egocentric and embodied assistance, and omnimodal integration. This tutorial aims to equip participants with both a historical perspective and a forward-looking roadmap for advancing Video QA in the LLM era.

This workshop aims to explore the potential of large generative models to revolutionize the way we interact with multimodal information. A Large Language Model (LLM) represents a sophisticated form of artificial intelligence engineered to comprehend and produce natural language text, exemplified by technologies such as GPT, LLaMA, Flan-T5, ChatGLM, and Qwen, etc. These models undergo training on extensive text datasets, exhibiting commendable attributes including robust language generation, zeroshot transfer capabilities, and In-Context Learning (ICL). With the surge in multimodal content-encompassing images, videos, audio, and 3D models-over the recent period, Large MultiModal Models (LMMs) have seen significant enhancements. These improvements enable the augmentation of conventional LLMs to accommodate multimodal inputs or outputs, as seen in BLIP, Flamingo, KOSMOS, LLaVA, Gemini, GPT-4, etc. Concurrently, certain research initiatives have delved into generating specific modalities, with Kosmos2 and MiniGPT-5 focusing on image generation, and SpeechGPT on speech production. There are also endeavors to integrate LLMs with external tools to achieve a near 'any-to-any' multimodal comprehension and generation capacity, illustrated by projects like Visual-ChatGPT, ViperGPT, MMREACT, HuggingGPT, and AudioGPT. Collectively, these models, spanning not only text and image generation but also other modalities, are referred to as large generative models. This workshop will provide an opportunity for researchers, practitioners, and industry professionals to explore the latest trends and best practices in the field of multimodal applications of large generative models. We also remark that the submissions are not limited to the use of such models. The workshop will also focus on exploring the challenges and opportunities of integrating large language models with other AI technologies such as computer vision and speech recognition. Additionally, the workshop will provide a platform for participants to present their research, share their experiences, and discuss potential collaborations.

Long-form video understanding (LVU) addresses the challenge of answering complex questions over extended video length, where informative cues are sparse and easily overwhelmed by redundant content. To tackle this, it requires selecting a small set of question-relevant keyframes and reasoning over long-range, temporally dispersed visual evidence. However, current methods typically extract frame-level features with limited temporal context and store them in sequential memory structures. As a result, they struggle to capture the evolving relations among entities and fail to maintain identity consistency when entities temporarily leave and later reappear in the video. These limitations prevent accurate keyframe localization and coherent reasoning.

AI as a concept has been around since the 1950s. With the recent advancements in machine learning technology, and the availability of big data and large computing resources, the scene is set for the explosive growth of AI. In particular, the emergence of Multimodal Foundation Models that offer significant capabilities in content comprehension, generation and reasoning has opened up opportunities for multimodal research and applications. The talk first reviews the trends and developments in Multimodal Foundation Models. It then outlines the advances in multilingual and multimodal alignments and discusses the emergence of language- and media-agnostic signals that appear to represent abstract concepts commonly used in human languages. These signals have been shown to have a positive impact on both the performance and safety of the resulting models, especially in enhancing those with low-training samples. To further improve the performance of the models, most current approaches employ reinforcement learning with various reward functions to achieve better controllable content generation and trust. To facilitate effective reinforcement learning, quality assessment is of pivotal importance, while it has largely been overlooked. This talk further presents recent approaches to quality assessment and its role in the generation of textual descriptions, videos, and 3D media. As research on Multimodal Foundation Models is still in an early stage, this talk concludes with directions for future research.

Video-language models are pivotal for tasks such as moment retrieval and highlight detection, yet they often struggle to capture the dynamic, non-linear interactions between temporal video sequences and textual semantics. Existing approaches, relying on static cross-attention or prompt-tuning mechanisms, fail to adaptively model the evolving relationships between modalities, leading to suboptimal alignment and limited generalization. Inspired by systems biology, we propose Reaction-Diffusion Multimodal Fusion (RDMF), a novel framework that reimagines video-language alignment as a reaction-diffusion (RD) process, drawing on the principles of pattern formation introduced by Alan Turing. In RDMF, video features diffuse across time to capture temporal context, while text-video interactions are modeled as non-linear reactions that amplify relevant features and suppress noise, forming emergent patterns akin to biological systems. Leveraging the Gray-Scott RD model, we design a computationally efficient fusion module that integrates video and text representations, supported by rigorous mathematical analysis of stability and convergence using Turing instability criteria. Our framework is theoretically grounded, employing advanced mathematical tools to ensure stable pattern formation, and is practically viable, incorporating standard components like pretrained encoders and DETR-style heads for moment retrieval and saliency prediction. RDMF represents a pioneering interdisciplinary approach, bridging systems biology and multimedia research to address the limitations of conventional multimodal fusion. Preliminary experiments demonstrate its potential to outperform existing methods in identifying salient video moments, offering a new paradigm for video-language tasks. Beyond its immediate applications, RDMF opens avenues for exploring biologically inspired architectures in multimedia, with implications for real-time video analysis, interactive media systems, and cross-disciplinary collaboration between multimedia and systems biology communities.

Recent advancements in Multi-modal Large Language Models (MLLMs) have led to significant progress in developing GUI agents for general tasks such as web browsing and mobile phone use. However, their application in professional domains remains under-explored. These specialized workflows introduce unique challenges for GUI perception models, including high-resolution displays and complex environments which lead to smaller target sizes. In this paper, we introduce ScreenSpot-Pro, a new benchmark designed to rigorously evaluate the grounding capabilities of MLLMs in high-resolution professional settings. The benchmark comprises authentic high-resolution images from a variety of professional domains with expert annotations. It spans 23 applications across five industries and three operating systems. Existing GUI grounding models perform poorly on this dataset, with the best model achieving only 18.9%. Our experiments reveal that strategically reducing the search area enhances accuracy. Based on this insight, we propose ScreenSeekeR, a visual search method that utilizes the GUI knowledge of a strong planner to guide a cascaded search, achieving state-of-the-art performance with 48.1% without any additional training. We hope that our benchmark and findings will advance the development of GUI agents for professional settings.

Personalized image generation has emerged as a promising direction in multimodal content creation. It aims to synthesize images tailored to individual style preferences (e.g. color schemes, character appearances, layout) and semantic intentions (e.g. emotion, action, scene contexts) by leveraging user-interacted history images and multimodal instructions. Despite notable progress, existing methods -- whether based on diffusion models, large language models, or Large Multimodal Models (LMMs) -- struggle to accurately capture and composite user style preferences and semantic intentions. In particular, the state-of-the-art LMM-based method suffers from the entanglement of visual features, leading to Guidance Collapse, where the generated images fail to preserve user-preferred styles or reflect the specified semantics.

Finance decision-making often relies on in-depth data analysis across various data sources, including financial tables, news articles, stock prices, etc. In this work, we introduce FINTMMBench, the first comprehensive benchmark for evaluating temporal-aware multi-modal Retrieval-Augmented Generation (RAG) systems in finance. Built from heterologous data of NASDAQ 100 companies, FINTMMBench offers three significant advantages. 1) Multi-modal Corpus: It encompasses a hybrid of financial tables, news articles, daily stock prices, and visual technical charts as the corpus. 2) Temporal-aware Questions: Each question requires the retrieval and interpretation of its relevant data over a specific time period, including daily, weekly, monthly, quarterly, and annual periods. 3) Diverse Financial Analysis Tasks: The questions involve 10 different financial analysis tasks designed by domain experts, including information extraction, trend analysis, sentiment analysis and event detection, etc. We further propose a novel TMMHybridRAG method, which first leverages a multi-modal LLM to convert data from other modalities (e.g., tabular, visual and time-series data) into textual format and then incorporates temporal information in each node when constructing graphs and dense indexes. Its effectiveness has been validated in extensive experiments, but notable gaps remain, highlighting the challenges presented by our FINTMMBench. The benchmark and source code will be made publicly available.

Hao Li, Chenghao Yang, An Zhang, Yang Deng, Xiang Wang, Tat-Seng Chua. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

As large language models (LLMs) continue to advance, their capacity to function effectively across a diverse range of languages has shown marked improvement. Preliminary studies observe that the hidden activations of LLMs often resemble English, even when responding to non-English prompts. This has led to the widespread assumption that LLMs may ``think'' in English. However, more recent results showing strong multilingual performance, even surpassing English performance on specific tasks in other languages, challenge this view. In this work, we find that LLMs progressively develop a core language-agnostic parameter space—a remarkably small subset of parameters whose deactivation results in significant performance degradation across all languages. This compact yet critical set of parameters underlies the model’s ability to generalize beyond individual languages, supporting the emergence of abstract thought that is not tied to any specific linguistic system. Specifically, we identify language-related neurons—those are consistently activated during the processing of particular languages, and categorize them as either shared (active across multiple languages) or exclusive (specific to one). As LLMs undergo continued development over time, we observe a marked increase in both the proportion and functional importance of shared neurons, while exclusive neurons progressively diminish in influence. These shared neurons constitute the backbone of the core language-agnostic parameter space, supporting the emergence of abstract thought. Motivated by these insights, we propose neuron-specific training strategies tailored to LLMs' language-agnostic levels at different development stages. Experiments across diverse LLM families support our approach. Our codes are available at https://anonymous.4open.science/status/S-C393.

Recent advances in Large Vision-Language Models (LVLMs) have showcased strong reasoning abilities across multiple modalities, achieving significant breakthroughs in various real-world applications. Despite this great success, the safety guardrail of LVLMs may not cover the unforeseen domains introduced by the visual modality. Existing studies primarily focus on eliciting LVLMs to generate harmful responses via carefully crafted image-based jailbreaks designed to bypass alignment defenses. In this study, we reveal that a safe image can be exploited to achieve the same jailbreak consequence when combined with additional safe images and prompts. This stems from two fundamental properties of LVLMs: universal reasoning capabilities and safety snowball effect. Building on these insights, we propose Safety Snowball Agent (SSA), a novel agent-based framework leveraging agents' autonomous and tool-using abilities to jailbreak LVLMs. SSA operates through two principal stages: (1) initial response generation, where tools generate or retrieve jailbreak images based on potential harmful intents, and (2) harmful snowballing, where refined subsequent prompts induce progressively harmful outputs. Our experiments demonstrate that SSA can use nearly any image to induce LVLMs to produce unsafe content, achieving high success jailbreaking rates against the latest LVLMs. Unlike prior works that exploit alignment flaws, SSA leverages the inherent properties of LVLMs, presenting a profound challenge for enforcing safety in generative multimodal systems.

The rapid development of Multimodal Large Language Models (MLLMs) poses increasing demands on the diversity and complexity of multimodal datasets. Yet manual annotation pipelines can no longer keep pace. Existing augmentation methods often follow fixed rules and lack verifiable control over sample diversity and reasoning complexity. To address this, we introduce Scalable COunterfactual Program Evolution (SCOPE), a framework that uses symbolic Visual Programming to guide program evolution via counterfactual reasoning. SCOPE performs the three steps of counterfactual inference: (1) Abduction, by generating verifiable programs to model reasoning associations; (2) Action, by intervening on program structure along three axes—reasoning path, visual context, and cross-instance composition; and (3) Prediction, by categorizing evolved instances by difficulty, structure, and input multiplicity. Based on this process, we build SCOPE-Train and SCOPE-Test, evolving benchmarks with expert validation. To support training, we propose MAP, a curriculum learning strategy that aligns model capacity with sample difficulty. Experiments show that SCOPE improves reasoning performance, exposes model blind spots, and enhances visual dialog capabilities.

Large Language Models (LLMs) have shown strong potential for recommendation by framing item prediction as a token-by-token language generation task. However, existing methods treat all item tokens equally, simply pursuing likelihood maximization during both optimization and decoding. This overlooks crucial token-level differences in decisiveness—many tokens contribute little to item discrimination yet can dominate optimization or decoding. To quantify token decisiveness, we propose a novel perspective that models item generation as a decision process, measuring token decisiveness by the Information Gain (IG) each token provides in reducing uncertainty about the generated item. Our empirical analysis reveals that most tokens have low IG but often correspond to high logits, disproportionately influencing training loss and decoding, which may impair model performance. Building on these insights, we introduce an Information Gain-based Decisiveness-aware Token handling (IGD) strategy that integrates token decisiveness into both tuning and decoding. Specifically, IGD downweights low-IG tokens during tuning and rebalances decoding to emphasize tokens with high IG. In this way, IGD moves beyond pure likelihood maximization, effectively prioritizing high-decisiveness tokens. Extensive experiments on four benchmark datasets with two LLM backbones demonstrate that IGD consistently improves recommendation accuracy, achieving significant gains on widely used ranking metrics compared to strong baselines. Our codes are available at \url{https://github.com/ZJLin2oo1/IGD}.

This paper presents JavisGPT, the first unified multimodal large language model (MLLM) for joint audio-video (JAV) comprehension and generation. JavisGPT has a concise encoder-LLM-decoder architecture, which has a SyncFusion module for spatio-temporal audio-video fusion and synchrony-aware learnable queries to bridge a pretrained JAV-DiT generator. This design enables temporally coherent video-audio understanding and generation from multimodal instructions. We design an effective three-stage training pipeline consisting of multimodal pretraining, audio-video fine-tuning, and large-scale instruction-tuning, to progressively build multimodal comprehension and generation from existing vision-language models. For instruction tuning, we construct JavisInst-Omni, a high-quality instruction dataset with over 200K GPT-4o-curated audio-video-text dialogues that cover diverse and multi-level comprehension and generation scenarios. On JAV comprehension and generation benchmarks, our experiments show that JavisGPT outperforms existing MLLMs, particularly in complex and temporally synchronized settings.

Multimodal Contrastive Learning (MCL) advances in aligning different modalities and generating multimodal representations in a joint space. By leveraging contrastive learning across diverse modalities, large-scale multimodal data enhances representational quality. However, a critical yet often overlooked challenge remains: multimodal data is rarely collected in a single process, and training from scratch is computationally expensive. Instead, emergent multimodal data can be used to optimize existing models gradually, \textit{i.e.}, models are trained on a sequence of modality pair data. We define this problem as Continual Multimodal Contrastive Learning (CMCL), an underexplored yet crucial research direction at the intersection of multimodal and continual learning. In this paper, we formulate CMCL through two specialized principles of stability and plasticity. We theoretically derive a novel optimization-based method, which projects updated gradients from dual sides onto subspaces where any gradient is prevented from interfering with the previously learned knowledge. Two upper bounds provide theoretical insights on both stability and plasticity in our solution. Beyond our theoretical contributions, we conduct experiments on multiple datasets by comparing our method against advanced continual learning baselines. The empirical results further support our claims and demonstrate the efficacy of our method. Our codes are available at https://github.com/Xiaohao-Liu/CMCL.

Large language models (LLMs) have achieved notable progress. Despite their success, next-token prediction (NTP), the dominant method for LLM training and inference, is constrained in both contextual coverage and inference efficiency due to its inherently sequential process. To overcome these challenges, we propose leap multi-token prediction~(L-MTP), an innovative token prediction method that extends the capabilities of multi-token prediction (MTP) by introducing a leap-based mechanism. Unlike conventional MTP, which generates multiple tokens at adjacent positions, L-MTP strategically skips over intermediate tokens, predicting non-sequential ones in a single forward pass. This structured leap not only enhances the model's ability to capture long-range dependencies but also enables a decoding strategy specially optimized for non-sequential leap token generation, effectively accelerating inference. We theoretically demonstrate the benefit of L-MTP in improving inference efficiency. Experiments across diverse benchmarks validate its merit in boosting both LLM performance and inference speed. The source code is available at https://github.com/Xiaohao-Liu/L-MTP.

3D molecule generation is crucial for drug discovery and material science, requiring models to process complex multi-modalities, including atom types, chemical bonds, and 3D coordinates. A key challenge is integrating these modalities of different shapes while maintaining SE(3) equivariance for 3D coordinates. To achieve this, existing approaches typically maintain separate latent spaces for invariant and equivariant modalities, reducing efficiency in both training and sampling. In this work, we propose **U**nified Variational **A**uto-**E**ncoder for **3D** Molecular Latent Diffusion Modeling (**UAE-3D**), a multi-modal VAE that compresses 3D molecules into latent sequences from a unified latent space, while maintaining near-zero reconstruction error. This unified latent space eliminates the complexities of handling multi-modality and equivariance when performing latent diffusion modeling. We demonstrate this by employing the Diffusion Transformer--a general-purpose diffusion model without any molecular inductive bias--for latent generation. Extensive experiments on GEOM-Drugs and QM9 datasets demonstrate that our method significantly establishes new benchmarks in both *de novo* and conditional 3D molecule generation, achieving leading efficiency and quality. On GEOM-Drugs, it reduces FCD by 72.6% over the previous best result, while achieving over 70% relative average improvements in geometric fidelity. Our code is released at [https://github.com/lyc0930/UAE-3D/](https://github.com/lyc0930/UAE-3D/).

Recent studies empirically reveal that large reasoning models (LRMs) can automatically allocate more reasoning strengths (\ie the number of reasoning tokens) for harder problems, exhibiting difficulty-awareness for better task performance. While this automatic reasoning strength allocation phenomenon has been widely observed, its underlying mechanism remains largely unexplored. To this end, we provide explanations for this phenomenon from the perspective of model activations. \textbf{We find evidence that LRMs pre-plan the reasoning strengths in their activations even before generation, with this reasoning strength causally controlled by the magnitude of a pre-allocated directional vector.} Specifically, we show that the number of reasoning tokens is predictable solely based on the question activations using linear probes, indicating that LRMs estimate the required reasoning strength in advance. We then uncover that LRMs encode this reasoning strength through a pre-allocated directional vector embedded in the activations of the model, where the vector’s magnitude modulates the reasoning strength. Subtracting this vector can lead to reduced reasoning token number and performance, while adding this vector can lead to increased reasoning token number and even improved performance. We further reveal that this direction vector consistently yields positive reasoning length prediction, and it modifies the logits of end-of-reasoning token \texttt{</think>} to affect the reasoning length. Finally, we demonstrate two potential applications of our findings: overthinking behavior detection and enabling efficient reasoning on simple problems. Our work provides new insights into the internal mechanisms of reasoning in LRMs and offers practical tools for controlling their reasoning behaviors. Our code is available at \url{https://anonymous.4open.science/r/LRM-plans-CoT-7E04}.

Multilingual reasoning remains a significant challenge for large language models (LLMs), with performance disproportionately favoring high-resource languages. Drawing inspiration from cognitive neuroscience, which suggests that human reasoning functions largely independently of language processing, we hypothesize that LLMs similarly encode reasoning and language as separable components that can be disentangled to enhance multilingual reasoning. To evaluate this, we perform a causal intervention by ablating language-specific representations at inference time. Experiments on 10 open-weight LLMs spanning 11 typologically diverse languages show that this language-specific ablation consistently boosts multilingual reasoning performance. Layer-wise analyses further confirm that language and reasoning representations can be effectively disentangled throughout the model, yielding improved multilingual reasoning capabilities, while preserving top-layer language features remains essential for maintaining linguistic fidelity. Compared to post-training methods such as supervised fine-tuning or reinforcement learning, our training-free language-reasoning disentanglement achieves comparable or superior results with minimal computational overhead. These findings shed light on the internal mechanisms underlying multilingual reasoning in LLMs and suggest a lightweight and interpretable strategy for improving cross-lingual generalization.

Large Language Models (LLMs) continue to exhibit vulnerabilities despite deliberate safety alignment efforts, posing significant risks to users and society. To safeguard against the risk of policy-violating content, system-level moderation via external guard models—designed to monitor LLM inputs and outputs and block potentially harmful content—has emerged as a prevalent mitigation strategy. Existing approaches of training guard models rely heavily on extensive human curated datasets and struggle with out-of-distribution threats, such as emerging harmful categories or jailbreak attacks. To address these limitations, we propose RSafe, an adaptive reasoning-based safeguard that conducts guided safety reasoning to provide robust protection within the scope of specified safety policies. RSafe operates in two stages: (1) guided reasoning, where it analyzes safety risks of input content through policy-guided step-by-step reasoning, and (2) reinforced alignment, where rule-based RL optimizes its reasoning paths to align with accurate safety prediction. This two-stage training paradigm enables RSafe to internalize safety principles to generalize safety protection capability over unseen or adversarial safety violation scenarios. During inference, RSafe accepts user-specified safety policies to provide enhanced safeguards tailored to specific safety requirements. Experiments demonstrate that RSafe matches state-of-the-art guard models using limited amount of public data in both prompt- and response-level harmfulness detection, while achieving superior out-of-distribution generalization on both emerging harmful category and jailbreak attacks. Furthermore, RSafe provides human-readable explanations for its safety judgments for better interpretability. RSafe offers a robust, adaptive, and interpretable solution for LLM safety moderation, advancing the development of reliable safeguards in dynamic real-world environments. Our code is available at https://anonymous.4open.science/r/RSafe-996D.

Text-to-image retrieval is a fundamental task in multimedia retrieval.Traditional studies have typically approached this task as a discriminative problem, matching the text and image via the cross-attention mechanism (one-tower framework) or in a common embedding space (two-tower framework).The one-tower framework excels in effectiveness but falls short in efficiency, whereas the two-tower framework is efficient but struggles to maintain competitive effectiveness.In this study, we aim to enhance both effectiveness and efficiency by transforming the text-to-image retrieval task into a token-to-voken generation problem, where fine-grained interactions are incorporated to improve effectiveness while maintaining high efficiency.Despite its potential advantages, this paradigm shift presents significant challenges: 1) misalignment with high-level semantics and 2) learning gap towards the retrieval target.To address the challenges, we propose AVG, which discretizes images into vokens while aligning with both the visual information and high-level semantics.Additionally, to bridge the learning gap between generative training and the retrieval target, AVG incorporates discriminative training to modify the learning direction during token-to-voken training.Experiments demonstrate that the benefits of paradigm innovation are realized: compared with the classical two-tower method, CLIP, AVG achieves the 7.53% relative effectiveness improvement and also 4× efficiency improvement.We release code at the GitHub repository.

Leveraging Large Language Models (LLMs) for generative recommendation has attracted significant research interest, where item tokenization is a critical step. It involves assigning item identifiers for LLMs to encode user history and generate the next item. Existing approaches leverage either token-sequence identifiers, representing items as discrete token sequences, or single-token identifiers, using ID or semantic embeddings. Token-sequence identifiers face issues such as the local optima problem in beam search and low generation efficiency due to step-by-step generation. In contrast, single-token identifiers fail to capture rich semantics or encode Collaborative Filtering (CF) information, resulting in suboptimal performance.

Fairness is an increasingly important factor in re-ranking tasks. Prior work has identified a trade-off between ranking accuracy and item fairness. However, the underlying mechanisms are still not fully understood. An analogy can be drawn between re-ranking and the dynamics of economic transactions. The accuracy-fairness trade-off parallels the coupling of the commodity tax transfer process. Fairness considerations in re-ranking, similar to a commodity tax on suppliers, ultimately translate into a cost passed on to consumers. Analogously, item-side fairness constraints result in a decline in user-side accuracy. In economics, the extent to which commodity tax on the supplier (item fairness) transfers to commodity tax on users (accuracy loss) is formalized using the notion of elasticity. The re-ranking fairness-accuracy trade-off is similarly governed by the elasticity of utility between item groups. This insight underscores the limitations of current fair re-ranking evaluations, which often rely solely on a single fairness metric, hindering comprehensive assessment of fair re-ranking algorithms.

Generative retrieval reformulates retrieval as an autoregressive generation task, where large language models (LLMs) generate target documents directly from a query. As a novel paradigm, the mechanisms that underpin its performance and scalability remain largely unexplored. We systematically investigate training and inference scaling laws in generative retrieval, exploring how model size, training data scale, and inference-time compute jointly influence performance. We propose a novel evaluation metric inspired by contrastive entropy and generation loss, providing a continuous performance signal that enables robust comparisons across diverse generative retrieval methods. Our experiments show that n-gram-based methods align strongly with training and inference scaling laws. We find that increasing model size, training data scale, and inference-time compute all contribute to improved performance, highlighting the complementary roles of these factors in enhancing generative retrieval. Across these settings, LLaMA models consistently outperform T5 models, suggesting a particular advantage for larger decoder-only models in generative retrieval. Our findings underscore that model sizes, data availability, and inference computation interact to unlock the full potential of generative retrieval, offering new insights for designing and optimizing future systems. We release code at SLGR GitHub repository.

Generative AI search driven by large language models (LLMs) is reshaping information retrieval by offering end-to-end answers to complex queries, reducing users' reliance on manually browsing and summarizing multiple web pages. However, while this paradigm enhances convenience, it disrupts the feedback-driven improvement loop that has historically powered the evolution of traditional Web search. Web search can continuously improve their ranking models by collecting large-scale, fine-grained user feedback (e.g., clicks, dwell time) at the document level. In contrast, generative AI search operates through a much longer search pipeline-spanning query decomposition, document retrieval, and answer generation-yet typically receives only coarse-grained feedback on the final answer. This introduces a feedback loop disconnect, where user feedback for the final output cannot be effectively mapped back to specific system components, making it difficult to improve each intermediate stage and sustain the feedback loop.

In the era of vast digital information, the sheer volume and heterogeneity of available information present significant challenges for intricate information seeking. Users frequently face multistep web search tasks that involve navigating vast and varied data sources. This complexity demands every step remains comprehensive, accurate, and relevant. However, traditional search methods often struggle to balance the need for localized precision with the broader context required for holistic understanding, leaving critical facets of intricate queries underexplored. In this paper, we introduce an LLM-based search assistant that adopts a new information seeking paradigm with holistically guided Monte Carlo tree search (HG-MCTS). We reformulate the task as a progressive information collection process with a knowledge memory and unite an adaptive checklist with multi-perspective reward modeling in MCTS. The adaptive checklist provides explicit sub-goals to guide the MCTS process toward comprehensive coverage of complex user queries. Simultaneously, our multi-perspective reward modeling offers both exploration and retrieval rewards, along with progress feedback that tracks completed and remaining sub-goals, refining the checklist as the tree search progresses. By striking a balance between localized tree expansion and global guidance, HG-MCTS reduces redundancy in search paths and ensures that all crucial aspects of an intricate query are properly addressed. Extensive experiments on real-world intricate information seeking tasks demonstrate that HG-MCTS acquires thorough knowledge collections and delivers more accurate final responses compared with existing baselines.

Recent advancements in Generative AI, such as Large Language Models (LLMs), have demonstrated remarkable success across various general tasks. Extensive studies have explored leveraging generative models in finance, but significant challenges persist. This half-day workshop explores potential approaches and research directions to address these challenges by equipping generative models with advanced Information Retrieval (IR) models. Specifically, this workshop seeks to provide a platform for discussing innovative ideas that facilitate the advancement of IR technology to enrich generative models in finance from four key perspectives: (i) financial IR techniques (ii) financial IR benchmarking and evaluation (iii) financial systems and agents/assistants (iv) and trustworthiness, privacy and security when applying financial IR and generative models. This workshop aims to deepen understanding, accelerate progress, and support the advancement of IR technology to enhance generative models to address financial challenges.

Web agents have emerged as a promising direction to automate Web task completion based on user instructions, significantly enhancing user experience. Recently, Web agents have evolved from traditional agents to Large Language Models (LLMs)-based Web agents. Despite their success, existing LLM-based Web agents overlook the importance of personalized data (e.g., user profiles and historical Web behaviors) in assisting the understanding of users' personalized instructions and executing customized actions.

Retrieval-Augmented Generation (RAG) systems have become a crucial tool to augment large language models (LLMs) with external knowledge for better task performance.However, existing traditional RAG methods inject knowledge directly into the context, resulting in several limitations.First, these methods highly rely on the in-context learning capability of LLMs, which often leads to excessively long contexts.This is inefficient due to the quadratic complexity of self-attention, leading to significant increase in inference time.Second, the extended context and the nature of self-attention can cause the LLMs to lose important information in the context, thereby degrading the original capabilities of LLMs.Third, the effectiveness of knowledge injection is perturbed by the permutation of knowledge within the extended context, reducing the robustness of existing RAG methods.To tackle the above problems, we propose DecoupledRAG, a method that decouples external knowledge from the context within the RAG framework.Specifically, we introduce a cross-attention based method that injects retrieved knowledge directly into the inference process of LLM on the fly, without modifying its parameters or the input context, so that the external knowledge can be utilized robustly in a permutation-independent manner.To the best of our knowledge, this is the first work that explore how to utilize cross-attention to inject knowledge with low training cost in decoder-only LLM era.By leveraging cross-attention operation, DecoupledRAG enables seamless knowledge aggregation without creating extended context.Experimental results demonstrate that our method could achieve

Large language models (LLMs), with advanced linguistic capabilities, have been employed in reranking tasks through a sequence-to-sequence approach. In this paradigm, multiple passages are reranked in a listwise manner and a textual reranked permutation is generated. However, due to the limited context window of LLMs, this reranking paradigm requires a sliding window strategy to iteratively handle larger candidate sets. This not only increases computational costs but also restricts the LLM from fully capturing all the comparison information for all candidates. To address these challenges, we propose a novel self-calibrated listwise reranking method, which aims to leverage LLMs to produce global relevance scores for ranking. To achieve it, we first propose the relevance-aware listwise reranking framework, which incorporates explicit list-view relevance scores to improve reranking efficiency and enable global comparison across the entire candidate set. Second, to ensure the comparability of the computed scores, we propose self-calibrated training that uses point-view relevance assessments generated internally by the LLM itself to calibrate the list-view relevance assessments. Extensive experiments and comprehensive analysis on the BEIR benchmark and TREC Deep Learning Tracks demonstrate the effectiveness and efficiency of our proposed method.

Video Question Answering (VideoQA) is the task of answering questions about a video. At its core is the understanding of the alignments between video scenes and question semantics to yield the answer. In leading VideoQA models, the typical learning objective, empirical risk minimization (ERM), tends to over-exploit the spurious correlations between question-irrelevant scenes and answers, instead of inspecting the causal effect of question-critical scenes, which undermines the prediction with unreliable reasoning. In this work, we take a causal look at VideoQA and propose a modal-agnostic learning framework, named Invariant Grounding for VideoQA (IGV), to ground the question-critical scene, whose causal relations with answers are invariant across different interventions on the complement. With IGV, leading VideoQA models are forced to shield the answering from the negative influence of spurious correlations, which significantly improves their reasoning ability. To unleash the potential of this framework, we further provide a Transformer-Empowered Invariant Grounding for VideoQA (TIGV), a substantial instantiation of IGV framework that naturally integrates the idea of invariant grounding into a transformer-style backbone. Experiments on four benchmark datasets validate our design in terms of accuracy, visual explainability, and generalization ability over the leading baselines. Our code is available at https://github.com/yl3800/TIGV.

Generalized zero-shot learning (GZSL) endeavors to identify the unseen categories using knowledge from the seen domain, necessitating the intrinsic interactions between the visual features and attribute semantic features. However, GZSL suffers from insufficient visual-semantic correspondences due to the attribute diversity and instance diversity. Attribute diversity refers to varying semantic granularity in attribute descriptions, ranging from low-level (specific, directly observable) to high-level (abstract, highly generic) characteristics. This diversity challenges the collection of adequate visual cues for attributes under a uni-granularity. Additionally, diverse visual instances corresponding to the same sharing attributes introduce semantic ambiguity, leading to vague visual patterns. To tackle these problems, we propose a multi-granularity progressive semantic-visual mutual adaption (PSVMA+) network, where sufficient visual elements across granularity levels can be gathered to remedy the granularity inconsistency. PSVMA+ explores semantic-visual interactions at different granularity levels, enabling awareness of multi-granularity in both visual and semantic elements. At each granularity level, the dual semantic-visual transformer module (DSVTM) recasts the sharing attributes into instance-centric attributes and aggregates the semantic-related visual regions, thereby learning unambiguous visual features to accommodate various instances. Given the diverse contributions of different granularities, PSVMA+ employs selective cross-granularity learning to leverage knowledge from reliable granularities and adaptively fuses multi-granularity features for comprehensive representations. Experimental results demonstrate that PSVMA+ consistently outperforms state-of-the-art methods.

Graph augmentation is the key component to reveal instance-discriminative features of a graph as its rationale—an interpretation for it—in graph contrastive learning (GCL). Existing rationale-aware augmentation mechanisms in GCL frameworks roughly fall into two categories and suffer from inherent limitations: (1) non-heuristic methods with the guidance of domain knowledge to preserve salient features, which require expensive expertise and lack generality, or (2) heuristic augmentations with a co-trained auxiliary model to identify crucial substructures, which face not only the dilemma between system complexity and transformation diversitybut also the instability stemming from the co-training of two separated sub-models. Inspired by recent studies on transformers, we propose self-attentive rationale-guided GCL (SR-GCL), which integrates rationale generator and encoder together, leverages the self-attention values in transformer module as a natural guidance to delineate semantically informative substructures from both node- and edge-wise perspectives, and contrasts on rationale-aware augmented pairs. On real-world biochemistry datasets, visualization results verify the effectiveness and interpretability of self-attentive rationalization, and the performance on downstream tasks demonstrates the state-of-the-art performance of SR-GCL for graph model pre-training. Codes are available at https://github.com/lsh0520/SR-GCL .

Traffic Accident Anticipation (TAA) in traffic scenes is a challenging problem for achieving zero fatalities in the future. Current approaches typically treat TAA as a supervised learning task needing the laborious annotation of accident occurrence duration. However, the inherent long-tailed, uncertain, and fast-evolving nature of traffic scenes has the problem that real causal parts of accidents are difficult to identify and are easily dominated by data bias, resulting in a background confounding issue. Thus, we propose an Attentive Video Diffusion (AVD) model that synthesizes additional accident video clips by generating the causal part in dashcam videos, i.e., from normal clips to accident clips. AVD aims to generate causal video frames based on accident or accident-free text prompts while preserving the style and content of frames for TAA after video generation. This approach can be trained using datasets collected from various driving scenes without any extra annotations. Additionally, AVD facilitates an Equivariant TAA (EQ-TAA) with an equivariant triple loss for an anchor accident-free video clip, along with the generated pair of contrastivepseudo-normalandpseudo-accidentclips. Extensive experiments have been conducted to evaluate the performance of AVD and EQ-TAA, and competitive performance compared to state-of-the-art methods has been obtained.

Facial expression recognition (FER) is a critical task in multimedia with significant implications across various domains. However, analyzing the causes of facial expressions is essential for accurately recognizing them. Current approaches, such as those based on facial action units (AUs), typically provide AU names and intensities but lack insight into the interactions and relationships between AUs and the overall expression. In this paper, we propose a novel method called ExpLLM, which leverages large language models to generate an accurate chain of thought (CoT) for facial expression recognition. Specifically, we have designed the CoT mechanism from three key perspectives: key observations, overall emotional interpretation, and conclusion. The key observations describe the AU's name, intensity, and associated emotions. The overall emotional interpretation provides an analysis based on multiple AUs and their interactions, identifying the dominant emotions and their relationships. Finally, the conclusion presents the final expression label derived from the preceding analysis. Furthermore, we also introduce the Exp-CoT Engine, designed to construct this expression CoT and generate instruction-description data for training our ExpLLM. Extensive experiments on the RAF-DB and AffectNet datasets demonstrate that ExpLLM outperforms current state-of-the-art FER methods. ExpLLM also surpasses the latest GPT-4o in expression CoT generation, particularly in recognizing micro-expressions where GPT-4o frequently fails.

Acquiring nutrition information and health-related knowledge about food is a common need among individuals. However, using conventional food names as search queries often fails to yield accurate matches to entries within food nutrition knowledge bases (FoodnKB), which frequently utilize scientific or product names. In this study, we present a method for enriching FoodnKB entries with imagery and facilitating visual access to food-related knowledge through image recognition. We start with an official food nutrition database and propose a consensus-based approach using Large Language Models to identify visually discernible and directly edible foods, expanding food synonyms and harnessing diverse web-based food images for comprehensive visual representation. To minimize manual annotation of noisy web images, we introduce a cyclic training-based area under the margin metric (cAUM) approach that effectively distinguishes appropriate images, including rare instances, from noisy ones. Additionally, we design a generic accuracy gap (AccGap) algorithm to automatically estimate the noise ratio of the web-harnessed data. Our integrated cAUM and AccGap method demonstrates superior performance in noise detection and enhancement of image recognition accuracy compared to existing noise-robust frameworks. Furthermore, we successfully apply the visually enriched FoodnKB and food recognition capabilities within a smart nutritionist mobile application.

Dialogue systems are designed to offer human users social support or functional services through natural language interactions. Traditional conversation research has put significant emphasis on a system’s response-ability, including its capacity to understand dialogue context and generate appropriate responses. However, the key element of proactive behavior—a crucial aspect of intelligent conversations—is often overlooked in these studies. Proactivity empowers conversational agents to lead conversations towards achieving pre-defined targets or fulfilling specific goals on the system side. Proactive dialogue systems are equipped with advanced techniques to handle complex tasks, requiring strategic and motivational interactions, thus representing a significant step towards artificial general intelligence. Motivated by the necessity and challenges of building proactive dialogue systems, we provide a comprehensive review of various prominent problems and advanced designs for implementing proactivity into different types of dialogue systems, including open-domain dialogues, task-oriented dialogues, and information-seeking dialogues. We also discuss real-world challenges that require further research attention to meet application needs in the future, such as proactivity in dialogue systems that are based on large language models, proactivity in hybrid dialogues, evaluation protocols and ethical considerations for proactive dialogue systems. By providing a quick access and overall picture of the proactive dialogue systems domain, we aim to inspire new research directions and stimulate further advancements towards achieving the next level of conversational AI capabilities, paving the way for more dynamic and intelligent interactions within various application domains.

Interactive Intelligent Tutoring Systems (ITSs) enhance the learning experience in online education by fostering effective learning through interactive problem-solving. However, many current ITS models do not fully incorporate proactive engagement strategies that optimize educational resources through thoughtful planning and assessment. In this work, we propose a novel and practical task of Goal-oriented Intelligent Tutoring Systems (GITS), designed to help students achieve proficiency in specific concepts through a tailored sequence of exercises and evaluations. We introduce a novel graph-based reinforcement learning framework, named Planning-Assessment-Interaction ( PAI ), to tackle the challenges of goal-oriented policy learning within GITS. This framework utilizes cognitive structure information to refine state representation and guide the selection of subsequent actions, whether that involves presenting an exercise or conducting an assessment. Additionally, PAI employs a cognitive diagnosis model that dynamically updates to predict student reactions to exercises and assessments. We construct three benchmark datasets covering different subjects to facilitate offline GITS research. Experimental results validate PAI ’s effectiveness and efficiency, and we present comprehensive analyses of its performance with different student types, highlighting the unique challenges presented by this task.

Dialogue disentanglement aims to detach the chronologically ordered utterances into several independent sessions. Conversation utterances are essentially organized and described by the underlying discourse, and thus dialogue disentanglement requires the full understanding and harnessing of the intrinsic discourse attribute. In this article, we propose enhancing dialogue disentanglement by taking full advantage of the dialogue discourse characteristics. First of all, in feature encoding stage , we construct the heterogeneous graph representations to model the various dialogue-specific discourse structural features, including the static speaker-role structures (i.e., speaker-utterance and speaker-mentioning structure) and the dynamic contextual structures (i.e., the utterance-distance and partial-replying structure). We then develop a structure-aware framework to integrate the rich structural features for better modeling the conversational semantic context. Second, in model learning stage , we perform optimization with a hierarchical ranking loss mechanism, which groups dialogue utterances into different discourse levels and carries training covering pairwise and session-wise levels hierarchically. Third, in inference stage , we devise an easy-first decoding algorithm, which performs utterance pairing under the easy-to-hard manner with a global context, breaking the constraint of traditional sequential decoding order. On two benchmark datasets, our overall system achieves new state-of-the-art performances on all evaluations. In-depth analyses further demonstrate the efficacy of each proposed idea and also reveal how our methods help advance the task. Our work has great potential to facilitate broader multi-party multi-thread dialogue applications.

Textual response generation is a pivotal yet challenging task for multimodal task-oriented dialog systems, which targets at generating the appropriate textual response given the multimodal context. Although existing efforts have obtained remarkable advancements, they ignore the potential of the domain information in revealing the key points of the user intention and the user’s history dialogs in indicating the user’s characteristics. To address this issue, in this work, we propose a novel domain-aware multimodal dialog system with distribution-based user characteristic modeling (named DMDU). In particular, DMDU contains three vital components: context-knowledge embedding extraction , domain-aware response generation , and distribution-based user characteristic injection . Specifically, the context-knowledge embedding extraction component aims to extract the embedding of multimodal context and related knowledge following existing studies. The domain-aware response generation component targets at conducting domain-aware fine-grained intention modeling based on the context and knowledge embedding, and thus fulfills the textual response generation. Moreover, the distribution-based user characteristic injection component first captures the user’s characteristics and current intention with the Gaussian distribution and then conducts the sampling-based contrastive semantic regularization to promote the context representation learning. Experimental results on the public dataset demonstrate the effectiveness of DMDU. We release codes to promote other researchers.

The multimodal emotion-cause pair extraction (MECPE) task aims to detect the emotions, causes, and emotion-cause pairs from multimodal conversations. Existing methods for this task typically concatenate representations of each utterance from distinct modalities and then predict emotion-cause pairs directly. This approach struggles to effectively integrate multimodal features and capture the subtleties of emotion transitions, which are crucial for accurately identifying causes—thereby limiting overall performance. To address these challenges, we propose a novel model that captures holistic interaction and label constraint (HiLo) features for the MECPE task. HiLo facilitates cross-modality and cross-utterance feature interaction with various attention mechanisms, establishing a robust foundation for precise cause extraction. Notably, our model innovatively leverages emotion transition features as pivotal cues to enhance causal inference within conversations. The experimental results demonstrate the superior performance of HiLo, evidenced by an increase of more than 2% in the F1 score compared to existing benchmarks. Further analysis reveals that our approach adeptly utilizes multimodal and dialogue features, making a significant contribution to the field of emotion-cause analysis. Our code is publicly available at https://is.gd/MVdYmx .

Collaborative Filtering (CF) recommender models highly depend on user-item interactions to learn CF representations, thus falling short of recommending cold-start items. To address this issue, prior studies mainly introduce item features (e.g., thumbnails) for cold-start item recommendation. They learn a feature extractor on warm-start items to align feature representations with interactions, and then leverage the feature extractor to extract the feature representations of cold-start items for interaction prediction. Unfortunately, the features of cold-start items, especially the popular ones, tend to diverge from those of warm-start ones due to temporal feature shifts, preventing the feature extractor from accurately learning feature representations of cold-start items. To alleviate the impact of temporal feature shifts, we consider using Distributionally Robust Optimization (DRO) to enhance the generation ability of the feature extractor. Nonetheless, existing DRO methods face an inconsistency issue: the worse-case warm-start items emphasized during DRO training might not align well with the cold-start item distribution. To capture the temporal feature shifts and combat this inconsistency issue, we propose a novel temporal DRO with new optimization objectives, namely, 1) to integrate a worst-case factor to improve the worst-case performance, and 2) to devise a shifting factor to capture the shifting trend of item features and enhance the optimization of the potentially popular groups in cold-start items. Substantial experiments on three real-world datasets validate the superiority of our temporal DRO in enhancing the generalization ability of cold-start recommender models.

In recent years, spectral graph neural networks, characterized by polynomial filters, have garnered increasing attention and have achieved remarkable performance in tasks such as node classification. These models typically assume that eigenvalues for the normalized Laplacian matrix are distinct from each other, thus expecting a polynomial filter to have a high fitting ability. However, this paper empirically observes that normalized Laplacian matrices frequently possess repeated eigenvalues. Moreover, we theoretically establish that the number of distinguishable eigenvalues plays a pivotal role in determining the expressive power of spectral graph neural networks. In light of this observation, we propose an eigenvalue correction strategy that can free polynomial filters from the constraints of repeated eigenvalue inputs. Concretely, the proposed eigenvalue correction strategy enhances the uniform distribution of eigenvalues, thus mitigating repeated eigenvalues, and improving the fitting capacity and expressive power of polynomial filters. Extensive experimental results on both synthetic and real-world datasets demonstrate the superiority of our method.

Graph neural networks (GNNs) have found widespread application in modeling graph data across diverse domains. While GNNs excel in scenarios where the testing data shares the distribution of their training counterparts (in distribution, ID), they often exhibit incorrect predictions when confronted with samples from an unfamiliar distribution (out-of-distribution, OOD). To identify and reject OOD samples with GNNs, recent studies have explored graph OOD detection, often focusing on training a specific model or modifying the data on top of a well-trained GNN. Despite their effectiveness, these methods come with heavy training resources and costs, as they need to optimize the GNN-based models on training data. Moreover, their reliance on modifying the original GNNs and accessing training data further restricts their universality. To this end, this paper introduces a method to detect Graph Out-of-Distribution At Test-time (namely GOODAT), a data-centric, unsupervised, and plug-and-play solution that operates independently of training data and modifications of GNN architecture. With a lightweight graph masker, GOODAT can learn informative subgraphs from test samples, enabling the capture of distinct graph patterns between OOD and ID samples. To optimize the graph masker, we meticulously design three unsupervised objective functions based on the graph information bottleneck principle, motivating the masker to capture compact yet informative subgraphs for OOD detection. Comprehensive evaluations confirm that our GOODAT method outperforms state-of-the-art benchmarks across a variety of real-world datasets.

Web agents powered by Large Language Models (LLMs) have demonstrated remarkable abilities in planning and executing multi-step interactions within complex web-based environments, fulfilling a wide range of web navigation tasks.Despite these advancements, the potential for LLM-powered agents to effectively engage with sequential user instructions in real-world scenarios has not been fully explored.In this work, we introduce a new task of Conversational Web Navigation, which necessitates sophisticated interactions that span multiple turns with both the users and the environment, supported by a specially developed dataset named Multi-Turn Mind2Web (MT-Mind2Web).To tackle the limited context length of LLMs and the context-dependency issue of the conversational tasks, we further propose a novel framework, named selfreflective memory-augmented planning (Self-MAP), which employs memory utilization and self-reflection techniques.Extensive experiments are conducted to benchmark the MT-Mind2Web dataset, and validate the effectiveness of the proposed method. 1

The recent advancements in generative language models have demonstrated their ability to memorize knowledge from documents and recall knowledge to respond to user queries effectively.Building upon this capability, we propose to enable multimodal large language models (MLLMs) to memorize and recall images within their parameters.Given a user query for visual content, the MLLM is anticipated to "recall" the relevant image from its parameters as the response.Achieving this target presents notable challenges, including inbuilt visual memory and visual recall schemes within MLLMs.To address these challenges, we introduce a generative cross-modal retrieval framework, which assigns unique identifier strings to represent images and involves two training steps: learning to memorize and learning to retrieve.The first step focuses on training the MLLM to memorize the association between images and their respective identifiers.The latter step teaches the MLLM to generate the corresponding identifier of the target image, given the textual query input.By memorizing images in MLLMs, we introduce a new paradigm to crossmodal retrieval, distinct from previous discriminative approaches.The experiments demonstrate that the generative paradigm performs effectively and efficiently even with large-scale image candidate sets.The code is released at https://github.com/liyongqi67/GRACE.

Zhiyuan Liu, An Zhang, Hao Fei, Enzhi Zhang, Xiang Wang, Kenji Kawaguchi, Tat-Seng Chua. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Multiple-choice questions (MCQs) are important in enhancing concept learning and student engagement for educational purposes.Despite the multimodal nature of educational content, current methods focus mainly on text-based inputs and often neglect the integration of visual information.In this work, we study the problem of multimodal educational question generation, which aims at generating subjectspecific educational questions with plausible yet incorrect distractors based on multimodal educational content.To tackle this problem, we introduce a novel framework, named Chainof-Exemplar (CoE), which utilizes multimodal large language models (MLLMs) with Chainof-Thought reasoning to improve the generation of challenging distractors.Furthermore, CoE leverages three-stage contextualized exemplar retrieval to retrieve exemplary questions as guides for generating more subject-specific educational questions.Experimental results on the ScienceQA benchmark demonstrate the superiority of CoE in both question generation and distractor generation over existing methods across various subjects and educational levels.

The digital landscape is rapidly evolving with an ever-increasing volume of online news, emphasizing the need for swift and precise analysis of complex events.We refer to the complex events composed of many news articles over an extended period as Temporal Complex Event (TCE).This paper proposes a novel approach using Large Language Models (LLMs) to systematically extract and analyze the event chain within TCE, characterized by their key points and timestamps.We establish a benchmark, named TCELongBench, to evaluate the proficiency of LLMs in handling temporal dynamics and understanding extensive text.This benchmark encompasses three distinct tasks -reading comprehension, temporal sequencing, and future event forecasting.In the experiment, we leverage retrieval-augmented generation (RAG) method and LLMs with long context window to deal with lengthy news articles of TCE.Our findings indicate that models with suitable retrievers exhibit comparable performance with those utilizing long context window.

Tong Zhang, Peixin Qin, Yang Deng, Chen Huang, Wenqiang Lei, Junhong Liu, Dingnan Jin, Hongru Liang, Tat-Seng Chua. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Utilizing powerful Large Language Models (LLMs) for generative recommendation has attracted much attention. Nevertheless, a crucial challenge is transforming recommendation data into the language space of LLMs through effective item tokenization. Current approaches, such as ID, textual, and codebook-based identifiers, exhibit shortcomings in encoding semantic information, incorporating collaborative signals, or handling code assignment bias. To address these limitations, we propose LETTER (a LEarnable Tokenizer for generaTivE Recommendation), which integrates hierarchical semantics, collaborative signals, and code assignment diversity to satisfy the essential requirements of identifiers. LETTER incorporates Residual Quantized VAE for semantic regularization, a contrastive alignment loss for collaborative regularization, and a diversity loss to mitigate code assignment bias. We instantiate LETTER on two models and propose a ranking-guided generation loss to augment their ranking ability theoretically. Experiments on three datasets validate the superiority of LETTER, advancing the state-of-the-art in the field of LLM-based generative recommendation.

Table-text document (e.g., financial reports) understanding has attracted increasing attention in recent two years. TAT-DQA is a realistic setting for the understanding of visually-rich table-text documents, which involves answering associated questions requiring discrete reasoning. Most existing work relies on token-level semantics, falling short in the reasoning across document elements such as quantities and dates. To address this limitation, we propose a novel Doc2SoarGraph model that exploits element-level semantics and employs Semantic-oriented hierarchical Graph structures to capture the differences and correlations among different elements within the given document and question. Extensive experiments on the TAT-DQA dataset reveal that our model surpasses the state-of-the-art conventional method (i.e., MHST) and large language model (i.e., ChatGPT) by 17.73 and 6.49 points respectively in terms of Exact Match (EM) metric, demonstrating exceptional effectiveness.

Text-to-video (T2V) synthesis has gained increasing attention in the community, in which the recently emerged diffusion models (DMs) have promisingly shown stronger performance than the past approaches. While existing state-of-the-art DMs are competent to achieve high-resolution video generation, they may largely suffer from key limitations (e.g., action occurrence disorders, crude video motions) with respect to the intricate temporal dynamics modeling, one of the crux of video synthesis. In this work, we investigate strengthening the awareness of video dynamics for DMs, for high-quality T2V generation. Inspired by human intuition, we design an innovative dynamic scene manager (dubbed as Dysen) module, which includes (step-1) extracting from input text the key actions with proper time-order arrangement, (step-2) transforming the action schedules into the dynamic scene graph (DSG) representations, and (step-3) enriching the scenes in the DSG with sufficient and reasonable details. Taking advantage of the existing powerful LLMs (e.g., ChatGPT) via in-context learning, Dysen realizes (nearly) human-level temporal dynamics understanding. Finally, the resulting video DSG with rich action scene details is encoded as fine-grained spatio-temporal features, integrated into the backbone T2V DM for video generating. Experiments on popular T2V datasets suggest that our Dysen- VDM consistently outperforms prior arts with significant margins, especially in scenarios with complex actions. Codes at http://haofei.vip/Dysen-VDM/.

Segmenting affordance in 3D data is key for bridging perception and action in robots. Existing efforts mostly focus on the visual side and overlook the affordance knowledge from a semantic aspect. This oversight not only limits their generalization to unseen objects, but more importantly, hinders their synergy with large language models (LLMs) which are excellent task planners that can decompose an overarching command into agent-actionable instructions. With this regard, we propose a novel task, Language-guided Affordance Segmentation on 3D Object (LASO), which challenges a model to segment a 3D object's part relevant to a given affordance question. To facilitate the task, we contribute a dataset comprising 19,751 point-question pairs, covering 8434 object shapes and 870 expert-crafted questions. As a pioneer solution, we further propose PointRefer, which highlights an adaptive fusion module to identify target affordance regions at different scales. To ensure a text-aware segmentation, we adopt a set of affordance queries conditioned on linguistic cues to generate dynamic kernels. These kernels are further used to convolute with point features and generate a segmentation mask. Comprehensive experiments and analyses validate PointRefer's effectiveness. With these efforts, We hope that LASO can steer the direction of 3D affordance, guiding it towards enhanced integration with the evolving capabilities of LLMs. Code and data are available at https://github.com/yl3800/LASO.

We present MM-AU, a novel dataset for Multi-Modal Accident video Understanding. MM-AU contains 11,727 in-the-wild ego-view accident videos, each with temporally aligned text descriptions. We annotate over 2.23 mil-lion object boxes and 58,650 pairs of video-based accident reasons, covering 58 accident categories. MM-AU supports various accident understanding tasks, particularly multimodal video diffusion to understand accident cause-effect chains for safe driving. With MM-AU, we present an Abductive accident Video unders tanding framework for Safe Driving perception (AdVersa-SD). AdVersa-SD performs video diffusion via an Object-Centric Video Diffusion (OAVD) method which is driven by an abductive CLIP model. This model involves a contrastive interaction loss to learn the pair co-occurrence of normal, near-accident, accident frames with the corresponding text descriptions, such as accident reasons, prevention advice, and accident categories. OAVD enforces the object region learning while fixing the content of the original frame background in video generation, to find the dominant objects for certain accidents. Extensive experiments verify the abductive ability of AdVersa-SD and the superiority of OAVD against the state-of-the-art diffusion models. Additionally, we provide care-ful benchmark evaluations for object detection and accident reason answering since AdVersa-SD relies on precise object and accident reason information.

Despite advancements in text-to-image generation (T2I), prior methods often face text-image misalignment problems such as relation confusion in generated images. Existing solutions involve cross-attention manipulation for better compositional understanding or integrating large language models for improved layout planning. However, the inherent alignment capabilities of T2I models are still inadequate. By reviewing the link between generative and discriminative modeling, we posit that T2I models' discriminative abilities may reflect their text-image alignment proficiency during generation. In this light, we advocate bolstering the discriminative abilities of T2I models to achieve more precise text-to-image alignment for generation. We present a discriminative adapter built on T2I models to probe their discriminative abilities on two representative tasks and leverage discriminative fine-tuning to improve their text-image alignment. As a bonus of the discriminative adapter, a self-correction mechanism can leverage discriminative gradients to better align generated images to text prompts during inference. Comprehensive evaluations across three benchmark datasets, including both in-distribution and out-of-distribution scenarios, demonstrate our method's superior generation performance. Meanwhile, it achieves state-of-the-art discriminative performance on the two discriminative tasks compared to other generative models. The code is available at https://dpt-t2i.github.io/.

We study visually grounded VideoQA in response to the emerging trends of utilizing pretraining techniques for video-language understanding. Specifically, by forcing vision-language models (VLMs) to answer questions and simultaneously provide visual evidence, we seek to ascertain the extent to which the predictions of such techniques are genuinely anchored in relevant video content, versus spurious correlations from language or irrelevant visual context. Towards this, we construct NExT-GQA - an extension of NExT-QA with 10.5K temporal grounding (or location) labels tied to the original QA pairs. With NExT-GQA, we scrutinize a series of state-of-the-art VLMs. Through post-hoc attention analysis, we find that these models are extremely weak in substantiating the answers despite their strong QA performance. This exposes the limitation of current VLMs in making reliable predictions. As a remedy, we further explore and propose a grounded-QA method via Gaussian mask optimization and cross-modal learning. Experiments with different backbones demonstrate that this grounding mechanism improves both grounding and QA. With these efforts, we aim to push towards trustworthy VLMs in VQA systems. Our dataset and code are available at https://github.com/doc-doc/NExT-GQA.

The Graph Transformer (GT) has shown significant ability in processing graph-structured data, addressing limitations in graph neural networks, such as over-smoothing and over-squashing. However, the implementation of GT in real-world heterogeneous graphs (HGs) with complex topology continues to present numerous challenges. Firstly, a challenge arises in designing a tokenizer that is compatible with heterogeneity. Secondly, the complexity of the transformer hampers the acquisition of high-order neighbor information in HGs. In this paper, we propose a novel Hop-based Heterogeneous Graph Transformer (H2Gormer) framework, paving a promising path for HGs to benefit from the capabilities of Transformers. We propose a Heterogeneous Hop-based Token Generation module to obtain high-order information in a flexible way. Specifically, to enrich the fine-grained heterogeneous semantics of each token, we propose a tailored multi-relational encoder to encode the hop-based neighbors. In this way, the resulting token embeddings are input to the Hop-based Transformer to obtain node representations, which are then combined with position embeddings to obtain the final encoding. Extensive experiments on four datasets are conducted to demonstrate the effectiveness of H2Gormer.

Despite the remarkable abilities of Large Language Models (LLMs) to answer questions, they often display a considerable level of overconfidence even when the question does not have a definitive answer.To avoid providing hallucinated answers to these unknown questions, existing studies typically investigate approaches to refusing to answer these questions.In this work, we propose a novel and scalable self-alignment method to utilize the LLM itself to enhance its response-ability to different types of unknown questions, being capable of not just refusing to answer but further proactively providing explanations to the unanswerability of unknown questions.Specifically, the Self-Align method first employ a two-stage classaware self-augmentation approach to generate a large amount of unknown question-response data.Then we conduct disparity-driven selfcuration to select qualified data for fine-tuning the LLM itself for aligning the responses to unknown questions as desired.Experimental results on two datasets across four types of unknown questions validate the superiority of the Self-Aligned method over existing baselines in terms of three types of task formulation. 1 * Equal contribution.Q: What animal can be found at the top of the men's Wimbledon trophy? Direct AnswerA: The animal that can be found at the top of the men's Wimbledon trophy is a falcon.

We investigate non-collaborative dialogue agents, which are expected to engage in strategic conversations with diverse users, for securing a mutual agreement that leans favorably towards the system's objectives.This poses two main challenges for existing dialogue agents:1) The inability to integrate user-specific characteristics into the strategic planning, and 2) The difficulty of training strategic planners that can be generalized to diverse users.To address these challenges, we propose TRIP to enhance the capability in tailored strategic planning, incorporating a user-aware strategic planning module and a population-based training paradigm.Through experiments on benchmark non-collaborative dialogue tasks, we demonstrate the effectiveness of TRIP in catering to diverse users.

Proactive dialogues serve as a practical yet challenging dialogue problem in the era of large language models (LLMs), where the dialogue policy planning is the key to improving the proactivity of LLMs. Most existing studies enable the dialogue policy planning of LLMs using various prompting schemes or iteratively enhance this capability in handling the given case with verbal AI feedback. However, these approaches are either bounded by the policy planning capability of the frozen LLMs or hard to be transferred to new cases. In this work, we introduce a new dialogue policy planning paradigm to strategize LLMs for proactive dialogue problems with a tunable language model plug-in as a plug-and-play dialogue policy planner, named PPDPP. Specifically, we develop a novel training framework to facilitate supervised fine-tuning over available human-annotated data as well as reinforcement learning from goal-oriented AI feedback with dynamic interaction data collected by the LLM-based self-play simulation. In this manner, the LLM-powered dialogue agent can not only be generalized to different cases after the training, but also be applicable to different applications by just substituting the learned plug-in. In addition, we propose to evaluate the policy planning capability of dialogue systems under the interactive setting. Experimental results demonstrate that PPDPP consistently and substantially outperforms existing approaches on three different proactive dialogue applications, including negotiation, emotional support, and tutoring dialogues.

Recent advancements in Multimodal Large Language Models (MLLMs) have been utilizing Visual Prompt Generators (VPGs) to convert visual features into tokens that LLMs can recognize. This is achieved by training the VPGs on millions of image-caption pairs, where the VPG-generated tokens of images are fed into a frozen LLM to generate the corresponding captions. However, this image-captioning based training objective inherently biases the VPG to concentrate solely on the primary visual contents sufficient for caption generation, often neglecting other visual details. This shortcoming results in MLLMs’ underperformance in comprehending demonstrative instructions consisting of multiple, interleaved, and multimodal instructions that demonstrate the required context to complete a task. To address this issue, we introduce a generic and lightweight Visual Prompt Generator Complete module (VPG-C), which can infer and complete the missing details essential for comprehending demonstrative instructions. Further, we propose a synthetic discriminative training strategy to fine-tune VPG-C, eliminating the need for supervised demonstrative instructions. As for evaluation, we build DEMON, a comprehensive benchmark for demonstrative instruction understanding. Synthetically trained with the proposed strategy, VPG-C achieves significantly stronger zero-shot performance across all tasks of DEMON. Further evaluation on the MME and OwlEval benchmarks also demonstrate the superiority of VPG-C. The code and models are available at https://github.com/DCDmllm/Cheetah.

We investigate composed image retrieval with text feedback. Users gradually look for the target of interest by moving from coarse to fine-grained feedback. However, existing methods merely focus on the latter, i.e., fine-grained search, by harnessing positive and negative pairs during training. This pair-based paradigm only considers the one-to-one distance between a pair of specific points, which is not aligned with the one-to-many coarse-grained retrieval process and compromises the recall rate. In an attempt to fill this gap, we introduce a unified learning approach to simultaneously modeling the coarse- and fine-grained retrieval by considering the multi-grained uncertainty. The key idea underpinning the proposed method is to integrate fine- and coarse-grained retrieval as matching data points with small and large fluctuations, respectively. Specifically, our method contains two modules: uncertainty modeling and uncertainty regularization. (1) The uncertainty modeling simulates the multi-grained queries by introducing identically distributed fluctuations in the feature space. (2) Based on the uncertainty modeling, we further introduce uncertainty regularization to adapt the matching objective according to the fluctuation range. Compared with existing methods, the proposed strategy explicitly prevents the model from pushing away potential candidates in the early stage and thus improves the recall rate. On the three public datasets, i.e., FashionIQ, Fashion200k, and Shoes, the proposed method has achieved +4.03%, + 3.38%, and + 2.40% Recall@50 accuracy over a strong baseline, respectively.

Language Models (LMs) have greatly influenced diverse domains. However, their inherent limitation in comprehending 3D molecular structures has considerably constrained their potential in the biomolecular domain. To bridge this gap, we focus on 3D molecule-text interpretation, and propose 3D-MoLM: 3D-Molecular Language Modeling. Specifically, 3D-MoLM enables an LM to interpret and analyze 3D molecules by equipping the LM with a 3D molecular encoder. This integration is achieved by a 3D molecule-text projector, bridging the 3D molecular encoder’s representation space and the LM’s input space. Moreover, to enhance 3DMoLM’s ability of cross-modal molecular understanding and instruction following, we meticulously curated a 3D molecule-centric instruction tuning dataset – 3D-MoIT. Through 3D molecule-text alignment and 3D molecule-centric instruction tuning, 3D-MoLM establishes an integration of 3D molecular encoder and LM. It significantly surpasses existing baselines on downstream tasks, including moleculetext retrieval, molecule captioning, and more challenging open-text molecular QA tasks, especially focusing on 3D-dependent properties. We will release our codes and datasets at https://github.com/lsh0520/3D-MoLM.

We introduce InstructVid2Vid, an end-to-end diffusion-based methodology for video editing guided by human language instructions. Our approach empowers video manipulation guided by natural language directives, eliminating the need for per-example fine-tuning or inversion. The proposed InstructVid2Vid model modifies a pretrained image generation model, Stable Diffusion, to generate a time-dependent sequence of video frames. By harnessing the collective intelligence of disparate models, we engineer a training dataset rich in video-instruction triplets, which is a more cost-efficient alternative to collecting data in real-world scenarios. To enhance the coherence between successive frames within the generated videos, we propose the Inter-Frames Consistency Loss and incorporate it during the training process. With multimodal classifier-free guidance during the inference stage, the generated videos is able to resonate with both the input video and the accompanying instructions. Experimental results demonstrate that InstructVid2Vid is capable of generating high-quality, temporally coherent videos and performing diverse edits, including attribute editing, background changes, and style transfer. These results underscore the versatility and effectiveness of our proposed method.

For multimodal LLMs, the synergy of visual comprehension (textual output) and generation (visual output) presents an ongoing challenge. This is due to a conflicting objective: for comprehension, an MLLM needs to abstract the visuals; for generation, it needs to preserve the visuals as much as possible. Thus, the objective is a dilemma for visual-tokens. To resolve the conflict, we propose encoding images into morph-tokens to serve a dual purpose: for comprehension, they act as visual prompts instructing MLLM to generate texts; for generation, they take on a different, non-conflicting role as complete visual-tokens for image reconstruction, where the missing visual cues are recovered by the MLLM. Extensive experiments show that morph-tokens can achieve a new SOTA for multimodal comprehension and generation simultaneously. Our project is available at https://github.com/DCDmllm/MorphTokens.

Large Language Models (LLMs) demonstrate remarkable proficiency in comprehending and handling text-based tasks. Many efforts are being made to transfer these attributes to video modality, which are termed Video-LLMs. However, existing Video-LLMs can only capture the coarse-grained semantics and are unable to effectively handle tasks related to comprehension or localization of specific video segments. In light of these challenges, we propose Momentor, a Video-LLM capable of accomplishing fine-grained temporal understanding tasks. To support the training of Momentor, we design an automatic data generation engine to construct Moment-10M, a large-scale video instruction dataset with segment-level instruction data. We train Momentor on Moment-10M, enabling it to perform segment-level reasoning and localization. Zero-shot evaluations on several tasks demonstrate that Momentor excels in fine-grained temporally grounded comprehension and localization.

While recently Multimodal Large Language Models (MM-LLMs) have made exciting strides, they mostly fall prey to the limitation of only input-side multimodal understanding, without the ability to produce content in multiple modalities. As we humans always perceive the world and communicate with people through various modalities, developing any-to-any MM-LLMs capable of accepting and delivering content in any modality becomes essential to human-level AI. To fill the gap, we present an end-to-end general-purpose any-to-any MM-LLM system, NExT-GPT. We connect an LLM with multimodal adaptors and different diffusion decoders, enabling NExT-GPT to perceive inputs and generate outputs in arbitrary combinations of text, image, video, and audio. By leveraging the existing well-trained high-performing encoders and decoders, NExT-GPT is tuned with only a small amount of parameter (1%) of certain projection layers, which not only benefits low-cost training but also facilitates convenient expansion to more potential modalities. Moreover, we introduce a modality-switching instruction tuning (MosIT) and manually curate a high-quality dataset for MosIT, based on which NExT-GPT is empowered with complex cross-modal semantic understanding and content generation. Overall, our research showcases the promising possibility of building a unified AI agent capable of modeling universal modalities, paving the way for more human-like AI research in the community.

The development of large language models (LLMs) has greatly advanced the field of multimodal understanding, leading to the emergence of large multimodal models (LMMs). In order to enhance visual comprehension, recent studies have equipped LMMs with region-level understanding capabilities by representing object bounding box coordinates as a series of text sequences (pix2seq). In this paper, we introduce a novel paradigm for object location modeling called the pix2emb method, where we ask the LMM to output the location embeddings and then decode them with different decoders. This paradigm allows us to use different location formats (such as bounding boxes and masks) in multimodal conversations. Leveraging the proposed pix2emb method, we train an LMM named NExT-Chat and demonstrate its capability of handling multiple tasks like visual grounding, region captioning, and grounded reasoning. Comprehensive experiments show the effectiveness of our NExT-Chat on various tasks, e.g., NExT-Chat (87.7) vs. Shikra (86.9) on POPE-Random, NExT-Chat (71.3) vs. LISA (67.9) on referring expression segmentation task, and NExT-Chat (79.6) vs. Kosmos-2 (62.3) on region caption task.

Android malware attacks have posed a severe threat to mobile users, necessitating a significant demand for the automated detection system. Among the various tools employed in malware detection, graph representations (e.g., function call graphs) have played a pivotal role in characterizing the behaviors of Android apps. However, though achieving impressive performance in malware detection, current state-of-the-art graph-based malware detectors are vulnerable to adversarial examples. These adversarial examples are meticulously crafted by introducing specific perturbations to normal malicious inputs. To defend against adversarial attacks, existing defensive mechanisms are typically supplementary additions to detectors and exhibit significant limitations, often relying on prior knowledge of adversarial examples and failing to defend against unseen types of attacks effectively. In this paper, we propose MASKDROID, a powerful detector with a strong discriminative ability to identify malware and remarkable robustness against adversarial attacks. Specifically, we introduce a masking mechanism into the Graph Neural Network (GNN) based framework, forcing MASKDROID to recover the whole input graph using a small portion (e.g., 20%) of randomly selected nodes.This strategy enables the model to understand the malicious semantics and learn more stable representations, enhancing its robustness against adversarial attacks. While capturing stable malicious semantics in the form of dependencies inside the graph structures, we further employ a contrastive module to encourage MASKDROID to learn more compact representations for both the benign and malicious classes to boost its discriminative power in detecting malware from benign apps and adversarial examples.

As one of the most advanced techniques in AI, Retrieval-Augmented Generation (RAG) can offer reliable and up-to-date external knowledge, providing huge convenience for numerous tasks. Particularly in the era of AI-Generated Content (AIGC), the powerful capacity of retrieval in providing additional knowledge enables RAG to assist existing generative AI in producing high-quality outputs. Recently, Large Language Models (LLMs) have demonstrated revolutionary abilities in language understanding and generation, while still facing inherent limitations such as hallucinations and out-of-date internal knowledge. Given the powerful abilities of RAG in providing the latest and helpful auxiliary information, Retrieval-Augmented Large Language Models (RA-LLMs) have emerged to harness external and authoritative knowledge bases, rather than solely relying on the model's internal knowledge, to augment the quality of the generated content of LLMs. In this survey, we comprehensively review existing research studies in RA-LLMs, covering three primary technical perspectives: Furthermore, to deliver deeper insights, we discuss current limitations and several promising directions for future research. Updated information about this survey can be found at: https://advanced-recommender-systems.github.io/RAG-Meets-LLMs/

Harnessing Large Language Models (LLMs) for recommendation is rapidly emerging, which relies on two fundamental steps to bridge the recommendation item space and the language space: 1) item indexing utilizes identifiers to represent items in the language space, and 2) generation grounding associates LLMs' generated token sequences to in-corpus items. However, previous methods exhibit inherent limitations in the two steps. Existing ID-based identifiers (e.g., numeric IDs) and description-based identifiers (e.g., titles) either lose semantics or lack adequate distinctiveness. Moreover, prior generation grounding methods might generate invalid identifiers, thus misaligning with in-corpus items. To address these issues, we propose a novel Transition paradigm for LLM-based Recommender (named TransRec) to bridge items and language. Specifically, TransRec presents multi-facet identifiers, which simultaneously incorporate ID, title, and attribute for item indexing to pursue both distinctiveness and semantics. Additionally, we introduce a specialized data structure for TransRec to ensure generating valid identifiers only and utilize substring indexing to encourage LLMs to generate from any position of identifiers. Lastly, TransRec presents an aggregated grounding module to leverage generated multi-facet identifiers to rank in-corpus items efficiently. We instantiate TransRec on two backbone models, BART-large and LLaMA-7B.

As online music consumption increasingly shifts towards playlist-based listening, the task of playlist continuation, in which an algorithm suggests songs to extend a playlist in a personalized and musically cohesive manner, has become vital to the success of music streaming services. Currently, many existing playlist continuation approaches rely on collaborative filtering methods to perform their recommendations. However, such methods will struggle to recommend songs that lack interaction data, an issue known as the cold-start problem. Current approaches to this challenge design complex mechanisms for extracting relational signals from sparse collaborative signals and integrating them into content representations. However, these approaches leave content representation learning out of scope and utilize frozen, pre-trained content models that may not be aligned with the distribution or format of a specific musical setting. Furthermore, even the musical state-of-the-art content modules are either (1) incompatible with the cold-start setting or (2) unable to effectively integrate cross-modal and relational signals. In this paper, we introduce LARP, a multi-modal cold-start playlist continuation model, to effectively overcome these limitations. LARP is a three-stage contrastive learning framework that integrates both multi-modal and relational signals into its learned representations. Our framework uses increasing stages of task-specific abstraction: within-track (language-audio) contrastive loss, track-track contrastive loss, and track-playlist contrastive loss. Experimental results on two publicly available datasets demonstrate the efficacy of LARP over uni-modal and multi-modal models for playlist continuation in a cold-start setting. Finally, this work pioneers the perspective of addressing cold-start recommendation via relational representation learning. Code and dataset are released at: https://github.com/Rsalganik1123/LARP/

The development of virtual try-on has revolutionized online shopping by allowing customers to visualize themselves in various fashion items, thus extending the in-store try-on experience to the cyber space. Although virtual try-on has attracted considerable research initiatives, existing systems only focus on the quality of image generation, overlooking whether the fashion item is a good match to the given person and clothes. Recognizing this gap, we propose to design a one-stop Smart Fitting Room, with the novel formulation of matching-aware virtual try-on. Following this formulation, we design a Hybrid Matching-aware Virtual Try-On Framework (HMaVTON), which combines retrieval-based and generative methods to foster a more personalized virtual try-on experience. This framework integrates a hybrid mix-and-match module and an enhanced virtual try-on module. The former can recommend fashion items available on the platform to boost sales and generate clothes that meets the diverse tastes of consumers. The latter provides high-quality try-on effects, delivering a one-stop shopping service. To validate the effectiveness of our approach, we enlist the expertise of fashion designers for a professional evaluation, assessing the rationality and diversity of the clothes combinations and conducting an evaluation matrix analysis. Our method significantly enhances the practicality of virtual try-on. The code is available at https://github.com/Yzcreator/HMaVTON.

In Large Language Models (LLMs), text generation that involves knowledge representation is often fraught with the risk of "hallucinations'', where models confidently produce erroneous or fabricated content. These inaccuracies often stem from intrinsic biases in the pre-training stage or from the incorporation of human preference biases during the fine-tuning process. To mitigate these issues, we take inspiration from Goldman's causal theory of knowledge, which asserts that knowledge is not merely about having a true belief but also involves a causal connection between the belief and the truth of the proposition. We instantiate this theory within the context of Knowledge Question Answering (KQA) by constructing a causal graph that delineates the pathways between the candidate knowledge and belief. Through the application of the do-calculus rules from structural causal models, we devise an unbiased estimation framework based on this causal graph, thereby establishing a methodology for knowledge modeling grounded in causal inference. The resulting CORE framework (short for "Causal knOwledge REasoning'') is comprised of four essential components: question answering, causal reasoning, belief scoring, and refinement. Together, they synergistically improve the KQA system by fostering faithful reasoning and introspection. Extensive experiments are conducted on ScienceQA and HotpotQA datasets, which demonstrate the effectiveness and rationality of the CORE framework.

The remarkable performance of Multimodal Large Language Models (MLLMs) has demonstrated their proficient understanding capabilities in handling various visual tasks. Nevertheless, the opaque nature of black-box reasoning processes persists as an enigma, rendering them uninterpretable and struggling with hallucination. Their ability to execute intricate reasoning tasks is also constrained, culminating in stagnation of progression. In this work, we introduce Fact, a novel paradigm designed to generate multimodal rationales that are faithful, concise, and transferable for teaching MLLMs. This paradigm utilizes verifiable visual programming to generate executable code guaranteeing faithfulness. Through a series of operations including pruning, merging, and bridging, the rationale enhances its conciseness. Furthermore, we filter rationales that can be transferred to end-to-end paradigms from programming paradigms to guarantee transferability. Empirical evidence from experiments demonstrates the superiority of Fact across models of varying parameter sizes, significantly enhancing their compositional reasoning and generalization ability and reducing hallucinations owing to its high correlation between images and text.

We study an emerging and intriguing problem of multimodal temporal event forecasting with large language models. Compared to using text or graph modalities, the investigation of utilizing images for temporal event forecasting has not been fully explored, especially in the era of large language models (LLMs). To bridge this gap, we are particularly interested in two key questions of: 1) why images will help in temporal event forecasting, and 2) how to integrate images into the LLM-based forecasting framework. To answer these research questions, we propose to identify two essential functions that images play in the scenario of temporal event forecasting, i.e., highlighting and complementary. Then, we develop a novel framework, named MM-Forecast. It employs an Image Function Identification module to recognize these functions as verbal descriptions using multimodal large language models (MLLMs), and subsequently incorporates these function descriptions into LLM-based forecasting models. To evaluate our approach, we construct a new multimodal dataset, MidEast-TE-mm, by extending an existing event dataset MidEast-TE-mini with images. Empirical studies demonstrate that our MM-Forecast can correctly identify the image functions, and further more, incorporating these verbal function descriptions significantly improves the forecasting performance. The dataset, code, and prompts are available at https://github.com/LuminosityX/MM-Forecast.

Product bundling has been a prevailing marketing strategy that is beneficial in the online shopping scenario. Effective product bundling methods depend on high-quality item representations capturing both the individual items' semantics and cross-item relations. However, previous item representation learning methods, either feature fusion or graph learning, suffer from inadequate cross-modal alignment and struggle to capture the cross-item relations for cold-start items. Multimodal pre-train models could be the potential solutions given their promising performance on various multimodal downstream tasks. However, the cross-item relations have been under-explored in the current multimodal pre-train models.

This paper presents a pilot study aimed at introducing multi-agent debate into multimodal reasoning. The study addresses two key challenges: the trivialization of opinions resulting from excessive summarization and the diversion of focus caused by distractor concepts introduced from images. These challenges stem from the inductive (bottom-up) nature of existing debating schemes. To address the issue, we propose a deductive (top-down) debating approach called Blueprint Debate on Graphs (BDoG). In BDoG, debates are confined to a blueprint graph to prevent opinion trivialization through world-level summarization. Moreover, by storing evidence in branches within the graph, BDoG mitigates distractions caused by frequent but irrelevant concepts. Extensive experiments validate that BDoG is able to achieve state-of-the-art results in ScienceQA and MMBench with significant improvements over previous methods. The source code can be accessed at https://github.com/thecharm/BDoG.

Recent developments of vision large language models (LLMs) have seen remarkable progress, yet still encounter challenges towards multimodal generalists, such as coarse-grained instance-level understanding, lack of unified support for both images and videos, and insufficient coverage across various vision tasks. In this paper we present Vitron, a universal pixel-level vision LLM designed for comprehensive understanding, generating, segmenting, and editing of both static images and dynamic videos. Building on top of an LLM backbone, Vitron incorporates encoders for images, videos, and pixel-level regional visuals within its frontend modules, while employing state-of-the-art visual specialists as its backend, via which Vitron supports a spectrum of vision end tasks, spanning visual comprehension to visual generation, from low level to high level. To ensure an effective and precise message passing from LLM to backend modules for function invocation, we propose a novel hybrid method by simultaneously integrating discrete textual instructions and continuous signal embeddings. Further, we design various pixel-level spatiotemporal vision-language alignment learning for Vitron to reach the best fine-grained visual capability. Finally, a cross-task synergy module is advised to learn to maximize the task-invariant fine-grained visual features, enhancing the synergy between different visual tasks. Demonstrated over 12 visual tasks and evaluated across 22 datasets, Vitron showcases its extensive capabilities in the four main vision task clusters. Overall, this work illuminates the great potential of developing a more unified multimodal generalist.

Recommender systems aim to predict personalized rankings based on user preference data. With the rise of Language Models (LMs), LM-based recommenders have been widely explored due to their extensive world knowledge and powerful reasoning abilities. Most of the LM-based recommenders convert historical interactions into language prompts, pairing with a positive item as the target response and fine-tuning LM with a language modeling loss. However, the current objective fails to fully leverage preference data and is not optimized for personalized ranking tasks, which hinders the performance of LM-based recommenders. Inspired by the current advancement of Direct Preference Optimization (DPO) in human preference alignment and the success of softmax loss in recommendations, we propose Softmax-DPO (\textbf{S-DPO}) to instill ranking information into the LM to help LM-based recommenders distinguish preferred items from negatives, rather than solely focusing on positives. Specifically, we incorporate multiple negatives in user preference data and devise an alternative version of DPO loss tailored for LM-based recommenders, which is extended from the traditional full-ranking Plackett-Luce (PL) model to partial rankings and connected to softmax sampling strategies. Theoretically, we bridge S-DPO with the softmax loss over negative sampling and find that it has an inherent benefit of mining hard negatives, which assures its exceptional capabilities in recommendation tasks. Empirically, extensive experiments conducted on three real-world datasets demonstrate the superiority of S-DPO to effectively model user preference and further boost recommendation performance while providing better rewards for preferred items. Our codes are available at https://github.com/chenyuxin1999/S-DPO.

As Large Language Models (LLMs) demonstrate impressive capabilities, demystifying their internal mechanisms becomes increasingly vital. Neuron attribution, which attributes LLM outputs to specific neurons to reveal the semantic properties they learn, has emerged as a key interpretability approach. However, while neuron attribution has made significant progress in deciphering text-only LLMs, its application to Multimodal LLMs (MLLMs) remains less explored. To address this gap, we propose a novel Neuron Attribution method tailored for MLLMs, termed NAM. Specifically, NAM not only reveals the modality-specific semantic knowledge learned by neurons within MLLMs, but also highlights several intriguing properties of neurons, such as cross-modal invariance and semantic sensitivity. These properties collectively elucidate the inner workings mechanism of MLLMs, providing a deeper understanding of how MLLMs process and generate multi-modal content. Through theoretical analysis and empirical validation, we demonstrate the efficacy of NAM and the valuable insights it offers. Furthermore, leveraging NAM, we introduce a multi-modal knowledge editing paradigm, underscoring the practical significance of our approach for downstream applications of MLLMs.

Large Language Models (LLMs) can elicit unintended and even harmful content when misaligned with human values, posing severe risks to users and society. To mitigate these risks, current evaluation benchmarks predominantly employ expert-designed contextual scenarios to assess how well LLMs align with human values. However, the labor-intensive nature of these benchmarks limits their test scope, hindering their ability to generalize to the extensive variety of open-world use cases and identify rare but crucial long-tail risks. Additionally, these static tests fail to adapt to the rapid evolution of LLMs, making it hard to evaluate timely alignment issues. To address these challenges, we propose ALI-Agent, an evaluation framework that leverages the autonomous abilities of LLM-powered agents to conduct in-depth and adaptive alignment assessments. ALI-Agent operates through two principal stages: Emulation and Refinement. During the Emulation stage, ALI-Agent automates the generation of realistic test scenarios. In the Refinement stage, it iteratively refines the scenarios to probe long-tail risks. Specifically, ALI-Agent incorporates a memory module to guide test scenario generation, a tool-using module to reduce human labor in tasks such as evaluating feedback from target LLMs, and an action module to refine tests. Extensive experiments across three aspects of human values--stereotypes, morality, and legality--demonstrate that ALI-Agent, as a general evaluation framework, effectively identifies model misalignment. Systematic analysis also validates that the generated test scenarios represent meaningful use cases, as well as integrate enhanced measures to probe long-tail risks.

In the evolving landscape of recommender systems, Graph Contrastive Learning (GCL) has become a prominent method for enhancing recommendation performance by alleviating the issue of data sparsity. However, existing GCL-based recommendations often overlook the control of shared information between the contrastive views. In this paper, we initially analyze and experimentally demonstrate these methods often lead to the issue of augmented representation collapse, where the representations between views become excessively similar, diminishing their distinctiveness. To address this issue, we propose the Information-Controllable Graph Contrastive Learning (IGCL) framework, a novel approach that focuses on optimizing the shared information between views to include as much relevant information for the recommendation task as possible while maintaining an appropriate level. In particular, we design the Collaborative Signals Enhanced Augmentation module to infuse the augmented representation with rich, task-relevant collaborative signals. Furthermore, the Information-Controllable Contrastive Learning module is designed to direct control over the magnitude of shared information between the contrastive views to avoid over-similarity. Extensive experiments on three public datasets demonstrate the effectiveness of IGCL, showcasing significant improvements in performance and the capability to alleviate augmented representation collapse.

The vital goal of information retrieval today extends beyond merely connecting users with relevant information they search for. It also aims to enrich the diversity, personalization, and interactivity of that connection, ensuring the information retrieval process is as seamless, beneficial, and supportive as possible in the global digital era. Current information retrieval systems often encounter challenges like a constrained understanding of queries, static and inflexible responses, limited personalization, and restricted interactivity. With the advent of large language models (LLMs), there's a transformative paradigm shift as we integrate LLM-powered agents into these systems. These agents bring forth crucial human capabilities like memory and planning to make them behave like humans in completing various tasks, effectively enhancing user engagement and offering tailored interactions. In this tutorial, we delve into the cutting-edge techniques of LLM-powered agents across various information retrieval fields, such as search engines, social networks, recommender systems, and conversational assistants. We will also explore the prevailing challenges in seamlessly incorporating these agents and hint at prospective research avenues that can revolutionize the way of information retrieval.

Recommender systems are the cornerstone of today's information dissemination, yet a disconnect between offline metrics and online performance greatly hinders their development. Addressing this challenge, we envision a recommendation simulator, capitalizing on recent breakthroughs in human-level intelligence exhibited by Large Language Models (LLMs). We propose Agent4Rec, a user simulator in recommendation, leveraging LLM-empowered generative agents equipped with user profile, memory, and actions modules specifically tailored for the recommender system. In particular, these agents' profile modules are initialized using real-world datasets (e.g., MovieLens, Steam, Amazon-Book), capturing users' unique tastes and social traits; memory modules log both factual and emotional memories and are integrated with an emotion-driven reflection mechanism; action modules support a wide variety of behaviors, spanning both taste-driven and emotion-driven actions. Each agent interacts with personalized recommender models in a page-by-page manner, relying on a pre-implemented collaborative filtering-based recommendation algorithm. We delve into both the capabilities and limitations of Agent4Rec, aiming to explore an essential research question: "To what extent can LLM-empowered generative agents faithfully simulate the behavior of real, autonomous humans in recommender systems?" Extensive and multi-faceted evaluations of Agent4Rec highlight both the alignment and deviation between agents and user-personalized preferences. Beyond mere performance comparison, we explore insightful experiments, such as emulating the filter bubble effect and discovering the underlying causal relationships in recommendation tasks.

Fair re-ranking aims to redistribute ranking slots among items more equitably to ensure responsibility and ethics. The exploration of redistribution problems has a long history in economics, offering valuable insights for conceptualizing fair re-ranking as a taxation process. Such a formulation provides us with a fresh perspective to re-examine fair re-ranking and inspire the development of new methods. From a taxation perspective, we theoretically demonstrate that most previous fair re-ranking methods can be reformulated as an item-level tax policy. Ideally, a good tax policy should be effective and conveniently controllable to adjust ranking resources. However, both empirical and theoretical analyses indicate that the previous item-level tax policy cannot meet two ideal controllable requirements: (1) continuity, ensuring minor changes in tax rates result in small accuracy and fairness shifts; (2) controllability over accuracy loss, ensuring precise estimation of the accuracy loss under a specific tax rate. To overcome these challenges, we introduce a new fair re-ranking method named Tax-rank, which levies taxes based on the difference in utility between two items. Then, we efficiently optimize such an objective by utilizing the Sinkhorn algorithm in optimal transport. Upon a comprehensive analysis, Our model Tax-rank offers a superior tax policy for fair re-ranking, theoretically demonstrating both continuity and controllability over accuracy loss. Experimental results show that Tax-rank outperforms all state-of-the-art baselines on two ranking tasks.

Recent research on proactive conversational agents (PCAs) mainly focuses on improving the system's capabilities in anticipating and planning action sequences to accomplish tasks and achieve goals before users articulate their requests. This perspectives paper highlights the importance of moving towards building human-centered PCAs that emphasize human needs and expectations, and that considers ethical and social implications of these agents, rather than solely focusing on technological capabilities. The distinction between a proactive and a reactive system lies in the proactive system's initiative-taking nature. Without thoughtful design, proactive systems risk being perceived as intrusive by human users. We address the issue by establishing a new taxonomy concerning three key dimensions of human-centered PCAs, namely Intelligence, Adaptivity, and Civility. We discuss potential research opportunities and challenges based on this new taxonomy upon the five stages of PCA system construction. This perspectives paper lays a foundation for the emerging area of conversational information retrieval research and paves the way towards advancing human-centered proactive conversational systems.

Leveraging Large Language Models (LLMs) for recommendation has recently garnered considerable attention, where fine-tuning plays a key role in LLMs' adaptation. However, the cost of fine-tuning LLMs on rapidly expanding recommendation data limits their practical application. To address this challenge, few-shot fine-tuning offers a promising approach to quickly adapt LLMs to new recommendation data. We propose the task of data pruning for efficient LLM-based recommendation, aimed at identifying representative samples tailored for LLMs' few-shot fine-tuning. While coreset selection is closely related to the proposed task, existing coreset selection methods often rely on suboptimal heuristic metrics or entail costly optimization on large-scale recommendation data.

Recent studies indicate that dense retrieval models struggle to perform well on a wide variety of retrieval tasks that lack dedicated training data, as different retrieval tasks often entail distinct search intents. To address this challenge, in this work we leverage instructions to flexibly describe retrieval intents and introduce I3, a unified retrieval system that performs Intent-Introspective retrieval across various tasks, conditioned on Instructions without any task-specific training. I3 innovatively incorporates a pluggable introspector in a parameter-isolated manner to comprehend specific retrieval intents by jointly reasoning over the input query and instruction, and seamlessly integrates the introspected intent into the original retrieval model for intent-aware retrieval. Furthermore, we propose progressively-pruned intent learning. It utilizes extensive LLM-generated data to train I3 phase-by-phase, embodying two key designs: progressive structure pruning and drawback extrapolation-based data refinement. Extensive experiments show that in the BEIR benchmark, I3 significantly outperforms baseline methods designed with task-specific retrievers, achieving state-of-the-art zero-shot performance without any task-specific tuning.

Composed image retrieval (CIR) aims to retrieve the target image based on a multimodal query, i.e., a reference image paired with corresponding modification text. Recent CIR studies leverage vision-language pre-trained (VLP) methods as the feature extraction backbone and perform nonlinear feature-level multimodal query fusion to retrieve the target image. Despite the promising performance, we argue that their nonlinear feature-level multimodal fusion may lead to the fused feature deviating from the original embedding space, potentially hurting the retrieval performance. To address this issue, in this work, we propose shifting the multimodal fusion from the feature level to the raw-data level to fully exploit the VLP model's multimodal encoding and cross-modal alignment abilities. In particular, we introduce a Dual Query Unification-based Composed Image Retrieval framework (DQU-CIR), whose backbone simply involves a VLP model's image encoder and a text encoder. Specifically, DQU-CIR first employs two training-free query unification components to derive a unified textual and visual query based on the raw data of the multimodal query, respectively. The unified textual query is derived by concatenating the modification text with the extracted reference image's textual description, while the unified visual query is created by writing the key modification words onto the reference image. Ultimately, to address diverse search intentions, DQU-CIR linearly combines the features of the two unified queries encoded by the VLP model to retrieve the target image. Extensive experiments on four real-world datasets validate the effectiveness of our proposed method.

Recommender systems often grapple with noisy implicit feedback. Most studies alleviate the noise issues from data cleaning perspective such as data resampling and reweighting, but they are constrained by heuristic assumptions. Another denoising avenue is from model perspective, which proactively injects noises into user-item interactions and enhances the intrinsic denoising ability of models. However, this kind of denoising process poses significant challenges to the recommender model's representation capacity to capture noise patterns.

Automatic bundle construction is a crucial prerequisite step in various bundle-aware online services. Previous approaches are mostly designed to model the bundling strategy of existing bundles. However, it is hard to acquire large-scale well-curated bundle dataset, especially for those platforms that have not offered bundle services before. Even for platforms with mature bundle services, there are still many items that are included in few or even zero bundles, which give rise to sparsity and cold-start challenges in the bundle construction models. To tackle these issues, we target at leveraging multimodal features, item-level user feedback signals, and the bundle composition information, to achieve a comprehensive formulation of bundle construction. Nevertheless, such formulation poses two new technical challenges: 1) how to learn effective representations by unifying multiple features optimally, and 2) how to address the problems of modality missing, noise, and sparsity problems induced by the incomplete query bundles. In this work, to address these technical challenges, we propose a Contrastive Learning-enhanced Hierarchical Encoder method (CLHE). Specifically, we use self-attention modules to combine the multimodal and multi-item features, and then leverage both item- and bundle-level contrastive learning to enhance the representation learning, thus to counter the modality missing, noise, and sparsity problems. Extensive experiments on four datasets in two application domains demonstrate that our method outperforms a list of SOTA methods. The code and dataset are available at https://github.com/Xiaohao-Liu/CLHE.

Explaining stock predictions is generally a difficult task for traditional non-generative deep learning models, where explanations are limited to visualizing the attention weights on important texts. Today, Large Language Models (LLMs) present a solution to this problem, given their known capabilities to generate human-readable explanations for their decision-making process. However, the task of stock prediction remains challenging for LLMs, as it requires the ability to weigh the varying impacts of chaotic social texts on stock prices. The problem gets progressively harder with the introduction of the explanation component, which requires LLMs to explain verbally why certain factors are more important than the others. On the other hand, to fine-tune LLMs for such a task, one would need expert-annotated samples of explanation for every stock movement in the training set, which is expensive and impractical to scale.

Recommender systems are vulnerable to injective attacks, which inject limited fake users into the platforms to manipulate the exposure of target items to all users. In this work, we identify that conventional injective attackers overlook the fact that each item has its unique potential audience, and meanwhile, the attack difficulty across different users varies. Blindly attacking all users will result in a waste of fake user budgets and inferior attack performance. To address these issues, we focus on an under-explored attack task called target user attacks, aiming at promoting target items to a particular user group. In addition, we formulate the varying attack difficulty as heterogeneous treatment effects through a causal lens and propose an Uplift-guided Budget Allocation (UBA) framework. UBA estimates the treatment effect on each target user and optimizes the allocation of fake user budgets to maximize the attack performance. Theoretical and empirical analysis demonstrates the rationality of treatment effect estimation methods of UBA. By instantiating UBA on multiple attackers, we conduct extensive experiments on three datasets under various settings with different target items, target users, fake user budgets, victim models, and defense models, validating the effectiveness and robustness of UBA.

Making the contents generated by Large Language Model (LLM), accurate, credible and traceable is crucial, especially in complex knowledge-intensive tasks that require multi-step reasoning and each step needs knowledge to solve. Retrieval-augmented generation is good potential to solve this problem. However, where and how to introduce Information Retrieval (IR) to LLM is a big challenge. Previous work has the problems that wrong knowledge retrieved by IR misleads the LLM and interaction between IR and LLM breaks the reasoning chain of LLM. This paper proposes a novel framework named Search-in-the-Chain (SearChain) for the interaction between LLM and IR to solve the challenges. First, LLM generates the reasoning chain named Chain-of-Query (CoQ) where each node consists of an IR-oriented query-answer pair. Second, IR verifies the answer of each node of CoQ. It corrects the answer that is not consistent with the retrieved information when IR gives high confidence, which improves the credibility. Third, LLM can indicate its missing knowledge in CoQ and rely on IR to provide this knowledge to LLM. These operations improve the accuracy in terms of reasoning and knowledge. Finally, SearChain generates the reasoning process and marks references to supporting documents for each reasoning step, which improves traceability. Interaction with IR in SearChain forms a novel reasoning path based on a tree, which enables LLM to dynamically modify the direction of reasoning. Experiments show that SearChain outperforms state-of-the-art baselines on complex knowledge-intensive tasks including multi-hop Q&A, slot filling, fact checking, and long-form Q&A.

While pre-training large-scale video-language models (VLMs) has shown remarkable potential for various downstream video-language tasks, existing VLMs can still suffer from certain commonly seen limitations, e.g., coarse-grained cross-modal aligning, under-modeling of temporal dynamics, detached video-language view. In this work, we target enhancing VLMs with a fine-grained structural spatio-temporal alignment learning method (namely Finsta). First of all, we represent the input texts and videos with fine-grained scene graph (SG) structures, both of which are further unified into a holistic SG (HSG) for bridging two modalities. Then, an SG-based framework is built, where the textual SG (TSG) is encoded with a graph Transformer, while the video dynamic SG (DSG) and the HSG are modeled with a novel recurrent graph Transformer for spatial and temporal feature propagation. A spatial-temporal Gaussian differential graph Transformer is further devised to strengthen the sense of the changes in objects across spatial and temporal dimensions. Next, based on the fine-grained structural features of TSG and DSG, we perform object-centered spatial alignment and predicate-centered temporal alignment respectively, enhancing the video-language grounding in both the spatiality and temporality. We design our method as a plug&play system, which can be integrated into existing well-trained VLMs for further representation augmentation, without training from scratch or relying on SG annotations in downstream applications. On 6 representative VL modeling tasks over 12 datasets in both standard and long-form video scenarios, Finsta consistently improves the existing 13 strong-performing VLMs persistently, and refreshes the current state-of-the-art end task performance significantly in both the fine-tuning and zero-shot settings.

The aspect term extraction (ATE) task aims to extract aspect terms describing a part or an attribute of a product from review sentences. Most existing works rely on either general or domain embedding to address this problem. Despite the promising results, the importance of general and domain embeddings is still ignored by most methods, resulting in degraded performances. Besides, word embedding is also related to downstream tasks, and how to regularize word embeddings to capture context-aware information is an unresolved problem. To solve these issues, we first propose context-aware dynamic word embedding (CDWE), which could simultaneously consider general meanings, domain-specific meanings, and the context information of words. Based on CDWE, we propose an attention-based convolution neural network, called ADWE-CNN for ATE, which could adaptively capture the previous meanings of words by utilizing an attention mechanism to assign different importance to the respective embeddings. The experimental results show that ADWE-CNN achieves a comparable performance with the state-of-the-art approaches. Various ablation studies have been conducted to explore the benefit of each component. Our code is publicly available athttp://github.com/xiejiajia2018/ADWE-CNN.

In driving scenes, road agents often engage in frequent interaction and strive to understand their surroundings. Ego-agent (each road agent itself) predicts what behavior will be engaged by other road users all the time and expects a shared and consistent understanding for safe movement. To achieve this, Behavioral Intention Prediction (BIP) simulates such a human consideration process to anticipate specific behaviors, and the rapid development of BIP inevitably leads to new issues and challenges. To catalyze future research, this work provides a comprehensive review of BIP from the available datasets, key factors, challenges, pedestrian-centric and vehicle-centric BIP approaches, and BIP-aware applications. The investigation reveals that data-driven deep learning approaches have become the primary pipelines, while the behavioral intention types are still limited in most current datasets and methods (e.g., Crossing (C) and Not Crossing (NC) for pedestrians and Lane Changing (LC) for vehicles) in this field. In addition, current research on BIP in safe-critical scenarios (e.g., near-crashing situations) is limited. Through this investigation, we identify open issues in behavioral intention prediction and suggest possible insights for future research.

In graph classification, attention- and pooling-based graph neural networks (GNNs) predominate to extract salient features from the input graph and support the prediction. They mostly follow the paradigm of “learning to attend,” which maximizes the mutual information between the attended graph and the ground-truth label. However, this paradigm causes GNN classifiers to indiscriminately absorb all statistical correlations between input features and labels in the training data without distinguishing the causal and noncausal effects of features. Rather than emphasizing causal features, the attended graphs tend to rely on noncausal features as shortcuts to predictions. These shortcut features may easily change outside the training distribution, thereby leading to poor generalization for GNN classifiers. In this article, we take a causal view on GNN modeling. Under our causal assumption, the shortcut feature serves as a confounder between the causal feature and prediction. It misleads the classifier into learning spurious correlations that facilitate prediction in in-distribution (ID) test evaluation while causing significant performance drop in out-of-distribution (OOD) test data. To address this issue, we employ the backdoor adjustment from causal theory—combining each causal feature with various shortcut features, to identify causal patterns and mitigate the confounding effect. Specifically, we employ attention modules to estimate the causal and shortcut features of the input graph. Then, a memory bank collects the estimated shortcut features, enhancing the diversity of shortcut features for combination. Simultaneously, we apply the prototype strategy to improve the consistency of intra-class causal features. We term our method as CAL+, which can promote stable relationships between causal estimation and prediction, regardless of distribution changes. Extensive experiments on synthetic and real-world OOD benchmarks demonstrate our method’s effectiveness in improving OOD generalization. Our codes are released at https://github.com/shuyao-wang/CAL-plus .

To empower the trust of current recommender systems, the counterfactual explanation (CE) method is adopted to generate the counterfactual instance for each input and take their changes causing the different outcomes as the explanation. Although promising results have been achieved by existing CE-based methods, we propose to generate the attribute-oriented counterfactual explanation. Different from them, we aim to generate the counterfactual instance by performing the intervention on the attributes, and then build an attribute-oriented counterfactual explainable recommender system. Considering the correlation and categorical values of attributes, how to efficiently generate the reliable counterfactual instances on the attributes challenges us. To alleviate such a problem, we propose to extract the decision rules over the attributes to guide the attribute-oriented counterfactual generation. Specifically, we adopt the gradient boosting decision tree (GBDT) to pre-build the decision rules over the attributes and develop a Rule-guided Counterfactual Explainable Recommendation model (RCER) to predict the user-item interaction and generate the counterfactual instances for the user-item pairs. We finally conduct extensive experiments on four publicly datasets, including NYC, LON, Amazon, and Movielens datasets. Experimental results have qualitatively and quantitatively justified the superiority of our model over existing cutting-edge baselines. We release the code:https://github.com/quxiaoyang0zero/RCER.

Recommendation performance usually exhibits a long-tail distribution over users — a small portion of head users enjoy much more accurate recommendation services than the others. We reveal two sources of this performance heterogeneity problem: the uneven distribution of historical interactions (a natural source); and the biased training of recommender models (a model source). As addressing this problem cannot sacrifice the overall performance, a wise choice is to eliminate the model bias while maintaining the natural heterogeneity. The key to debiased training lies in eliminating the effect of confounders that influence both the user's historical behaviors and the next behavior. The emerging causal recommendation methods achieve this by modeling the causal effect between user behaviors, however potentially neglect unobserved confounders (e.g., friend suggestions) that are hard to measure in practice. To address unobserved confounders, we resort to the front-door adjustment (FDA) in causal theory and propose a causal multi-teacher distillation framework (CausalD). FDA requires proper mediators in order to estimate the causal effects of historical behaviors on the next behavior. To achieve this, we equip CausalD with multiple heterogeneous recommendation models to model the mediator distribution. Then, the causal effect estimated by FDA is the expectation of recommendation prediction over the mediator distribution and the prior distribution of historical behaviors, which is technically achieved by multi-teacher ensemble. To pursue efficient inference, CausalD further distills multiple teachers into one student model to directly infer the causal effect for making recommendations. We instantiate CausalD on two representative models, DeepFM and DIN, and conduct extensive experiments on three real-world datasets, which validate the superiority of CausalD over state-of-the-art methods. Through in-depth analysis, we find that CausalD largely improves the performance of tail users, reduces the performance heterogeneity, and enhances the overall performance.

The present machine learning schema typically uses a one-pass model inference (e.g., forward propagation) to make predictions in the testing phase. It is inherently different from human students who double-check the answer during examinations especially when the confidence is low. To bridge this gap, we propose a learning to double-check (L2D) framework, which formulates double check as a learnable procedure with two core operations: recognizing unreliable predictions and revising predictions. To judge the correctness of a prediction, we resort to counterfactual faithfulness in causal theory and design a contrastive faithfulness measure. In particular, L2D generates counterfactual features by imagining: "what would the sample features be if its label was the predicted class" and judges the prediction by the faithfulness of the counterfactual features. Furthermore, we design a simple and effective revision module to revise the original model prediction according to the faithfulness. We apply the L2D framework to three classification models and conduct experiments on two public datasets for image classification, validating the effectiveness of L2D in prediction correctness judgment and revision.

Rumor verification on social media aims to identify the truth value of a rumor, which is important to decrease the detrimental public effects. A rumor might arouse heated discussions and replies, conveying different stances of users that could be helpful in identifying the rumor. Thus, several works have been proposed to verify a rumor by modelling its entire stance sequence in the time domain. However, these works ignore that such a stance sequence could be decomposed into controversies with different intensities, which could be used to cluster the stance sequences with the same consensus. In addition, the existing stance extractors fail to consider both the impact of all previously posted tweets and the reply chain on obtaining the stance of a new reply. To address the above problems, in this article, we propose a novel stance-based network to aggregate the controversies of the stance sequence for rumor verification, termed Filter-based Stance Network (FSNet). As controversies with different intensities are reflected as the different changes of stances, it is convenient to represent different controversies in the frequency domain, but it is hard in the time domain. Our proposed FSNet decomposes the stance sequence into multiple controversies in the frequency domain and obtains the weighted aggregation of them. Specifically, FSNet consists of two modules: the stance extractor and the filter block. To obtain better stance features toward the source, the stance extractor contains two stages. In the first stage, the tweet representation of each reply is obtained by aggregating information from all previously posted tweets in a conversation. Then, the features of stance toward the source, i.e., rumor-aware stance, are extracted with the reply chains in the second stage. In the filter block module, a rumor-aware stance sequence is constructed by sorting all the tweets of a conversation in chronological order. Fourier Transform thereafter is employed to convert the stance sequence into the frequency domain, where different frequency components reflect controversies of different intensities. Finally, a frequency filter is applied to explore the different contributions of controversies. We supervise our FSNet with both stance labels and rumor labels to strengthen the relations between rumor veracity and crowd stances. Extensive experiments on two benchmark datasets demonstrate that our model substantially outperforms all the baselines.

Multimodal information (e.g., visual, acoustic, and textual) has been widely used to enhance representation learning for micro-video recommendation. For integrating multimodal information into a joint representation of micro-video, multimodal fusion plays a vital role in the existing micro-video recommendation approaches. However, the static multimodal fusion used in previous studies is insufficient to model the various relationships among multimodal information of different micro-videos. In this article, we develop a novel meta-learning-based multimodal fusion framework called Meta Multimodal Fusion (MetaMMF), which dynamically assigns parameters to the multimodal fusion function for each micro-video during its representation learning. Specifically, MetaMMF regards the multimodal fusion of each micro-video as an independent task. Based on the meta information extracted from the multimodal features of the input task, MetaMMF parameterizes a neural network as the item-specific fusion function via a meta learner. We perform extensive experiments on three benchmark datasets, demonstrating the significant improvements over several state-of-the-art multimodal recommendation models, like MMGCN, LATTICE, and InvRL. Furthermore, we lighten our model by adopting canonical polyadic decomposition to improve the training efficiency, and validate its effectiveness through experimental results. Codes are available at https://github.com/hanliu95/MetaMMF .

Bundle recommendation seeks to recommend a bundle of related items to users to improve both user experience and the profits of platform. Existing bundle recommendation models have progressed from capturing only user-bundle interactions to the modeling of multiple relations among users, bundles, and items. CrossCBR, in particular, incorporates cross-view contrastive learning into a two-view preference learning framework, significantly improving SOTA performance. It does, however, have two limitations: (1) the two-view formulation does not fully exploit all the heterogeneous relations among users, bundles, and items; and (2) the “early contrast and late fusion” framework is less effective in capturing user preference and difficult to generalize to multiple views. In this article, we present MultiCBR, a novel Multi -view C ontrastive learning framework for B undle R ecommendation. First, we devise a multi-view representation learning framework capable of capturing all the user-bundle, user-item, and bundle-item relations, especially better utilizing the bundle-item affiliations to enhance sparse bundles’ representations. Second, we innovatively adopt an “early fusion and late contrast” design that first fuses the multi-view representations before performing self-supervised contrastive learning. In comparison to existing approaches, our framework reverses the order of fusion and contrast, introducing the following advantages: (1) Our framework is capable of modeling both cross-view and ego-view preferences, allowing us to achieve enhanced user preference modeling; and (2) instead of requiring quadratic number of cross-view contrastive losses, we only require two self-supervised contrastive losses, resulting in minimal extra costs. Experimental results on three public datasets indicate that our method outperforms SOTA methods. The code and dataset can be found in the github repo https://github.com/HappyPointer/MultiCBR .

Recommender systems easily face the issue of user preference shifts. User representations will become out-of-date and lead to inappropriate recommendations if user preference has shifted over time. To solve the issue, existing work focuses on learning robust representations or predicting the shifting pattern. There lacks a comprehensive view to discover the underlying reasons for user preference shifts. To understand the preference shift, we abstract a causal graph to describe the generation procedure of user interaction sequences. Assuming user preference is stable within a short period, we abstract the interaction sequence as a set of chronological environments. From the causal graph, we find that the changes of some unobserved factors (e.g., becoming pregnant) cause preference shifts between environments. Besides, the fine-grained user preference over item categories sparsely affects the interactions with different items. Inspired by the causal graph, our key considerations to handle preference shifts lie in modeling the interaction generation procedure by: (1) capturing the preference shifts across environments for accurate preference prediction and (2) disentangling the sparse influence from user preference to interactions for accurate effect estimation of preference. To this end, we propose a Causal Disentangled Recommendation (CDR) framework, which captures preference shifts via a temporal variational autoencoder and learns the sparse influence from multiple environments. Specifically, an encoder is adopted to infer the unobserved factors from user interactions while a decoder is to model the interaction generation process. Besides, we introduce two learnable matrices to disentangle the sparse influence from user preference to interactions. Last, we devise a multi-objective loss to optimize CDR. Extensive experiments on three datasets show the superiority of CDR in enhancing the generalization ability under user preference shifts.

Benefiting many information retrieval applications, named entity recognition (NER) has shown impressive progress. Recently, there has been a growing trend to decompose complex NER tasks into two subtasks (e.g., entity span detection (ESD) and entity type classification (ETC), to achieve better performance. Despite the remarkable success, from the perspective of representation, existing methods do not explicitly distinguish non-entities and entities, which may lead to ESD errors. Meanwhile, they do not explicitly distinguish entities with different entity types, which may lead to entity type misclassification. As such, the limited representation abilities may challenge some competitive NER methods, leading to unsatisfactory performance, especially in the low-resource setting (e.g., cross-domain NER). In light of these challenges, we propose to utilize contrastive learning to refine the original chaotic representations and learn the generalized representations for cross-domain NER. In particular, this article proposes a dual contrastive learning model (Dual-CL), which respectively utilizes a token-level contrastive learning module and a sentence-level contrastive learning module to enhance ESD, ETC for cross-domain NER. Empirical results on 10 domain pairs under two different settings show that Dual-CL achieves better performances than compared baselines in terms of several standard metrics. Moreover, we conduct detailed analyses to are presented to better understand each component’s effectiveness.

In recommender systems, click behaviors play a fundamental role in mining users’ interests and training models (clicked items as positive samples). Such signals are implicit feedback and are arguably less representative of users’ inherent interests. Most existing works denoise implicit feedback by introducing external signals, such as gaze, dwell time, and “like” behaviors. However, such explicit feedback is not always routinely available, or might be problematic to collect on a large scale. In this paper, we identify that an interaction’s related structural patterns in its neighborhood graph are potentially correlated with some outcome of implicit feedback (i.e., users’ ratings after consuming items), analogous to findings in other domains such as social networks. Inspired by this finding, we propose a novel Structure LEarning based Denoising (SLED) framework for denoising recommendation without explicit signals, which consists of two phases: center-aware graph structure learning and denoised recommendation . Phase 1 pre-trains a structural encoder in a self-supervised manner and learns to capture an interaction’s related structural patterns in its neighborhood graph. Phase 2 transfers the structure encoder to downstream recommendation datasets, which helps to down-weight the effect of noisy interactions on user interest modeling and loss calculation. We collect a relatively noisy industrial dataset across several days during a period of product promotion festival. Extensive experiments on this dataset and multiple public datasets demonstrate that the proposed SLED framework can significantly improve the recommendation quality over various base recommendation models.

In leading collaborative filtering (CF) models, representations of users and items are prone to learn popularity bias in the training data as shortcuts. The popularity shortcut tricks are good for in-distribution (ID) performance but poorly generalized to out-of-distribution (OOD) data, i.e., when popularity distribution of test data shifts w.r.t. the training one. To close the gap, debiasing strategies try to assess the shortcut degrees and mitigate them from the representations. However, there exist two deficiencies: (1) when measuring the shortcut degrees, most strategies only use statistical metrics on a single aspect (i.e., item frequency on item and user frequency on user aspect), failing to accommodate the compositional degree of a user–item pair; (2) when mitigating shortcuts, many strategies assume that the test distribution is known in advance. This results in low-quality debiased representations. Worse still, these strategies achieve OOD generalizability with a sacrifice on ID performance. In this work, we present a simple yet effective debiasing strategy, PopGo , which quantifies and reduces the interaction-wise popularity shortcut without any assumptions on the test data. It first learns a shortcut model, which yields a shortcut degree of a user–item pair based on their popularity representations. Then, it trains the CF model by adjusting the predictions with the interaction-wise shortcut degrees. By taking both causal- and information-theoretical looks at PopGo, we can justify why it encourages the CF model to capture the critical popularity-agnostic features while leaving the spurious popularity-relevant patterns out. We use PopGo to debias two high-performing CF models (matrix factorization [ 28 ] and LightGCN [ 19 ]) on four benchmark datasets. On both ID and OOD test sets, PopGo achieves significant gains over the state-of-the-art debiasing strategies (e.g., DICE [ 71 ] and MACR [ 58 ]). Codes and datasets are available at https://github.com/anzhang314/PopGo .

In recent years, various dense retrieval methods have been developed to improve the performance of search engines with a vectorized index. However, these approaches require a large pre-computed index and have a limited capacity to memorize all semantics in a document within a single vector. To address these issues, researchers have explored end-to-end generative retrieval models that use a seq-to-seq generative model to directly return identifiers of relevant documents. Although these models have been effective, they are often trained with the MLE method. It only encourages the model to assign a high probability to the relevant document identifier, ignoring the relevance comparisons of other documents. This may lead to performance degradation in ranking tasks, where the core is to compare the relevance between documents. To address this issue, we propose a ranking-oriented generative retrieval model that incorporates relevance signals to better estimate the relative relevance of different documents in ranking tasks. Based upon the analysis of the optimization objectives of dense retrieval and generative retrieval, we propose utilizing dense retrieval to provide relevance feedback for generative retrieval. Under an alternate training framework, the generative retrieval model gradually acquires higher-quality ranking signals to optimize the model. Experimental results show that our approach increasing Recall@1 by 12.9% with respect to the baselines on MS MARCO dataset.

Existing video-audio understanding models are trained and evaluated in an intra-domain setting, facing performance degeneration in real-world applications where multiple domains and distribution shifts naturally exist. The key to video-audio domain generalization (VADG) lies in alleviating spurious correlations over multi-modal features. To achieve this goal, we resort to causal theory and attribute such correlation to confounders affecting both video-audio features and labels. We propose a DeVADG framework that conducts uni-modal and cross-modal deconfounding through back-door adjustment. DeVADG performs cross-modal disentanglement and obtains fine-grained confounders at both class-level and domain-level using half-sibling regression and unpaired domain transformation, which essentially identifies domain-variant factors and class-shared factors that cause spurious correlations between features and false labels. To promote VADG research, we collect a VADG-Action dataset for video-audio action recognition with over 5,000 video clips across four domains (e.g., cartoon and game) and ten action classes (e.g., cooking and riding). We conduct extensive experiments, i.e., multi-source DG, single-source DG, and qualitative analysis, validating the rationality of our causal analysis and the effectiveness of the DeVADG framework.

Short video platforms have become an important channel for news sharing, but also a new breeding ground for fake news. To mitigate this problem, research of fake news video detection has recently received a lot of attention. Existing works face two roadblocks: the scarcity of comprehensive and largescale datasets and insufficient utilization of multimodal information. Therefore, in this paper, we construct the largest Chinese short video dataset about fake news named FakeSV, which includes news content, user comments, and publisher profiles simultaneously. To understand the characteristics of fake news videos, we conduct exploratory analysis of FakeSV from different perspectives. Moreover, we provide a new multimodal detection model named SV-FEND, which exploits the cross-modal correlations to select the most informative features and utilizes the social context information for detection. Extensive experiments evaluate the superiority of the proposed method and provide detailed comparisons of different methods and modalities for future works. Our dataset and codes are available in https://github.com/ICTMCG/FakeSV.

Large-scale commonsense knowledge bases empower a broad range of AI applications, where the automatic extraction of commonsense knowledge (CKE) is a fundamental and challenging problem. CKE from text is known for suffering from the inherent sparsity and reporting bias of commonsense in text. Visual perception, on the other hand, contains rich commonsense knowledge about real-world entities, e.g., (person, can_hold, bottle), which can serve as promising sources for acquiring grounded commonsense knowledge. In this work, we present CLEVER, which formulates CKE as a distantly supervised multi-instance learning problem, where models learn to summarize commonsense relations from a bag of images about an entity pair without any human annotation on image instances. To address the problem, CLEVER leverages vision-language pre-training models for deep understanding of each image in the bag, and selects informative instances from the bag to summarize commonsense entity relations via a novel contrastive attention mechanism. Comprehensive experimental results in held-out and human evaluation show that CLEVER can extract commonsense knowledge in promising quality, outperforming pre-trained language model-based methods by 3.9 AUC and 6.4 mAUC points. The predicted commonsense scores show strong correlation with human judgment with a 0.78 Spearman coefficient. Moreover, the extracted commonsense can also be grounded into images with reasonable interpretability. The data and codes can be obtained at https://github.com/thunlp/CLEVER.

In this work, we investigate a more realistic unsupervised multimodal machine translation (UMMT) setup, inference-time image-free UMMT, where the model is trained with sourcetext image pairs, and tested with only sourcetext inputs.First, we represent the input images and texts with the visual and language scene graphs (SG), where such fine-grained visionlanguage features ensure a holistic understanding of the semantics.To enable pure-text input during inference, we devise a visual scene hallucination mechanism that dynamically generates pseudo visual SG from the given textual SG.Several SG-pivoting based learning objectives are introduced for unsupervised translation training.On the benchmark Multi30K data, our SG-based method outperforms the best-performing baseline by significant BLEU scores on the task and setup, helping yield translations with better completeness, relevance and fluency without relying on paired images.Further in-depth analyses reveal how our model advances in the task setting.

Existing research on multimodal relation extraction (MRE) faces two co-existing challenges, internal-information over-utilization and external-information under-exploitation.To combat that, we propose a novel framework that simultaneously implements the idea of internal-information screening and externalinformation exploiting.First, we represent the fine-grained semantic structures of the input image and text with the visual and textual scene graphs, which are further fused into a unified cross-modal graph (CMG).Based on CMG, we perform structure refinement with the guidance of the graph information bottleneck principle, actively denoising the less-informative features.Next, we perform topic modeling over the input image and text, incorporating latent multimodal topic features to enrich the contexts.On the benchmark MRE dataset, our system outperforms the current best model significantly.With further in-depth analyses, we reveal the great potential of our method for the MRE task.Our codes are open at https://github.com/ChocoWu/MRE-ISE.

Unpaired cross-lingual image captioning has long suffered from irrelevancy and disfluency issues, due to the inconsistencies of the semantic scene and syntax attributes during transfer.In this work, we propose to address the above problems by incorporating the scene graph (SG) structures and the syntactic constituency (SC) trees.Our captioner contains the semantic structure-guided image-to-pivot captioning and the syntactic structure-guided pivot-to-target translation, two of which are joined via pivot language.We then take the SG and SC structures as pivoting, performing cross-modal semantic structure alignment and cross-lingual syntactic structure alignment learning.We further introduce cross-lingual&cross-modal back-translation training to fully align the captioning and translation stages.Experiments on English↔Chinese transfers show that our model shows great superiority in improving captioning relevancy and fluency.

Visual spatial description (VSD) aims to generate texts that describe the spatial relations of the given objects within images.Existing VSD work merely models the 2D geometrical vision features, thus inevitably falling prey to the problem of skewed spatial understanding of target objects.In this work, we investigate the incorporation of 3D scene features for VSD.With an external 3D scene extractor, we obtain the 3D objects and scene features for input images, based on which we construct a target object-centered 3D spatial scene graph (GO3D-S 2 G), such that we model the spatial semantics of target objects within the holistic 3D scenes.Besides, we propose a scene subgraph selecting mechanism, sampling topologically-diverse subgraphs from GO3D-S 2 G, where the diverse local structure features are navigated to yield spatially-diversified text generation.Experimental results on two VSD datasets demonstrate that our framework outperforms the baselines significantly, especially improving on the cases with complex visual spatial relations.Meanwhile, our method can produce more spatially-diversified generation.Code is available at https://github.com/zhaoyucs/VSD.

Multi-step stock price prediction over a long-term horizon is crucial for forecasting its volatility, allowing financial institutions to price and hedge derivatives, and banks to quantify the risk in their trading books. Additionally, most financial regulators also require a liquidity horizon of several days for institutional investors to exit their risky assets, in order to not materially affect market prices. However, the task of multi-step stock price prediction is challenging, given the highly stochastic nature of stock data. Current solutions to tackle this problem are mostly designed for single-step, classification-based predictions, and are limited to low representation expressiveness. The problem also gets progressively harder with the introduction of the target price sequence, which also contains stochastic noise and reduces generalizability at test-time.

Collaborative Filtering (CF) has been widely applied for personalized recommendations in various industrial applications. However, due to the training strategy of Empirical Risk Minimization, CF models tend to favor popular items, resulting in inferior performance on sparse users and items. To enhance the CF representation learning of sparse users and items without sacrificing the performance of popular items, we propose a novel Popularity- aware Distributionally Robust Optimization (PDRO) framework. In particular, PDRO emphasizes the optimization of sparse users/items, while incorporating item popularity to preserve the performance of popular items through two modules. First, an implicit module develops a new popularity-aware DRO objective, paying more attention to items that will potentially become popular over time. Second, an explicit module that directly predicts the popularity of items to help the estimation of user-item matching scores. We apply PDRO to a micro-video recommendation scenario and implement it on two representative backend models. Extensive experiments on a real-world industrial dataset, as well as two public benchmark datasets, validate the efficacy of our proposed PDRO. Additionally, we perform an offline A/B test on the industrial dataset, further demonstrating the superiority of PDRO in real-world application scenarios.

Recent research on video moment retrieval has mostly focused on enhancing the performance of accuracy, efficiency, and robustness, all of which largely rely on the abundance of high-quality annotations. While the precise frame-level annotations are time-consuming and cost-expensive, few attentions have been paid to the labeling process. In this work, we explore a new interactive manner to stimulate the process of human-in-the-loop annotation in video moment retrieval task. The key challenge is to select “ambiguous” frames and videos for binary annotations to facilitate the network training. To be specific, we propose a new hierarchical uncertainty-based modeling that explicitly considers modeling the uncertainty of each frame within the entire video sequence corresponding to the query description, and selecting the frame with the highest uncertainty. Only selected frame will be annotated by the human experts, which can largely reduce the workload. After obtaining a small number of labels provided by the expert, we show that it is sufficient to learn a competitive video moment retrieval model in such a harsh environment. Moreover, we treat the uncertainty score of frames in a video as a whole, and estimate the difficulty of each video, which can further relieve the burden of video selection. In general, our active learning strategy for video moment retrieval works not only at the frame level but also at the sequence level. Experiments on two public datasets validate the effectiveness of our proposed method. Our code is released at https://github.com/renjie-liang/HUAL.

Large language models (LLMs) outperform information retrieval techniques for downstream knowledge-intensive tasks when being prompted to generate world knowledge.Yet, community concerns abound regarding the factuality and potential implications of using this uncensored knowledge.In light of this, we introduce CONNER, a COmpreheNsive kNowledge Evaluation fRamework, designed to systematically and automatically evaluate generated knowledge from six important perspectives -Factuality, Relevance, Coherence, Informativeness, Helpfulness and Validity.We conduct an extensive empirical analysis of the generated knowledge from three different types of LLMs on two widely-studied knowledge-intensive tasks, i.e., open-domain question answering and knowledge-grounded dialogue.Surprisingly, our study reveals that the factuality of generated knowledge, even if lower, does not significantly hinder downstream tasks.Instead, the relevance and coherence of the outputs are more important than small factual mistakes.Further, we show how to use CONNER to improve knowledge-intensive tasks by designing two strategies: Prompt Engineering and Knowledge Selection.Our evaluation code and LLMgenerated knowledge with human annotations will be released 1 to facilitate future research.

Large Language Model (LLM) has demonstrated significant ability in various Natural Language Processing tasks.However, their effectiveness is highly dependent on the phrasing of the task prompt, leading to research on automatic prompt optimization using labeled task data.We reveal that these prompt optimization techniques are vulnerable to distribution shifts such as subpopulation shifts, which are common for LLMs in real-world scenarios such as customer reviews analysis.In this light, we propose a new problem of robust prompt optimization for LLMs against distribution shifts, which requires the prompt optimized over the labeled source group can simultaneously generalize to an unlabeled target group.To solve this problem, we propose Generalized Prompt Optimization framework , which incorporates the unlabeled data from the target group into prompt optimization.Extensive experimental results demonstrate the effectiveness of the proposed framework with significant performance improvement on the target group and comparable performance on the source group.

Language Models (LMs) have demonstrated impressive molecule understanding ability on various 1D text-related tasks.However, they inherently lack 2D graph perception -a critical ability of human professionals in comprehending molecules' topological structures.To bridge this gap, we propose MolCA: Molecular Graph-Language Modeling with Cross-Modal Projector and Uni-Modal Adapter.MolCA enables an LM (i.e., Galactica) to understand both text-and graph-based molecular contents via the cross-modal projector.Specifically, the cross-modal projector is implemented as a Q-Former to connect a graph encoder's representation space and an LM's text space.Further, MolCA employs a uni-modal adapter (i.e., LoRA) for the LM's efficient adaptation to downstream tasks.Unlike previous studies that couple an LM with a graph encoder via cross-modal contrastive learning, MolCA retains the LM's ability of open-ended text generation and augments it with 2D graph information.To showcase its effectiveness, we extensively benchmark MolCA on tasks of molecule captioning, IUPAC name prediction, and molecule-text retrieval, on which MolCA significantly outperforms the baselines.Our codes and checkpoints can be found at https: //github.com/acharkq/MolCA.

This paper strives to solve complex video question answering (VideoQA) which features long video containing multiple objects and events at different time. To tackle the challenge, we highlight the importance of identifying question-critical temporal moments and spatial objects from the vast amount of video content. Towards this, we propose a Spatio-Temporal Rationalization (STR), a differentiable selection module that adaptively collects question-critical moments and objects using cross-modal interaction. The discovered video moments and objects are then served as grounded rationales to support answer reasoning. Based on STR, we further propose TranSTR, a Transformerstyle neural network architecture that takes STR as the core and additionally underscores a novel answer interaction mechanism to coordinate STR for answer decoding. Experiments on four datasets show that TranSTR achieves new state-of-the-art (SoTA). Especially, on NExT-QA and Causal-VidQA which feature complex VideoQA, it significantly surpasses the previous SoTA by 5.8% and 6.8%, respectively. We then conduct extensive studies to verify the importance of STR as well as the proposed answer interaction mechanism. With the success of TranSTR and our comprehensive analysis, we hope this work can spark more future efforts in complex VideoQA. Code will be released at https://github.com/yl3800/TranSTR.

Prompt tuning, a recently emerging paradigm, enables the powerful vision-language pre-training models to adapt to downstream tasks in a parameter- and data- efficient way, by learning the "soft prompts" to condition frozen pretraining models. Though effective, it is particularly problematic in the few-shot scenario, where prompt tuning performance is sensitive to the initialization and requires a time-consuming process to find a good initialization, thus restricting the fast adaptation ability of the pre-training models. In addition, prompt tuning could undermine the generalizability of the pre-training models, because the learnable prompt tokens are easy to overfit to the limited training samples. To address these issues, we introduce a novel Gradient-RegulAted Meta-prompt learning (GRAM) framework that jointly meta-learns an efficient soft prompt initialization for better adaptation and a lightweight gradient regulating function for strong cross-domain generalizability in a meta-learning paradigm using only the unlabeled image-text pre-training data. Rather than designing a specific prompt tuning method, our GRAM can be easily incorporated into various prompt tuning methods in a model-agnostic way, and comprehensive experiments show that GRAM brings about consistent improvement for them in several settings (i.e., few-shot learning, cross-domain generalization, cross-dataset generalization, etc.) over 11 datasets. Further, experiments show that GRAM enables the orthogonal methods of textual and visual prompt tuning to work in a mutually-enhanced way, offering better generalizability beyond the uni-modal prompt tuning methods.

Proactive dialogue systems, related to a wide range of real-world conversational applications, equip the conversational agent with the capability of leading the conversation direction towards achieving pre-defined targets or fulfilling certain goals from the system side. It is empowered by advanced techniques to progress to more complicated tasks that require strategical and motivational interactions. In this survey, we provide a comprehensive overview of the prominent problems and advanced designs for conversational agent's proactivity in different types of dialogues. Furthermore, we discuss challenges that meet the real-world application needs but require a greater research focus in the future. We hope that this first survey of proactive dialogue systems can provide the community with a quick access and an overall picture to this practical problem, and stimulate more progresses on conversational AI to the next level.

Graph Neural Networks (GNNs) have achieved remarkable success in various domains but most of them are developed under the in-distribution assumption. Under out-of-distribution (OOD) settings, they suffer from the distribution shift between the training set and the test set and may not generalize well to the test distribution. Several methods have tried the invariance principle to improve the generalization of GNNs in OOD settings. However, in previous solutions, the graph encoder is immutable after the invariant learning and cannot be adapted to the target distribution flexibly. Confronting the distribution shift, a flexible encoder with refinement to the target distribution can generalize better on the test set than the stable invariant encoder. To remedy these weaknesses, we propose a Flexible invariant Learning framework for Out-Of-Distribution generalization on graphs (FLOOD), which comprises two key components, invariant learning and bootstrapped learning. The invariant learning component constructs multiple environments from graph data augmentation and learns invariant representation under risk extrapolation. Besides, the bootstrapped learning component is devised to be trained in a self-supervised way with a shared graph encoder with the invariant learning part. During the test phase, the shared encoder is flexible to be refined with the bootstrapped learning on the test set. Extensive experiments are conducted for both transductive and inductive node classification tasks. The results demonstrate that FLOOD consistently outperforms other graph OOD generalization methods and effectively improves the generalization ability.

Discovering causal structure from purely observational data (i.e., causal discovery), aiming to identify causal relationships among variables, is a fundamental task in machine learning.The recent invention of differentiable score-based DAG learners is a crucial enabler, which reframes the combinatorial optimization problem into a differentiable optimization with a DAG constraint over directed graph space. Despite their great success, these cutting-edge DAG learners incorporate DAG-ness independent score functions to evaluate the directed graph candidates, lacking in considering graph structure. As a result, measuring the data fitness alone regardless of DAG-ness inevitably leads to discovering suboptimal DAGs and model vulnerabilities.

Event forecasting has been a demanding and challenging task throughout the entire human history. It plays a pivotal role in crisis alarming and disaster prevention in various aspects of the whole society. The task of event forecasting aims to model the relational and temporal patterns based on historical events and makes forecasting to what will happen in the future. Most existing studies on event forecasting formulate it as a problem of link prediction on temporal event graphs. However, such pure structured formulation suffers from two main limitations: 1) most events fall into general and high-level types in the event ontology, and therefore they tend to be coarse-grained and offers little utility which inevitably harms the forecasting accuracy; and 2) the events defined by a fixed ontology are unable to retain the out-of-ontology contextual information.

This paper identifies two kinds of redundancy in the current VideoQA paradigm. Specifically, the current video encoders tend to holistically embed all video clues at different granularities in a hierarchical manner, which inevitably introducesneighboring-frame redundancy that can overwhelm detailed visual clues at the object level. Subsequently, prevailing vision-language fusion designs introduce thecross-modal redundancy by exhaustively fusing all visual elements with question tokens without explicitly differentiating their pairwise vision-language interactions, thus making a pernicious impact on the answering. To this end, we propose a novel transformer-based architecture, that aims to model VideoQA in a redundancy-aware manner. To address the neighboring-frame redundancy, we introduce a video encoder structure that emphasizes the object-level change in neighboring frames, while adopting an out-of-neighboring message-passing scheme that imposes attention only on distant frames. As for the cross-modal redundancy, we equip our fusion module with a novel adaptive sampling, which explicitly differentiates the vision-language interactions by identifying a small subset of visual elements that exclusively support the answer. Upon these advancements, we find this \underlineR edundancy-\underlinea ware trans\underlineformer (RaFormer) can achieve state-of-the-art results on multiple VideoQA benchmarks.

Information retrieval (IR) is a fundamental technique that aims to acquire information from a collection of documents, web pages, or other sources. While traditional text-based IR has achieved great success, the under-utilization of varied data sources in different modalities (i.e., text, images, audio, and video) would hinder IR techniques from giving its full advancement and thus limits the applications in the real world. Within recent years, the rapid development of deep multimodal learning paves the way for advancing IR with multi-modality. Benefiting from a variety of data types and modalities, some latest prevailing techniques are invented to show great facilitation in multi-modal and IR learning, such as CLIP, ChatGPT, GPT4, etc. In the context of IR, deep multi-modal learning has shown the prominent potential to improve the performance of retrieval systems, by enabling them to better understand and process the diverse types of data that they encounter. Given the great potential shown by multimodal-empowered IR, there can be still unsolved challenges and open questions in the related directions. With this workshop, we aim to provide a platform for discussion about multi-modal IR among scholars, practitioners, and other interested parties.

Unsupervised Domain Adaptation (UDA) aims to enhance the generalization of the learned model to other domains. The domain-invariant knowledge is transferred from the model trained on labeled source domain, e.g., video game, to unlabeled target domains, e.g., real-world scenarios, saving annotation expenses. Existing UDA methods for semantic segmentation usually focus on minimizing the inter-domain discrepancy of various levels, e.g., pixels, features, and predictions, for extracting domain-invariant knowledge. However, the primary intra-domain knowledge, such as context correlation inside an image, remains under-explored. In an attempt to fill this gap, we revisit the current pixel contrast in semantic segmentation and propose a unified pixel- and patch-wise self-supervised learning framework, called PiPa, for domain adaptive semantic segmentation that facilitates intra-image pixel-wise correlations and patch-wise semantic consistency against different contexts. The proposed framework exploits the inherent structures of intra-domain images, which: (1) explicitly encourages learning the discriminative pixel-wise features with intra-class compactness and inter-class separability, and (2) motivates the robust feature learning of the identical patch against different contexts or fluctuations. Extensive experiments verify the effectiveness of the proposed method, which obtains competitive accuracy on the two widely-used UDA benchmarks, e.g., 75.6 mIoU on GTA→Cityscapes and 68.2 mIoU on Synthia→Cityscapes. Moreover, our method is compatible with other UDA approaches to further improve the performance without introducing extra parameters.

It has been a hot research topic to enable machines to understand human emotions in multimodal contexts under dialogue scenarios, which is tasked with multimodal emotion analysis in conversation (MM-ERC). MM-ERC has received consistent attention in recent years, where a diverse range of methods has been proposed for securing better task performance. Most existing works treat MM-ERC as a standard multimodal classification problem and perform multimodal feature disentanglement and fusion for maximizing feature utility. Yet after revisiting the characteristic of MM-ERC, we argue that both the feature multimodality and conversational contextualization should be properly modeled simultaneously during the feature disentanglement and fusion steps. In this work, we target further pushing the task performance by taking full consideration of the above insights. On the one hand, during feature disentanglement, based on the contrastive learning technique, we devise a Dual-level Disentanglement Mechanism (DDM) to decouple the features into both the modality space and utterance space. On the other hand, during the feature fusion stage, we propose a Contribution-aware Fusion Mechanism (CFM) and a Context Refusion Mechanism (CRM) for multimodal and context integration, respectively. They together schedule the proper integrations of multimodal and context features. Specifically, CFM explicitly manages the multimodal feature contributions dynamically, while CRM flexibly coordinates the introduction of dialogue contexts. On two public MM-ERC datasets, our system achieves new state-of-the-art performance consistently. Further analyses demonstrate that all our proposed mechanisms greatly facilitate the MM-ERC task by making full use of the multimodal and context features adaptively. Note that our proposed methods have the great potential to facilitate a broader range of other conversational multimodal tasks.

In the text-to-image generation field, recent remarkable progress in Stable Diffusion makes it possible to generate rich kinds of novel photorealistic images. However, current models still face misalignment issues (e.g., problematic spatial relation understanding and numeration failure) in complex natural scenes, which impedes the high-faithfulness text-to-image generation. Although recent efforts have been made to improve controllability by giving fine-grained guidance (e.g., sketch and scribbles), this issue has not been fundamentally tackled since users have to provide such guidance information manually. In this work, we strive to synthesize high-fidelity images that are semantically aligned with a given textual prompt without any guidance. Toward this end, we propose a coarse-to-fine paradigm to achieve layout planning and image generation. Concretely, we first generate the coarse-grained layout conditioned on a given textual prompt via in-context learning based on Large Language Models. Afterward, we propose a fine-grained object-interaction diffusion method to synthesize high-faithfulness images conditioned on the prompt and the automatically generated layout. Extensive experiments demonstrate that our proposed method outperforms the state-of-the-art models in terms of layout and image generation. Our code and settings are available at https://layoutllm-t2i.github.io/.

Recommender systems rely on user-item interactions to learn Collaborative Filtering (CF) signals and easily under-recommend the cold-start items without historical interactions. To boost cold-start item recommendation, previous studies usually incorporate item features (e.g., micro-video content features) into CF models. They essentially align the feature representations of warm-start items with CF representations during training, and then adopt the feature representations of cold-start items to make recommendations. However, cold-start items might have feature distribution shifts from warm-start ones due to different upload times. As such, these cold-start item features fall into the underrepresented feature space, where their feature representations cannot align well with CF signals, causing poor cold-start recommendation.

A large language model is a type of artificial intelligence model designed to understand and generate natural language text, such as GPT, T5, RoBERTa, BERT, etc. These models are trained on vast amounts of text data, allowing them to learn the patterns and structures of human language. With the increasing amount of multimodal information such as audio, visual, and text data generated, there is a growing need of leveraging large generative language model for multimodal applications. Recently, a few notable multimodal models (e.g., BLIP, Flamingo, KOSMOS, PaLM-E, LLaVA, Visual ChatGPT, GPT-4, etc.) with a combination of large language models significantly enhanced their understanding and generate more accurate and nuanced responses. The workshop will provide an opportunity for researchers, practitioners, and industry professionals to explore the latest trends and best practices in the field of multimodal applications of large generative models. The workshop will also focus on exploring the challenges and opportunities of integrating large language models with other AI technologies such as computer vision and speech recognition.

The task of spatio-temporal video grounding involves identifying the spatial and temporal regions in a video that correspond to the objects or actions described in a given textual description. However, current models used for spatio-temporal video grounding often rely heavily on spatio-temporal priors to make the predictions. As a result, they may suffer from spurious correlations and lack the ability to generalize well to new or diverse scenarios. To overcome this limitation, we introduce a deconfounded multimodal learning framework, which utilizes a structural causal model to treat dataset biases as a confounder and subsequently remove their confounding effect. Through this framework, we can perform causal intervention on the multimodal input and derive an unbiased estimation formula through the do-calculus technique. In order to tackle the challenge of diverse and often unobservable confounders, we further propose a novel retrieval-based approach with a causal mask mechanism. The proposed method leverages analogical reasoning to facilitate deconfounded learning and mitigate dataset biases, enabling unbiased spatio-temporal prediction without explicitly modeling the confounding factors. Extensive experiments on two challenging benchmarks have well verified the effectiveness and rationality of our proposed solution.

As one of the core video semantic understanding tasks, Video Semantic Role Labeling (VidSRL) aims to detect the salient events from given videos, by recognizing the predict-argument event structures and the interrelationships between events. While recent endeavors have put forth methods for VidSRL, they can be mostly subject to two key drawbacks, including the lack of fine-grained spatial scene perception and the insufficiently modeling of video temporality. Towards this end, this work explores a novel holistic spatio-temporal scene graph (namely HostSG) representation based on the existing dynamic scene graph structures, which well model both the fine-grained spatial semantics and temporal dynamics of videos for VidSRL. Built upon the HostSG, we present a nichetargeting VidSRL framework. A scene-event mapping mechanism is first designed to bridge the gap between the underlying scene structure and the high-level event semantic structure, resulting in an overall hierarchical scene-event (termed ICE) graph structure. We further perform iterative structure refinement to optimize the ICE graph, e.g., filtering noisy branches and newly building informative connections, such that the overall structure representation can best coincide with end task demand. Finally, three subtask predictions of VidSRL are jointly decoded, where the end-to-end paradigm effectively avoids error propagation. On the benchmark dataset, our framework boosts significantly over the current best-performing model. Further analyses are shown for a better understanding of the advances of our methods. Our HostSG representation shows greater potential to facilitate a broader range of other video understanding tasks.

Unmanned Aerial Vehicles (UAVs), also known as drones, have become increasingly popular in recent years due to their ability to capture high-quality multimedia data from the sky. With the rise of multimedia applications, such as aerial photography, cinematography, and mapping, UAVs have emerged as a powerful tool for gathering rich and diverse multimedia content. This workshop aims to bring together researchers, practitioners, and enthusiasts interested in UAV multimedia to explore the latest advancements, challenges, and opportunities in this exciting field. The workshop covers various topics related to UAV multimedia, including aerial image and video processing, machine learning for UAV data analysis, UAV swarm technology, and UAV-based multimedia applications. In the context of the ACM Multimedia conference, this workshop is highly relevant as multimedia data from UAVs is becoming an increasingly important source of content for many multimedia applications. The workshop provides a platform for researchers to share their work and discuss potential collaborations, as well as an opportunity for practitioners to learn about the latest developments in UAV multimedia technology. Overall, this workshop provides a unique opportunity to explore the exciting and rapidly evolving field of UAV multimedia and its potential impact on the wider multimedia community.

Masked graph modeling excels in the self-supervised representation learning of molecular graphs. Scrutinizing previous studies, we can reveal a common scheme consisting of three key components: (1) graph tokenizer, which breaks a molecular graph into smaller fragments (\ie subgraphs) and converts them into tokens; (2) graph masking, which corrupts the graph with masks; (3) graph autoencoder, which first applies an encoder on the masked graph to generate the representations, and then employs a decoder on the representations to recover the tokens of the original graph. However, the previous MGM studies focus extensively on graph masking and encoder, while there is limited understanding of tokenizer and decoder. To bridge the gap, we first summarize popular molecule tokenizers at the granularity of node, edge, motif, and Graph Neural Networks (GNNs), and then examine their roles as the MGM's reconstruction targets. Further, we explore the potential of adopting an expressive decoder in MGM. Our results show that a subgraph-level tokenizer and a sufficiently expressive decoder with remask decoding have a \yuan{large impact on the encoder's representation learning}. Finally, we propose a novel MGM method SimSGT, featuring a Simple GNN-based Tokenizer (SGT) and an effective decoding strategy. We empirically validate that our method outperforms the existing molecule self-supervised learning methods. Our codes and checkpoints are available at https://github.com/syr-cn/SimSGT.

In this work, we investigate the task of text-to-image (T2I) synthesis under the abstract-to-intricate setting, i.e., generating intricate visual content from simple abstract text prompts. Inspired by human imagination intuition, we propose a novel scene-graph hallucination (SGH) mechanism for effective abstract-to-intricate T2I synthesis. SGH carries out scene hallucination by expanding the initial scene graph (SG) of the input prompt with more feasible specific scene structures, in which the structured semantic representation of SG ensures high controllability of the intrinsic scene imagination. To approach the T2I synthesis, we deliberately build an SG-based hallucination diffusion system. First, we implement the SGH module based on the discrete diffusion technique, which evolves the SG structure by iteratively adding new scene elements. Then, we utilize another continuous-state diffusion model as the T2I synthesizer, where the overt image-generating process is navigated by the underlying semantic scene structure induced from the SGH module. On the benchmark COCO dataset, our system outperforms the existing best-performing T2I model by a significant margin, especially improving on the abstract-to-intricate T2I generation. Further in-depth analyses reveal how our methods advance.

Since developing a new multimodal LLM (MLLM) by pre-training on tremendous image-text pairs from scratch can be exceedingly resource-consuming, connecting an existing LLM with a comparatively lightweight visual prompt generator (VPG) becomes a feasible paradigm. However, further tuning the VPG component of the MLLM still incurs significant computational costs, such as thousands of GPU hours and millions of training data points. An alternative solution is transferring an existing VPG from one MLLM to the target MLLM. In this work, we investigate VPG transferability across LLMs for the first time, aiming to reduce the cost of VPG training. Specifically, we explore VPG transfer across different LLM sizes (e.g., small-to-large) and types. We identify key factors to maximize transfer efficiency, based on which we develop a simple yet highly effective two-stage transfer framework, called VPGTrans. Notably, it enables VPG transfer from BLIP-2 OPT 2.7B to BLIP-2 OPT 6.7B with less than 10% of the GPU hours using only 10.7% of the training data compared to training a VPG for OPT 6.7B from scratch. Furthermore, we provide a series of intriguing findings and discuss potential explanations behind them. Finally, we showcase the practical value of our VPGTrans approach, by customizing two novel MLLMs, including VL-LLaMA and VL-Vicuna, with recently released LLaMA and Vicuna LLMs.

Contrastive Learning (CL) has achieved impressive performance in self-supervised learning tasks, showing superior generalization ability. Inspired by the success, adopting CL into collaborative filtering (CF) is prevailing in semi-supervised topK recommendations. The basic idea is to routinely conduct heuristic-based data augmentation and apply contrastive losses (e.g., InfoNCE) on the augmented views. Yet, some CF-tailored challenges make this adoption suboptimal, such as the issue of out-of-distribution, the risk of false negatives, and the nature of top-K evaluation. They necessitate the CL-based CF scheme to focus more on mining hard negatives and distinguishing false negatives from the vast unlabeled user-item interactions, for informative contrast signals. Worse still, there is limited understanding of contrastive loss in CF methods, especially w.r.t. its generalization ability. To bridge the gap, we delve into the reasons underpinning the success of contrastive loss in CF, and propose a principled Adversarial InfoNCE loss (AdvInfoNCE), which is a variant of InfoNCE, specially tailored for CF methods. AdvInfoNCE adaptively explores and assigns hardness to each negative instance in an adversarial fashion and further utilizes a fine-grained hardness-aware ranking criterion to empower the recommender’s generalization ability. Training CF models with AdvInfoNCE, we validate the effectiveness of AdvInfoNCE on both synthetic and real-world benchmark datasets, thus showing its generalization ability to mitigate out-of-distribution problems. Given the theoretical guarantees and empirical superiority of AdvInfoNCE over most contrastive loss functions, we advocate its adoption as a standard loss in recommender systems, particularly for the out-of-distribution tasks. Codes are available at https://github.com/LehengTHU/AdvInfoNCE.

Textual response generation is an essential task for multimodal task-oriented dialog systems. Although existing studies have achieved fruitful progress, they still suffer from two critical limitations: 1) focusing on the attribute knowledge but ignoring the relation knowledge that can reveal the correlations between different entities and hence promote the response generation, and 2)only conducting the cross-entropy loss based output-level supervision but lacking the representation-level regularization. To address these limitations, we devise a novel multimodal task-oriented dialog system (named MDS-S2). Specifically, MDS-S2 first simultaneously acquires the context related attribute and relation knowledge from the knowledge base, whereby the non-intuitive relation knowledge is extracted by the n-hop graph walk. Thereafter, considering that the attribute knowledge and relation knowledge can benefit the responding to different levels of questions, we design a multi-level knowledge composition module in MDS-S^2 to obtain the latent composed response representation. Moreover, we devise a set of latent query variables to distill the semantic information from the composed response representation and the ground truth response representation, respectively, and thus conduct the representation-level semantic regularization. Extensive experiments on a public dataset have verified the superiority of our proposed MDS-S2. We have released the codes and parameters to facilitate the research community.

Image-text retrieval aims to bridge the modality gap and retrieve cross-modal content based on semantic similarities. Prior work usually focuses on the pairwise relations (i.e., whether a data sample matches another) but ignores the higher-order neighbor relations (i.e., a matching structure among multiple data samples). Re-ranking, a popular post-processing practice, has revealed the superiority of capturing neighbor relations in single-modality retrieval tasks. However, it is ineffective to directly extend existing re-ranking algorithms to image-text retrieval. In this paper, we analyze the reason from four perspectives, i.e., generalization, flexibility, sparsity, and asymmetry, and propose a novel learnable pillar-based re-ranking paradigm. Concretely, we first select top-ranked intra- and intermodal neighbors as pillars, and then reconstruct data samples with the neighbor relations between them and the pillars. In this way, each sample can be mapped into a multimodal pillar space only using similarities, ensuring generalization. After that, we design a neighbor-aware graph reasoning module to flexibly exploit the relations and excavate the sparse positive items within a neighborhood. We also present a structure alignment constraint to promote crossmodal collaboration and align the asymmetric modalities. On top of various base backbones, we carry out extensive experiments on two benchmark datasets, i.e., Flickr30K and MS-COCO, demonstrating the effectiveness, superiority, generalization, and transferability of our proposed re-ranking paradigm.

Generative models such as Generative Adversarial Networks (GANs) and Variational Auto-Encoders (VAEs) are widely utilized to model the generative process of user interactions. However, they suffer from intrinsic limitations such as the instability of GANs and the restricted representation ability of VAEs. Such limitations hinder the accurate modeling of the complex user interaction generation procedure, such as noisy interactions caused by various interference factors. In light of the impressive advantages of Diffusion Models (DMs) over traditional generative models in image synthesis, we propose a novel Diffusion Recommender Model (named DiffRec) to learn the generative process in a denoising manner. To retain personalized information in user interactions, DiffRec reduces the added noises and avoids corrupting users' interactions into pure noises like in image synthesis. In addition, we extend traditional DMs to tackle the unique challenges in recommendation: high resource costs for large-scale item prediction and temporal shifts of user preference. To this end, we propose two extensions of DiffRec: L-DiffRec clusters items for dimension compression and conducts the diffusion processes in the latent space; and T-DiffRec reweights user interactions based on the interaction timestamps to encode temporal information. We conduct extensive experiments on three datasets under multiple settings (e.g., clean training, noisy training, and temporal training). The empirical results validate the superiority of DiffRec with two extensions over competitive baselines.

Multimedia recommendation methods aim to discover the user preference on the multi-modal information to enhance the collaborative filtering (CF) based recommender system. Nevertheless, they seldom consider the impact of feature extraction on the user preference modeling and prediction of the user-item interaction, as the extracted features contain excessive information irrelevant to the recommendation.

A bundle is a group of items that provides improved services to users and increased profits for sellers. However, locating the desired bundles that match the users' tastes still challenges us, due to the sparsity issue. Despite the remarkable performance of existing approaches, we argue that they seldom consider the bundling strategy (i.e., how the items within a bundle are associated with each other) in the bundle recommendation, resulting in the suboptimal user and bundle representations for their interaction prediction. Therefore, we propose to model the strategy-aware user and bundle representations for the bundle recommendation.

With the growing success of graph neural networks (GNNs), the explainability of GNN is attracting considerable attention. Current explainers mostly leverage feature attribution and selection to explain a prediction. By tracing the importance of input features, they select the salient subgraph as the explanation. However, their explainability is at the granularity of input features only, and cannot reveal the usefulness of hidden neurons. This inherent limitation makes the explainers fail to scrutinize the model behavior thoroughly, resulting in unfaithful explanations.

Dialogue State Tracking (DST) module is an essential component of task-oriented dialog systems to understand users’ goals and needs. Collecting dialogue state labels including slots and values can be costly, requiring experts to annotate all (slot, value) information for each turn in dialogues. It is also difficult to define all possible slots and values in advance, especially with the wide application of dialogue systems in more and more new-rising applications. In this paper, we focus on improving DST module to generate dialogue states in circumstances with limited annotations and knowledge about slot ontology. To this end, we design a dual prompt learning framework for few-shot DST. The dual framework aims to explore how to utilize the language understanding and generation capabilities of pre-trained language models for DST efficiently. Specifically, we consider the learning of slot generation and value generation as dual tasks, and two kinds of prompts are designed based on this dual structure to incorporate task-related knowledge of these two tasks respectively. In this way, the DST task can be formulated as a language modeling task efficiently under few-shot settings. To evaluate the proposed framework, we conduct experiments on two task-oriented dialogue datasets. The results demonstrate that the proposed method not only outperforms existing state-of-the-art few-shot methods, but also can generate unseen slots. It indicates that DST-related knowledge can be probed from pre-trained language models and utilized to address low-resource DST efficiently with the help of prompt learning.

Collaborative Filtering (CF) models, despite their great success, suffer from severe performance drops due to popularity distribution shifts, where these changes are ubiquitous and inevitable in real-world scenarios. Unfortunately, most leading popularity debiasing strategies, rather than tackling the vulnerability of CF models to varying popularity distributions, require prior knowledge of the test distribution to identify the degree of bias and further learn the popularity-entangled representations to mitigate the bias. Consequently, these models result in significant performance benefits in the target test set, while dramatically deviating the recommendation from users’ true interests without knowing the popularity distribution in advance. In this work, we propose a novel learning framework, Invariant Collaborative Filtering (InvCF), to discover disentangled representations that faithfully reveal the latent preference and popularity semantics without making any assumption about the popularity distribution. At its core is the distillation of unbiased preference representations (i.e., user preference on item property), which are invariant to the change of popularity semantics, while filtering out the popularity feature that is unstable or outdated. Extensive experiments on five benchmark datasets and four evaluation settings (i.e., synthetic long-tail, unbiased, temporal split, and out-of-distribution evaluations) demonstrate that InvCF outperforms the state-of-the-art baselines in terms of popularity generalization ability on real recommendations. Visualization studies shed light on the advantages of InvCF for disentangled representation learning. Our codes are available at https://github.com/anzhang314/InvCF.

Abstract This paper studies compositional 3D-aware image synthesis for both single-object and multi-object scenes. We observe that two challenges remain in this field: existing approaches (1) lack geometry constraints and thus compromise the multi-view consistency of the single object, and (2) can not scale to multi-object scenes with complex backgrounds. To address these challenges coherently, we propose multi-view consistent generative adversarial networks (MVCGAN) for compositional 3D-aware image synthesis. First, we build the geometry constraints on the single object by leveraging the underlying 3D information. Specifically, we enforce the photometric consistency between pairs of views, encouraging the model to learn the inherent 3D shape. Second, we adapt MVCGAN to multi-object scenarios. In particular, we formulate the multi-object scene generation as a “decompose and compose” process. During training, we adopt the top-down strategy to decompose training images into objects and backgrounds. When rendering, we deploy a reverse bottom-up manner by composing the generated objects and background into the holistic scene. Extensive experiments on both single-object and multi-object datasets show that the proposed method achieves competitive performance for 3D-aware image synthesis.

Temporal grounding is the task of locating a specific segment from an untrimmed video according to a query sentence. This task has achieved significant momentum in the computer vision community as it enables activity grounding beyond pre-defined activity classes by utilizing the semantic diversity of natural language descriptions. The semantic diversity is rooted in the principle of compositionality in linguistics, where novel semantics can be systematically described by combining known words in novel ways (compositional generalization). However, existing temporal grounding datasets are not carefully designed to evaluate the compositional generalizability. To systematically benchmark the compositional generalizability of temporal grounding models, we introduce a new Compositional Temporal Grounding task and construct two new dataset splits, i.e., Charades-CG and ActivityNet-CG. We empirically find that they fail to generalize to queries with novel combinations of seen words. We argue that the inherent compositional structure (i.e., composition constituents and their relationships) inside the videos and language is the crucial factor to achieve compositional generalization. Based on this insight, we propose a variational cross-graph reasoning framework that explicitly decomposes video and language into hierarchical semantic graphs, respectively, and learns fine-grained semantic correspondence between the two graphs. Meanwhile, we introduce a novel adaptive structured semantics learning approach to derive the structure-informed and domain-generalizable graph representations, which facilitate the fine-grained semantic correspondence reasoning between the two graphs. To further evaluate the understanding of the compositional structure, we also introduce a more challenging setting, where one of the components in the novel composition is unseen. This requires more sophisticated understanding of the compositional structure to infer the potential semantics of the unseen word based on the other learned composition constituents appearing in both the video and language context, and their relationships. Extensive experiments validate the superior compositional generalizability of our approach, demonstrating its ability to handle queries with novel combinations of seen words as well as novel words in the testing composition.

Explainability is crucial for probing graph neural networks (GNNs), answering questions like "Why the GNN model makes a certain prediction?". Feature attribution is a prevalent technique of highlighting the explanatory subgraph in the input graph, which plausibly leads the GNN model to make its prediction. Various attribution methods have been proposed to exploit gradient-like or attention scores as the attributions of edges, then select the salient edges with top attribution scores as the explanation. However, most of these works make an untenable assumption - the selected edges are linearly independent - thus leaving the dependencies among edges largely unexplored, especially their coalition effect. We demonstrate unambiguous drawbacks of this assumption - making the explanatory subgraph unfaithful and verbose. To address this challenge, we propose a reinforcement learning agent, Reinforced Causal Explainer (RC-Explainer). It frames the explanation task as a sequential decision process - an explanatory subgraph is successively constructed by adding a salient edge to connect the previously selected subgraph. Technically, its policy network predicts the action of edge addition, and gets a reward that quantifies the action's causal effect on the prediction. Such reward accounts for the dependency of the newly-added edge and the previously-added edges, thus reflecting whether they collaborate together and form a coalition to pursue better explanations. It is trained via policy gradient to optimize the reward stream of edge sequences. As such, RC-Explainer is able to generate faithful and concise explanations, and has a better generalization power to unseen graphs. When explaining different GNNs on three graph classification datasets, RC-Explainer achieves better or comparable performance to state-of-the-art approaches w.r.t. two quantitative metrics: predictive accuracy, contrastivity, and safely passes sanity checks and visual inspections. Codes and datasets are available at https://github.com/xiangwang1223/reinforced_causal_explainer.

We propose to perform video question answering (VideoQA) in a Contrastive manner via a Video Graph Transformer model (CoVGT). CoVGT's uniqueness and superiority are three-fold: 1) It proposes a dynamic graph transformer module which encodes video by explicitly capturing the visual objects, their relations and dynamics, for complex spatio-temporal reasoning. 2) It designs separate video and text transformers for contrastive learning between the video and text to perform QA, instead of multi-modal transformer for answer classification. Fine-grained video-text communication is done by additional cross-modal interaction modules. 3) It is optimized by the joint fully- and self-supervised contrastive objectives between the correct and incorrect answers, as well as the relevant and irrelevant questions respectively. With superior video encoding and QA solution, we show that CoVGT can achieve much better performances than previous arts on video reasoning tasks. Its performances even surpass those models that are pretrained with millions of external data. We further show that CoVGT can also benefit from cross-modal pretraining, yet with orders of magnitude smaller data. The results demonstrate the effectiveness and superiority of CoVGT, and additionally reveal its potential for more data-efficient pretraining.

In recommender systems, users' behavior data are driven by the interactions of user-item latent factors. To improve recommendation effectiveness and robustness, recent advances focus on latent factor disentanglement via variational inference. Despite significant progress, uncovering the underlying interactions, i.e., dependencies of latent factors, remains largely neglected by the literature. To bridge the gap, we investigate the joint disentanglement of user-item latent factors and the dependencies between them, namely latent structure learning. We propose to analyze the problem from the causal perspective, where a latent structure should ideally reproduce observational interaction data, and satisfy the structure acyclicity and dependency constraints, i.e., causal prerequisites. We further identify the recommendation-specific challenges for latent structure learning, i.e., the subjective nature of users' minds and the inaccessibility of private/sensitive user factors causing universally learned latent structure to be suboptimal for individuals. To address these challenges, we propose the personalized latent structure learning framework for recommendation, namely PlanRec, which incorporates 1) differentiable Reconstruction, Dependency, and Acyclicity regularizations to satisfy the causal prerequisites; 2) Personalized Structure Learning (PSL) which personalizes the universally learned dependencies through probabilistic modeling; and 3) uncertainty estimation which explicitly measures the uncertainty of structure personalization, and adaptively balances personalization and shared knowledge for different users. We conduct extensive experiments on two public benchmark datasets from MovieLens and Amazon, and a large-scale industrial dataset from Alipay. Empirical studies validate that PlanRec discovers effective shared/personalized structures, and successfully balances shared knowledge and personalization via rational uncertainty estimation.

Relation Extraction (RE) is a vital step to complete Knowledge Graph (KG) by extracting entity relations from texts. However, it usually suffers from the long-tail issue. This paper proposes a novel approach to learn relation prototypes from unlabeled texts, to facilitate long-tail RE by transferring knowledge from relation types with sufficient training data. We learn relation prototypes as an implicit factor between entities, which reflects meanings of relations and their proximities. We construct a co-occurrence graph from texts, and capture both first-order and second-order entity proximities for embedding learning. By optimize the distance from entity pairs to corresponding prototypes, our method can be easily adapted to almost arbitrary RE frameworks. Thus, the learning of infrequent or even unseen relation types will benefit from semantically proximate relations through pairs of entities and large-scale textual information. Extensive experiments on two publicly available datasets present promising improvements (4.1% F1 on average). Ablation studies on long-tail relations, main components, and different RE models demonstrate the effectiveness of the learned relation prototypes. Finally, we analyze several example cases to give intuitive impressions as qualitative analysis. Our codes and data can be found in https://github.com/CrisJk/PA-TRP.

Graph Convolutional Network (GCN) is an emerging technique that performs learning and reasoning on graph data. It operates feature learning on the graph structure, through aggregating the features of the neighbor nodes to obtain the embedding of each target node. Owing to the strong representation power, recent research shows that GCN achieves state-of-the-art performance on several tasks such as recommendation and linked document classification. Despite its effectiveness, we argue that existing designs of GCN forgo modeling cross features, making GCN less effective for tasks or data where cross features are important. Although neural network can approximate any continuous function, including the multiplication operator for modeling feature crosses, it can be rather inefficient to do so (i.e., wasting many parameters at the risk of overfitting) if there is no explicit design. To this end, we design a new operator named Cross-feature Graph Convolution, which explicitly models the arbitrary-order cross features with complexity linear to feature dimension and order size. We term our proposed architecture as Cross-GCN, and conduct experiments on three graphs to validate its effectiveness. Extensive analysis validates the utility of explicitly modeling cross features in GCN, especially for feature learning at lower layers.

Knowledge Graph (KG), as a side-information, tends to be utilized to supplement the collaborative filtering (CF) based recommendation model. By mapping items with the entities in KGs, prior studies mostly extract the knowledge information from the KGs and inject it into the representations of users and items. Despite their remarkable performance, they fail to model the user preference on attribute in the KG, since they ignore that (1) the structure information of KG may hinder the user preference learning, and (2) the user's interacted attributes will result in the bias issue on the similarity scores. With the help of causality tools, we construct the causal-effect relation between the variables in KG-based recommendation and identify the reasons causing the mentioned challenges. Accordingly, we develop a new framework, termed Knowledge Graph-based Causal Recommendation (KGCR), which implements the deconfounded user preference learning and adopts counterfactual inference to eliminate bias in the similarity scoring. Ultimately, we evaluate our proposed model on three datasets, including Amazon-book, LastFM, and Yelp2018 datasets. By conducting extensive experiments on the datasets, we demonstrate that KGCR outperforms several state-of-the-art baselines, such as KGNN-LS (Wang et al., 2019), KGAT (Wang et al., 2019) and KGIN (Wang et al., 2021).

Learning representations for multimedia content is critical for multimedia recommendation. Current representation learning methods roughly fall into two groups: (1) using the historical interactions to create ID embeddings of users and items, and (2) treating multi-modal data as the side information of items to enrich their ID embeddings. Each user-item interaction offers the supervisory signal to optimize the representation learning by the traditional supervised learning paradigm. Due to the overlook of the multi-modal patterns ($e.g.$, co-occurrence of visual, acoustic, textual features in micro-videos a user saw before, and her behavioral features) hidden in the data, these methods are insufficient to create powerful representations and obtain satisfactory recommendation accuracy. To capture multi-modal patterns in the data itself, we go beyond the supervised learning paradigm, and incorporate the idea of self-supervised learning (SSL) into multimedia recommendation. Specifically, SSL consists of two components: (1) data augmentation upon multi-modal contents, where we design three operators — feature dropout (FD), feature masking (FM), feature fine and coarse spaces (FAC) — to generate multiple views of individual items; and (2) contrastive learning, which differentiates the views of an item from the others’ to distill additional supervisory signals. Clearly, SSL enables us to explore and exhibit the underlying relations among modalities, thereby resulting in powerful representations. We denote the generic framework by Self-supervised Learning-guided Multimedia Recommendation (SLMRec). Extensive experiments are performed on three real-world datasets, showing that SLMRec achieves significant improvements over several state-of-the-art baselines like LightGCN [1], MMGCN [2]. Further analysis shows how SSL affects recommendation performance.

Conversational recommendation system (CRS) attracts increasing attention in various application domains such as retail and travel. It offers an effective way to capture users' dynamic preferences with multi-turn conversations. However, most current studies center on the recommendation aspect while over-simplifying the conversation process. The negligence of complexity in data structure and conversation flow hinders their practicality and utility. In reality, there exist various relationships among slots and values, while users' requirements may dynamically adjust or change. Moreover, the conversation often involves visual modality to facilitate the conversation. These actually call for a more advanced internal state representation of the dialogue and a proper reasoning scheme to guide the decision making process. In this paper, we explore multiple facets of multimodal conversational recommendation and try to address the above mentioned challenges. In particular, we represent the structured back-end database as a multimodal knowledge graph which captures the various relations and evidence in different modalities. The user preferences expressed via conversation utterances will then be gradually updated to the state graph with clear polarity. Based on these, we train an end-to-end State Graph-based Reasoning model SGR to perform reasoning over the whole state graph. The prediction of our proposed model benefits from the structure of the graph. It not only allows for zero-shot reasoning for items unseen in training conversations, but also provides a natural way to explain the policies. Extensive experiments show that our model achieves better performance compared with existing methods.

Recent years witnessed several advances in developing multi-goal conversational recommender systems (MG-CRS) that can proactively attract users’ interests and naturally lead user-engaged dialogues with multiple conversational goals and diverse topics. Four tasks are often involved in MG-CRS, including Goal Planning, Topic Prediction, Item Recommendation, and Response Generation. Most existing studies address only some of these tasks. To handle the whole problem of MG-CRS, modularized frameworks are adopted where each task is tackled independently without considering their interdependencies. In this work, we propose a novel Unified MultI-goal conversational recommeNDer system (UniMIND). Specifically, we unify these four tasks with different formulations into the same sequence-to-sequence paradigm. Prompt-based learning strategies are investigated to endow the unified model with the capability of multi-task learning. Finally, the overall learning and inference procedure consists of three stages, including multi-task learning, prompt-based tuning, and inference. Experimental results on two MG-CRS benchmarks (DuRecDial and TG-ReDial) show that UniMIND achieves state-of-the-art performance on all tasks with a unified model. Extensive analyses and discussions are provided for shedding some new perspectives for MG-CRS.

Aspect-based sentiment analysis (ABSA) aims at automatically inferring the specific sentiment polarities toward certain aspects of products or services behind the social media texts or reviews, which has been a fundamental application to the real-world society. Since the early 2010s, ABSA has achieved extraordinarily high accuracy with various deep neural models. However, existing ABSA models with strong in-house performances may fail to generalize to some challenging cases where the contexts are variable, i.e., low robustness to real-world environments. In this study, we propose to enhance the ABSA robustness by systematically rethinking the bottlenecks from all possible angles, including model, data, and training. First, we strengthen the current best-robust syntax-aware models by further incorporating the rich external syntactic dependencies and the labels with aspect simultaneously with a universal-syntax graph convolutional network. In the corpus perspective, we propose to automatically induce high-quality synthetic training data with various types, allowing models to learn sufficient inductive bias for better robustness. Last, we based on the rich pseudo data perform adversarial training to enhance the resistance to the context perturbation and meanwhile employ contrastive learning to reinforce the representations of instances with contrastive sentiments. Extensive robustness evaluations are conducted. The results demonstrate that our enhanced syntax-aware model achieves better robustness performances than all the state-of-the-art baselines. By additionally incorporating our synthetic corpus, the robust testing results are pushed with around 10% accuracy, which are then further improved by installing the advanced training strategies. In-depth analyses are presented for revealing the factors influencing the ABSA robustness.

As recommender systems become increasingly important in daily human decision-making, users are demanding convincing explanations to understand why they get the specific recommendation results. Although a number of explainable recommender systems have recently been proposed, there still lacks an understanding of what users really need in a recommendation explanation. The actual reason behind users’ intention to examine and consume (e.g., click and watch a movie) can be the window to answer this question and is named as self-explanation in this work. In addition, humans usually make recommendations accompanied by explanations, but there remain fewer studies on how humans explain and what we can learn from human-generated explanations. To investigate these questions, we conduct a novel multi-role, multi-session user study in which users interact with multiple types of system-generated explanations as well as human-generated explanations, namely peer-explanation . During the study, users’ intentions, expectations, and experiences are tracked in several phases, including before and after the users are presented with an explanation and after the content is examined. Through comprehensive investigations, three main findings have been made: First, we observe not only the positive but also the negative effects of explanations, and the impact varies across different types of explanations. Moreover, human-generated explanation, peer-explanation , performs better in increasing user intentions and helping users to better construct preferences, which results in better user satisfaction. Second, based on users’ self-explanation , the information accuracy is measured and found to be a major factor associated with user satisfaction. Some other factors, such as unfamiliarity and similarity, are also discovered and summarized. Third, through annotations of the information aspects used in the human-generated self-explanation and peer-explanation , patterns of how humans explain are investigated, including what information and how much information is utilized. In addition, based on the findings, a human-inspired explanation approach is proposed and found to increase user satisfaction, revealing the potential improvement of further incorporating more human patterns in recommendation explanations. These findings have shed light on the deeper understanding of the recommendation explanation and further research on its evaluation and generation. Furthermore, the collected data, including human-generated explanations by both the external peers and the users’ selves, will be released to support future research works on explanation evaluation.

Grounded Situation Recognition (GSR), i.e., recognizing the salient activity (or verb) category in an image (e.g.,buying) and detecting all corresponding semantic roles (e.g.,agent and goods), is an essential step towards “human-like” event understanding. Since each verb is associated with a specific set of semantic roles, all existing GSR methods resort to a two-stage framework: predicting the verb in the first stage and detecting the semantic roles in the second stage. However, there are obvious drawbacks in both stages: 1) The widely-used cross-entropy (XE) loss for object recognition is insufficient in verb classification due to the large intra-class variation and high inter-class similarity among daily activities. 2) All semantic roles are detected in an autoregressive manner, which fails to model the complex semantic relations between different roles. To this end, we propose a novel SituFormerfor GSR which consists of a Coarse-to-Fine Verb Model (CFVM) and a Transformer-based Noun Model (TNM). CFVM is a two-step verb prediction model: a coarse-grained model trained with XE loss first proposes a set of verb candidates, and then a fine-grained model trained with triplet loss re-ranks these candidates with enhanced verb features (not only separable but also discriminative). TNM is a transformer-based semantic role detection model, which detects all roles parallelly. Owing to the global relation modeling ability and flexibility of the transformer decoder, TNM can fully explore the statistical dependency of the roles. Extensive validations on the challenging SWiG benchmark show that SituFormer achieves a new state-of-the-art performance with significant gains under various metrics. Code is available at https://github.com/kellyiss/SituFormer.

Video question answering requires the models to understand and reason about both the complex video and language data to correctly derive the answers. Existing efforts have been focused on designing sophisticated cross-modal interactions to fuse the information from two modalities, while encoding the video and question holistically as frame and word sequences. Despite their success, these methods are essentially revolving around the sequential nature of video- and question-contents, providing little insight to the problem of question-answering and lacking interpretability as well. In this work, we argue that while video is presented in frame sequence, the visual elements (e.g., objects, actions, activities and events) are not sequential but rather hierarchical in semantic space. To align with the multi-granular essence of linguistic concepts in language queries, we propose to model video as a conditional graph hierarchy which weaves together visual facts of different granularity in a level-wise manner, with the guidance of corresponding textual cues. Despite the simplicity, our extensive experiments demonstrate the superiority of such conditional hierarchical graph architecture, with clear performance improvements over prior methods and also better generalization across different type of questions. Further analyses also demonstrate the model's reliability as it shows meaningful visual-textual evidences for the predicted answers.

Neural discrete reasoning (NDR) has shown remarkable progress in combining deep models with discrete reasoning.However, we find that existing NDR solution suffers from large performance drop on hypothetical questions, e.g., "what the annualized rate of return would be if the revenue in 2020 was doubled".The key to hypothetical question answering (HQA) is counterfactual thinking, which is a natural ability of human reasoning but difficult for deep models.In this work, we devise a Learning to Imagine (L2I) module, which can be seamlessly incorporated into NDR models to perform the imagination of unseen counterfactual.In particular, we formulate counterfactual thinking into two steps: 1) identifying the fact to intervene, and 2) deriving the counterfactual from the fact and assumption, which are designed as neural networks.Based on TAT-QA, we construct a very challenging HQA dataset with 8,283 hypothetical questions.We apply the proposed L2I to TAGOP, the state-of-theart solution on TAT-QA, validating the rationality and effectiveness of our approach.

The progress of recommender systems is hampered mainly by evaluation as it requires real-time interactions between humans and systems, which is too laborious and expensive. This issue is usually approached by utilizing the interaction history to conduct offline evaluation. However, existing datasets of user-item interactions are partially observed, leaving it unclear how and to what extent the missing interactions will influence the evaluation. To answer this question, we collect a fully-observed dataset from Kuaishou's online environment, where almost all 1,411 users have been exposed to all 3,327 items. To the best of our knowledge, this is the first real-world fully-observed data with millions of user-item interactions.

Video Question Answering (VideoQA) is the task of an-swering questions about a video. At its core is understanding the alignments between visual scenes in video and linguistic semantics in question to yield the answer. In leading VideoQA models, the typical learning objective, empirical risk minimization (ERM), latches on superficial correlations between video-question pairs and answers as the alignments. However, ERM can be problematic, because it tends to over-exploit the spurious correlations between question-irrelevant scenes and answers, instead of inspecting the causal effect of question-critical scenes. As a result, the VideoQA models suffer from unreliable reasoning. In this work, we first take a causal look at VideoQA and argue that invariant grounding is the key to ruling out the spurious correlations. Towards this end, we propose a new learning framework, Invariant Grounding for VideoQA (IGV), to ground the question-critical scene, whose causal relations with answers are invariant across different interventions on the complement. With IGV, the VideoQA mod-els are forced to shield the answering process from the negative influence of spurious correlations, which significantly improves the reasoning ability. Experiments on three benchmark datasets validate the superiority of IGV in terms of accuracy, visual explainability, and generalization ability over the leading baselines. Our code is available at https://github.com/y13800/IGV.

To facilitate conversational question answering (CQA) over hybrid contexts in finance, we present a new dataset, named PACIFIC.Compared with existing CQA datasets, PACIFIC exhibits three key features: (i) proactivity, (ii) numerical reasoning, and (iii) hybrid context of tables and text.A new task is defined accordingly to study Proactive Conversational Question Answering (PCQA), which combines clarification question generation and CQA.In addition, we propose a novel method, namely UniPCQA, to adapt a hybrid format of input and output content in PCQA into the Seq2Seq problem, including the reformulation of the numerical reasoning process as code generation.UniPCQA performs multi-task learning over all sub-tasks in PCQA and incorporates a simple ensemble strategy to alleviate the error propagation issue in the multi-task learning by cross-validating top-k sampled Seq2Seq outputs.We benchmark the PACIFIC dataset with extensive baselines and provide comprehensive evaluations on each sub-task of PCQA.The following tables present the recorded investment by portfolio segment and by region.($ in millions)

We study automatic Contract Clause Extraction (CCE) by modeling implicit relations in legal contracts.Existing CCE methods mostly treat contracts as plain text, creating a substantial barrier to understanding contracts of high complexity.In this work, we first comprehensively analyze the complexity issues of contracts and distill out three implicit relations commonly found in contracts, namely, 1) Long-range Context Relation that captures the correlations of distant clauses; 2) Term-Definition Relation that captures the relation between important terms with their corresponding definitions; and 3) Similar Clause Relation that captures the similarities between clauses of the same type.Then we propose a novel framework ConReader to exploit the above three relations for better contract understanding and improving CCE.Experimental results show that ConReader makes the prediction more interpretable and achieves new state-of-the-art on two CCE tasks in both conventional and zero-shot settings.1

Vision-language pre-training (VLP) has shown impressive performance on a wide range of cross-modal tasks, where VLP models without reliance on object detectors are becoming the mainstream due to their superior computation efficiency and competitive performance.However, the removal of object detectors also deprives the capability of VLP models in explicit object modeling, which is essential to various position-sensitive vision-language (VL) tasks, such as referring expression comprehension and visual commonsense reasoning.To address the challenge, we introduce PEVL that enhances the pre-training and prompt tuning of VLP models with explicit object position modeling.Specifically, PEVL reformulates discretized object positions and language in a unified language modeling framework, which facilitates explicit VL alignment during pretraining, and also enables flexible prompt tuning for various downstream tasks.We show that PEVL enables state-of-the-art performance of detector-free VLP models on position-sensitive tasks such as referring expression comprehension and phrase grounding, and also improves the performance on position-insensitive tasks with grounded inputs.We make the data and code for this paper publicly available at https://github.com/thunlp/PEVL.

This survey aims to organize the recent advances in video question answering (VideoQA) and point towards future directions.We firstly categorize the datasets into: 1) normal VideoQA, multi-modal VideoQA and knowledge-based VideoQA, according to the modalities invoked in the question-answer pairs, and 2) factoid VideoQA and inference VideoQA, according to the technical challenges in comprehending the questions and deriving the correct answers.We then summarize the VideoQA techniques, including those mainly designed for Factoid QA (such as the early spatio-temporal attention-based methods and the recent Transformer-based ones) and those targeted at explicit relation and logic inference (such as neural modular networks, neural symbolic methods, and graph-structured methods).Aside from the backbone techniques, we also delve into specific models and derive some common and useful insights either for video modeling, question answering, or for cross-modal correspondence learning.Finally, we present the research trends of studying beyond factoid VideoQA to inference VideoQA, as well as towards the robustness and interpretability.Additionally, we maintain a repository, https://github.com/VRU-NExT/ VideoQA, to keep trace of the latest VideoQA papers, datasets, and their open-source implementations if available.With these efforts, we strongly hope this survey could shed light on the follow-up VideoQA research.

Leading graph contrastive learning (GCL) methods perform graph augmentations in two fashions: (1) randomly corrupting the anchor graph, which could cause the loss of semantic information, or (2) using domain knowledge to maintain salient features, which undermines the generalization to other domains. Taking an invariance look at GCL, we argue that a high-performing augmentation should preserve the salient semantics of anchor graphs regarding instance-discrimination. To this end, we relate GCL with invariant rationale discovery, and propose a new framework, Rationale-aware Graph Contrastive Learning (RGCL). Specifically, without supervision signals, RGCL uses a rationale generator to reveal salient features about graph instance-discrimination as the rationale, and then creates rationale-aware views for contrastive learning. This rationale-aware pre-training scheme endows the backbone model with the powerful representation ability, further facilitating the fine-tuning on downstream tasks. On MNIST-Superpixel and MUTAG datasets, visual inspections on the discovered rationales showcase that the rationale generator successfully captures the salient features (\ie distinguishing semantic nodes in graphs). On biochemical molecule and social network benchmark datasets, the state-of-the-art performance of RGCL demonstrates the effectiveness of rationale-aware views for contrastive learning. Our codes are available at https://github.com/lsh0520/RGCL.

Bundle recommendation aims to recommend a bundle of related items to users, which can satisfy the users' various needs with one-stop convenience. Recent methods usually take advantage of both user-bundle and user-item interactions information to obtain informative representations for users and bundles, corresponding to bundle view and item view, respectively. However, they either use a unified view without differentiation or loosely combine the predictions of two separate views, while the crucial cooperative association between the two views' representations is overlooked.

Waterfall Recommender System (RS), a popular form of RS in mobile applications, is a stream of recommended items consisting of successive pages that can be browsed by scrolling. In waterfall RS, when a user finishes browsing a page, the edge (e.g., mobile phones) would send a request to the cloud server to get a new page of recommendations, known as the paging request mechanism. RSs typically put a large number of items into one page to reduce excessive resource consumption from numerous paging requests, which, however, would diminish the RSs' ability to timely renew the recommendations according to users' real-time interest and lead to a poor user experience. Intuitively, inserting additional requests inside pages to update the recommendations with a higher frequency can alleviate the problem. However, previous attempts, including only non-adaptive strategies (e.g., insert requests uniformly), would eventually lead to resource overconsumption. To this end, we envision a new learning task of edge intelligence named Intelligent Request Strategy Design (IRSD). It aims to improve the effectiveness of waterfall RSs by determining the appropriate occasions of request insertion based on users' real-time intention. Moreover, we propose a new paradigm of adaptive request insertion strategy named Uplift-based On-edge Smart Request Framework (AdaRequest). AdaRequest 1) captures the dynamic change of users' intentions by matching their real-time behaviors with their historical interests based on attention-based neural networks. 2) estimates the counterfactual uplift of user purchase brought by an inserted request based on causal inference. 3) determines the final request insertion strategy by maximizing the utility function under online resource constraints. We conduct extensive experiments on both offline dataset and online A/B test to verify the effectiveness of AdaRequest. Remarkably, AdaRequest has been deployed on the Waterfall RS of Taobao and brought over 3% lift on Gross Merchandise Value (GMV).

In graph classification, attention- and pooling-based graph neural networks (GNNs) prevail to extract the critical features from the input graph and support the prediction. They mostly follow the paradigm of learning to attend, which maximizes the mutual information between the attended graph and the ground-truth label. However, this paradigm makes GNN classifiers recklessly absorb all the statistical correlations between input features and labels in the training data, without distinguishing the causal and noncausal effects of features. Instead of underscoring the causal features, the attended graphs are prone to visit the noncausal features as the shortcut to predictions. Such shortcut features might easily change outside the training distribution, thereby making the GNN classifiers suffer from poor generalization.

Video Question Answering (VideoQA) is the task of answering the natural language questions about a video. Producing an answer requires understanding the interplay across visual scenes in video and linguistic semantics in question. However, most leading VideoQA models work as black boxes, which make the visual-linguistic alignment behind the answering process obscure. Such black-box nature calls for visual explainability that reveals "What part of the video should the model look at to answer the question?". Only a few works present the visual explanations in a post-hoc fashion, which emulates the target model's answering process via an additional method.

With the proliferation of online shopping, fashion recommendation, which aims to provide suitable suggestions to support the consumer's purchase in e-commerce platforms, has gained increasing research attention from both academia and industry. Although existing efforts have achieved great progress, they focus on the visual modality, lacking the exploration of other modalities of items, e.g., the textual descriptions and attributes of items. Accordingly, this workshop targets calling for a coordinated effort to promote the multimedia computing towards fashion recommendation. This workshop will showcase the innovative methodologies and ideas on new yet challenging research problems, including (not limited to) fashion recommendation, interactive fashion recommendation, interactive garment retrieval, and outfit compatibility modeling.

Multimodal dialogue systems attract much attention recently, but they are far from skills like: 1) automatically generate context- specific responses instead of safe but general responses; 2) naturally coordinate between the different information modalities (e.g. text and image) in responses; 3) intuitively explain the reasons for generated responses and improve a specific response without re-training the whole model. To approach these goals, we propose a different angle for the task - Reflecting Experiences for Response Generation (RERG). This is supported by the fact that generating a response from scratch can be hard, but much easier if we can access other similar dialogue contexts and the corresponding responses. In particular, RERG first uses a multimodal contrastive learning enhanced retrieval model for soliciting similar dialogue instances. It then employs a cross copy based reuse model to explore the current dialogue context (vertical) and similar dialogue instances' responses (horizontal) for response generation simultaneously. Experimental results demonstrate that our model outperforms other state-of-the-art models on both automatic metrics and human evaluation. Moreover, RERG naturally provides supporting dialogue instances for better explainability. It also has a strong capability in adapting to unseen dialogue settings by simply adding related samples to the retrieval datastore without re-training the whole model.

Document Visual Question Answering (VQA) aims to answer questions over visually-rich documents. In this work, we introduce a new Document VQA dataset, named TAT-DQA, which consists of 3,067 document pages comprising semi-structured table(s) and unstructured text as well as 16,558 question-answer pairs. The documents are sampled from financial reports and contain lots of numbers, which means discrete reasoning capability is demanded to answer the questions. Based on TAT-DQA, we further develop a novel model named MHST that takes into account the information in multi-modalities to intelligently address different types of questions with corresponding strategies, i.e., extraction or reasoning. The experiments show that MHST model significantly outperforms the baseline methods, demonstrating its effectiveness. However, the performance still lags far behind that of expert humans. We expect that our TAT-DQA dataset would facilitate the research on understanding of visually-rich documents, especially for scenarios that require discrete reasoning. Also, we hope the proposed model would inspire researchers to design more advanced Document VQA models in future.

Universally modeling all typical information extraction tasks (UIE) with one generative language model (GLM) has revealed great potential by the latest study, where various IE predictions are unified into a linearized hierarchical expression under a GLM. Syntactic structure information, a type of effective feature which has been extensively utilized in IE community, should also be beneficial to UIE. In this work, we propose a novel structure-aware GLM, fully unleashing the power of syntactic knowledge for UIE. A heterogeneous structure inductor is explored to unsupervisedly induce rich heterogeneous structural representations by post-training an existing GLM. In particular, a structural broadcaster is devised to compact various latent trees into explicit high-order forests, helping to guide a better generation during decoding. We finally introduce a task-oriented structure fine-tuning mechanism, further adjusting the learned structures to most coincide with the end-task's need. Over 12 IE benchmarks across 7 tasks our system shows significant improvements over the baseline UIE system. Further in-depth analyses show that our GLM learns rich task-adaptive structural bias that greatly resolves the UIE crux, the long-range dependence issue and boundary identifying.

Collaborative filtering (CF) models easily suffer from popularity bias, which makes recommendation deviate from users’ actual preferences. However, most current debiasing strategies are prone to playing a trade-off game between head and tail performance, thus inevitably degrading the overall recommendation accuracy. To reduce the negative impact of popularity bias on CF models, we incorporate Bias-aware margins into Contrastive loss and propose a simple yet effective BC Loss, where the margin tailors quantitatively to the bias degree of each user-item interaction. We investigate the geometric interpretation of BC loss, then further visualize and theoretically prove that it simultaneously learns better head and tail representations by encouraging the compactness of similar users/items and enlarging the dispersion of dissimilar users/items. Over six benchmark datasets, we use BC loss to optimize two high-performing CF models. In various evaluation settings (i.e., imbalanced/balanced, temporal split, fully-observed unbiased, tail/head test evaluations), BC loss outperforms the state-of-the-art debiasing and non-debiasing methods with remarkable improvements. Considering the theoretical guarantee and empirical success of BC loss, we advocate using it not just as a debiasing strategy, but also as a standard loss in recommender models. Codes are available at https://github.com/anzhang314/BC-Loss.

Proactive dialogue system is able to lead the conversation to a goal topic and has advantaged potential in bargain, persuasion, and negotiation. Current corpus-based learning manner limits its practical application in real-world scenarios. To this end, we contribute to advancing the study of the proactive dialogue policy to a more natural and challenging setting, i.e., interacting dynamically with users. Further, we call attention to the non-cooperative user behavior - the user talks about off-path topics when he/she is not satisfied with the previous topics introduced by the agent. We argue that the targets of reaching the goal topic quickly and maintaining a high user satisfaction are not always converged, because the topics close to the goal and the topics user preferred may not be the same. Towards this issue, we propose a new solution named I-Pro that can learn Proactive policy in the Interactive setting. Specifically, we learn the trade-off via a learned goal weight, which consists of four factors (dialogue turn, goal completion difficulty, user satisfaction estimation, and cooperative degree). The experimental results demonstrate I-Pro significantly outperforms baselines in terms of effectiveness and interpretability.

Recommender systems usually face the issue of filter bubbles: over-recommending homogeneous items based on user features and historical interactions. Filter bubbles will grow along the feedback loop and inadvertently narrow user interests. Existing work usually mitigates filter bubbles by incorporating objectives apart from accuracy such as diversity and fairness. However, they typically sacrifice accuracy, hurting model fidelity and user experience. Worse still, users have to passively accept the recommendation strategy and influence the system in an inefficient manner with high latency, e.g., keeping providing feedback (e.g., like and dislike) until the system recognizes the user intention.

Generating fluent and informative natural responses while main- taining representative internal states for search optimization is critical for conversational search systems. Existing approaches ei- ther 1) predict structured dialog acts first and then generate natural response; or 2) map conversation context to natural responses di- rectly in an end-to-end manner. Both kinds of approaches have shortcomings. The former suffers from error accumulation while the semantic associations between structured acts and natural re- sponses are confined in single direction. The latter emphasizes generating natural responses but fails to predict structured acts. Therefore, we propose a neural co-generation model that gener- ates the two concurrently. The key lies in a shared latent space shaped by two informed priors. Specifically, we design structured dialog acts and natural response auto-encoding as two auxiliary tasks in an interconnected network architecture. It allows for the concurrent generation and bidirectional semantic associations. The shared latent space also enables asynchronous reinforcement learn- ing for further joint optimization. Experiments show that our model achieves significant performance improvements.

System auditing provides a low-level view into cyber threats by monitoring system entity interactions. In response to advanced cyber-attacks, one prevalent solution is to apply data provenance analysis on audit records to search for anomalies (anomalous behaviors) or specifications of known attacks. However, existing approaches suffer from several limitations: 1) generating high volumes of false alarms, 2) relying on expert knowledge, or 3) producing coarse-grained detection signals. In this paper, we recognize the structural similarity between threat detection in cybersecurity and recommendation in information retrieval. By mapping security concepts of system entity interactions to recommendation concepts of user-item interactions, we identify cyber threats by predicting the preferences of a system entity on its interactive entities. Furthermore, inspired by the recent advances in modeling high-order connectivity via item side information in the recommendation, we transfer the insight to cyber threat analysis and customize an automated detection system, SHADEWATCHER. It fulfills the potential of high-order information in audit records via graph neural networks to improve detection effectiveness. Besides, we equip SHADEWATCHER with dynamic updates towards better generalization to false alarms. In our evaluation against both real-life and simulated cyber-attack scenarios, SHADEWATCHER shows its advantage in identifying threats with high precision and recall rates. Moreover, SHADEWATCHER is capable of pinpointing threats from nearly a million system entity interactions within seconds.

Modern recommender systems learn user representations from historical interactions, which suffer from the problem of user feature shifts, such as an income increase. Historical interactions will inject out-of-date information into the representation in conflict with the latest user feature, leading to improper recommendations. In this work, we consider the Out-Of-Distribution (OOD) recommendation problem in an OOD environment with user feature shifts. To pursue high fidelity, we set additional objectives for representation learning as: 1) strong OOD generalization and 2) fast OOD adaptation.

Effectively representing users lie at the core of modern recommender systems. Since users’ interests naturally exhibit multiple aspects, it is of increasing interest to develop multi-interest frameworks for recommendation, rather than represent each user with an overall embedding. Despite their effectiveness, existing methods solely exploit the encoder (the forward flow) to represent multiple aspects of interests. However, without explicit regularization, the interest embeddings may not be distinct from each other nor semantically reflect representative historical items. Towards this end, we propose the Re4 framework, which leverages the backward flow to reexamine each interest embedding. Specifically, Re4 encapsulates three backward flows, i.e., 1) Re-contrast, which drives each interest embedding to be distinct from other interests using contrastive learning; 2) Re-attend, which ensures the interest-item correlation estimation in the forward flow to be consistent with the criterion used in final recommendation; and 3) Re-construct, which ensures that each interest embedding can semantically reflect the information of representative items that relate to the corresponding interest. We demonstrate the novel forward-backward multi-interest paradigm on ComiRec, and perform extensive experiments on three real-world datasets. Empirical studies validate that Re4 helps to learn learning distinct and effective multi-interest representations.

Meta-learning has been proposed as a framework to address the challenging few-shot learning setting. The key idea is to leverage a large number of similar few-shot tasks in order to learn how to adapt a base-learner to a new task for which only a few labeled samples are available. As deep neural networks (DNNs) tend to overfit using a few samples only, typical meta-learning models use shallow neural networks, thus limiting its effectiveness. In order to achieve top performance, some recent works tried to use the DNNs pre-trained on large-scale datasets but mostly in straight-forward manners, e.g., (1) taking their weights as a warm start of meta-training, and (2) freezing their convolutional layers as the feature extractor of base-learners. In this paper, we propose a novel approach called meta-transfer learning (MTL), which learns to transfer the weights of a deep NN for few-shot learning tasks. Specifically, meta refers to training multiple tasks, and transfer is achieved by learning scaling and shifting functions of DNN weights (and biases) for each task. To further boost the learning efficiency of MTL, we introduce the hard task (HT) meta-batch scheme as an effective learning curriculum of few-shot classification tasks. We conduct experiments for five-class few-shot classification tasks on three challenging benchmarks, miniImageNet, tieredImageNet, and Fewshot-CIFAR100 (FC100), in both supervised and semi-supervised settings. Extensive comparisons to related works validate that our MTL approach trained with the proposed HT meta-batch scheme achieves top performance. An ablation study also shows that both components contribute to fast convergence and high accuracy.

Images can convey rich semantics and induce various emotions in viewers. Recently, with the rapid advancement of emotional intelligence and the explosive growth of visual data, extensive research efforts have been dedicated to affective image content analysis (AICA). In this survey, we will comprehensively review the development of AICA in the recent two decades, especially focusing on the state-of-the-art methods with respect to three main challenges - the affective gap, perception subjectivity, and label noise and absence. We begin with an introduction to the key emotion representation models that have been widely employed in AICA and description of available datasets for performing evaluation with quantitative comparison of label noise and dataset bias. We then summarize and compare the representative approaches on (1) emotion feature extraction, including both handcrafted and deep features, (2) learning methods on dominant emotion recognition, personalized emotion prediction, emotion distribution learning, and learning from noisy data or few labels, and (3) AICA based applications. Finally, we discuss some challenges and promising research directions in the future, such as image content and context understanding, group emotion clustering, and viewer-image interaction.

To explore the specific visual aspects and the language consistency at the same time, this paper introduces a new image captioning task, dubbedentity slot filling captioning (ESFCap). It is similar to the masked entity completion tasks in NLP, which are widely used to study language context and has been successfully employed to improve language understanding. Specifically, given a sentence with blank for describing an image, the ESFCap task aims to fill the blank with proper text content according to the visual information. The filled text should be grounded to correct visual entities and also in concordance with the sentence structure. To support the ESFCap research, we collect and release an entity slot filling captioning dataset,Flickr30k-EnFi, based on Flickr30k-Entities. The Flickr30k-EnFi dataset consists of 31,783 images and 565,750 masked sentences, as well as the text snippets for the masked slot. For tackling the ESFCap task, we propose a multi-modal fusion model equipped with a novel adaptive dynamic attention module, termed AdaMFN. The AdaMFN model effectively leverages both global and local information from vision and language. It is also able to adaptively focus on the key linguistic knowledge and visual regions to generate correct filling results. The experimental results and analysis demonstrate the effectiveness of our proposed model.

The task of video moment retrieval (VMR) is to retrieve the specific video moment from an untrimmed video, according to a textual query. It is a challenging task that requires effective modeling of complex cross-modal matching relationship. Recent efforts primarily model the cross-modal interactions by hand-crafted network architectures. Despite their effectiveness, they rely heavily on expert experience to select architectures and have numerous hyperparameters that need to be carefully tuned, which significantly limit their applications in real-world scenarios. How to design flexible architectures for modeling cross-modal interactions with less manual effort is crucial for the task of VMR but has received limited attention so far. To address this issue, we present a novel VMR approach that automatically searches for an optimal architecture to learn cross-modal matching relationship. Specifically, we develop a cross-modal architecture searching method. It first searches for repeatable cell network architectures based on a directed acyclic graph, which performs operation sampling over a customized task-specific operation set. Then, we adaptively modulate the edge importance in the graph by a query-aware attention network, which performs edge sampling softly in the searched cell. Different from existing neural architecture search methods, our approach can effectively exploit the query information to reach query-conditioned architectures for modeling cross modal matching. Extensive experiments on three benchmark datasets show that our approach can not only significantly outperform the state-of-the-art methods but also run more efficiently and robustly than manually crafted network architectures.

Although the single-image super-resolution (SISR) methods have achieved great success on the single degradation, they still suffer performance drop with multiple degrading effects in real scenarios. Recently, some blind and non-blind models for multiple degradations have been explored. However, these methods usually degrade significantly for distribution shifts between the training and test data. Towards this end, we propose a novel conditional hyper-network framework for super-resolution with multiple degradations (named CMDSR), which helps the SR framework learn how to adapt to changes in the degradation distribution of input. We extract degradation prior at the task-level with the proposed ConditionNet, which will be used to adapt the parameters of the basic SR network (BaseNet). Specifically, the ConditionNet of our framework first learns the degradation prior from a support set, which is composed of a series of degraded image patches from the same task. Then the adaptive BaseNet rapidly shifts its parameters according to the conditional features. Moreover, in order to better extract degradation prior, we propose a task contrastive loss to shorten the inner-task distance and enlarge the cross-task distance between task-level features. Without predefining degradation maps, our blind framework can conduct one single parameter update to yield considerable improvement in SR results. Extensive experiments demonstrate the effectiveness of CMDSR over various blind, and even several non-blind methods. The flexible BaseNet structure also reveals that CMDSR can be a general framework for a large series of SISR models. Our code is available at https://github.com/guanghaoyin/CMDSR.

Conversational systems have recently attracted significant attention. Both the research community and industry believe that it will exert huge impact on human-computer interaction, and specifically, the IR/RecSys community has begun to explore Conversational Recommendation. In real-life scenarios, such systems are often urgently needed in helping users accomplishing different tasks under various situations. However, existing works still face several shortcomings: (1) Most efforts are largely confined in single task setting. They fall short of hands in handling tasks across domains. (2) Aside from soliciting user preference from dialogue history, a conversational recommender naturally has access to the back-end data structure which should be fully leveraged to yield good recommendations. In this paper, we thus present a Topic-guided Conversational Recommender (TCR) which is specifically designed for the multi-domain setting. It augments the sequence-to-sequence (seq2seq) models with a neural latent topic component to better guide the response generation. To better leverage the dialogue history and the back-end data structure, we adopt a graph convolutional network (GCN) to model the relationships between different recommendation candidates while also capture the match between candidates and the dialogue history. We then seamlessly combine these two parts with the idea of pointer networks. We perform extensive evaluation on a large-scale task-oriented multi-domain dialogue dataset and the results show that our method achieves superior performance as compared to a wide range of baselines.

As sharing of accounts is quite common among family members or roommates, the design of efficient recommender schemes for shared accounts has raised much attention recently. Generally speaking, after each login, it is essential for a recommender system to identify the current user behind and leverage this information to make recommendations. One naive approach is first to identify the identity of the current user and then make recommendations. However, this two-stage based approach may not achieve satisfactory performance. The key is that the recommended items favoring identifying users in the first stage may not be interesting to the users, which can deplete the user's patience quickly and cause early termination of users. To address the problem, we propose a novel recommendation scheme that makes a tradeoff between recommending discriminating items (helpful for identifying the user) and recommending interesting ones to the user (helpful for increasing the number of clicks). Under this scheme, we develop a patience model to capture the user's dynamic patience level during the recommendation process. Moreover, considering the increasing popularity of mobile devices, we also incorporate mobile sensor data (i.e., angle, accelerometer, gyroscope, etc.) into our approach to further improve the performance of the system. We implemented the above system in an App on mobile devices and carried out extensive experiments. The results demonstrate that our proposed scheme significantly outperforms the existing state-of-the-art approaches.

Mixed dish is a food category that contains different dishes mixed in one plate, and is popular in Eastern and Southeast Asia. Recognizing the individual dishes in a mixed dish image is important for health related applications, e.g. to calculate the nutrition values of the dish. However, most existing methods that focus on single dish classification are not applicable to the recognition of mixed dish images. The main challenge of mixed dish recognition comes from three aspects: a wide range of dish types, the complex dish combination with severe overlap between different dishes and the large visual variances of same dish type caused by different cooking/cutting methods applied in different canteens. In order to tackle these problems, we propose the contextual relation network that encodes the implicit and explicit contextual relations among multiple dishes from region-level features and label-level co-occurrence respectively. Besides, to address the visual variances of dish instances from different canteens, we introduce the domain adaption networks to align both local and global features, and eliminating domain gaps of dish features across different canteens. In addition, we collect a mixed dish image dataset containing 9254 mixed dish images from 6 canteens in Singapore. Extensive experiments on both our dataset and public one validate that our methods can achieve top performance for localizing and recognizing multiple dishes and solve the domain shift problem to a certain extent in mixed dish images.

Fashion trend forecasting is of great research significance in providing useful suggestions for both fashion companies and fashion lovers. Although various studies have been devoted to tackling this challenging task, they only studied limited fashion elements with highly seasonal or simple patterns, which could hardly reveal the real complex fashion trends. Moreover, the mainstream solutions for this task are still statistical-based and solely focus on time-series data modeling, which limit the forecast accuracy. Towards insightful fashion trend forecasting, previous work[1]proposed to analyze more fine-grained fashion elements which can informatively reveal fashion trends. Specifically, it focused on detailed fashion element trend forecasting for specific user groups based on social media data. In addition, it proposed a neural network-based method, namely KERN, to address the problem of fashion trend modeling and forecasting. In this work, to extend the previous work[1], we propose an improved model named Relation Enhanced Attention Recurrent (REAR) network. Compared to KERN, the REAR model leverages not only the relations among fashion elements, but also those among user groups, thus capturing more types of correlations among various fashion trends. To further improve the performance of long-range trend forecasting, the REAR method devises a sliding temporal attention mechanism, which is able to capture temporal patterns on future horizons better. Extensive experiments and more analysis have been conducted on the FIT[1]and GeoStyle[2]datasets to evaluate the performance of REAR. Experimental and analytical results demonstrate the effectiveness of the proposed REAR model in fashion trend forecasting, which also show the improvement of REAR compared to the KERN.

Fashion recommendation, aiming to explore specific user preference in fashion, has become an important research topic for its practical significance to the fashion business sector. However, little work has been done on an important sub-task called sequential fashion recommendation, which aims to capture additional short-term fashion interest of users by modeling the item-to-item transitions. In this paper, we propose a novel Attentional Content-level Translation-based Recommender (ACTR) framework, which simultaneously models the instant user intent of each transition and the intent-specific transition probability. Specifically, we define instant intent with the relationships between adjacent items that the users interacted, which are the three fundamental domain-specific relationships of:match,substituteandothers. To further exploit the characteristics of fashion domain and alleviate the item transition sparsity problem, we augment the item-level transition modeling with multiple sub-transitions using various content-level attributes. An attention mechanism is further devised to effectively aggregate multiple content-level transitions. To the best of our knowledge, this is the first work that specifies the implicit user actions in online fashion shopping with explicit instant intent, which enhances the connectivity of fashion items and boosts the recommendation performance. Extensive experiments on two real-world fashion E-commerce datasets demonstrate the effectiveness of the proposed method in sequential fashion recommendation.

Understanding user preference on item context is the key to acquire a high-quality multimedia recommendation. Typically, the pre-existing features of items are derived from pre-trained models (e.g. visual features of micro-videos extracted from some neural networks), and then introduced into the recommendation framework (e.g. collaborative filtering) to capture user preference. However, we argue that such a paradigm is insufficient to output satisfactory user representations, which hardly profile personal interests well. The key reason is that present works largely leave user intents untouched, then failing to encode such informative representation of users. In this work, we aim to learn multi-level user intents from the co-interacted patterns of items, so as to obtain high-quality representations of users and items and further enhance the recommendation performance. Towards this end, we develop a novel framework,Hierarchical User Intent Graph Network, which exhibits user intents in a hierarchical graph structure, from the fine-grained to coarse-grained intents. In particular, we get the multi-level user intents by recursively performing two operations: 1) intra-level aggregation, which distills the signal pertinent to user intents from co-interacted item graphs; and 2) inter-level aggregation, which constitutes the supernode in higher levels to model coarser-grained user intents via gathering the nodes’ representations in the lower ones. Then, we refine the user and item representations as a distribution over the discovered intents, instead of simple pre-existing features. To demonstrate the effectiveness of our model, we conducted extensive experiments on three public datasets. Our model achieves significant improvements over the state-of-the-art methods, including MMGCN and DisenGCN. Furthermore, by visualizing the item representations, we provide the semantics of user intents.

introduction Share on Graph Technologies for User Modeling and Recommendation: Introduction to the Special Issue - Part 1 Authors: Xiangnan He University of Science and Technology of China, He Fei, China University of Science and Technology of China, He Fei, ChinaSearch about this author , Zhaochun Ren Shandong University, Qingdao, China Shandong University, Qingdao, ChinaSearch about this author , Emine Yilmaz University College London, United Kingdom University College London, United KingdomSearch about this author , Marc Najork Google Research, Mountain View, CA, United States Google Research, Mountain View, CA, United StatesSearch about this author , Tat-Seng Chua National University of Singapore, Republic of Singapore, Singapore National University of Singapore, Republic of Singapore, SingaporeSearch about this author Authors Info & Claims ACM Transactions on Information SystemsVolume 40Issue 2April 2022 Article No.: 21pp 1–5https://doi.org/10.1145/3477596Online:27 September 2021Publication History 0citation282DownloadsMetricsTotal Citations0Total Downloads282Last 12 Months282Last 6 weeks7 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my Alerts New Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

introduction Share on Introduction to the Special Section on Graph Technologies for User Modeling and Recommendation, Part 2 Authors: Xiangnan He University of Science and Technology of China, Hefei, China University of Science and Technology of China, Hefei, ChinaSearch about this author , Zhaochun Ren Shandong University, Qingdao, China Shandong University, Qingdao, ChinaSearch about this author , Emine Yilmaz Department of Computer Science, University College London, London, United Kingdom Department of Computer Science, University College London, London, United KingdomSearch about this author , Marc Najork Google Research, Mountain View, CA, United States Google Research, Mountain View, CA, United StatesSearch about this author , Tat-Seng Chua National University of Singapore, Singapore, Republic of Singapore National University of Singapore, Singapore, Republic of SingaporeSearch about this author Authors Info & Claims ACM Transactions on Information SystemsVolume 40Issue 3July 2022 Article No.: 42pp 1–5https://doi.org/10.1145/3490180Published:14 December 2021Publication History 0citation385DownloadsMetricsTotal Citations0Total Downloads385Last 12 Months385Last 6 weeks27 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my Alerts New Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

Lexical cohesion is a fundamental mechanism for text which requires a pair of words to be interpreted as a certain type of lexical relation (e.g., similarity) to understand a coherent context; we refer to such relations as the contextual lexical relation. However, work on lexical cohesion has not modeled context comprehensively in considering lexical relations due to the lack of linguistic resources. In this paper, we take initial steps to address contextual lexical relations by focusing on the contrast relation, as it is a well-known relation though it is more subtle and relatively less resourced. We present a corpus named Cont 2 Lex to make Contextual Lexical Contrast Recognition a computationally feasible task. We benchmark this task with widely-adopted semantic representations; we discover that contextual embeddings (e.g. BERT) generally outperform static embeddings (e.g. Glove), but barely go beyond 70% in accuracy performance. In addition, we find that all embeddings perform better when CLC occurs within the same sentence, suggesting possible limitations of current computational coherence models. Another intriguing discovery is the improvement of BERT in CLC is largely attributed to its modeling of CLC word pairs co-occurring with other word repetitions. Such observations imply that the progress made in lexical coherence modeling remains relatively primitive even for semantic representations such as BERT that have been empowering numerous standard NLP tasks to approach human benchmarks. Through presenting our corpus and benchmark, we attempt to seed initial discussions and endeavors in advancing semantic representations from modeling syntactic and semantic levels to coherence and discourse levels.

Word embedding can represent a word as a point vector or a Gaussian distribution in high-dimensional spaces. Gaussian distribution is innately more expressive than point vector owing to the ability to additionally capture semantic uncertainties of words, and thus can express asymmetric relations among words more naturally (e.g., animal entails cat but not the reverse. However, previous Gaussian embedders neglect inner-word conceptual knowledge and lack tailored Gaussian contextualizer, leading to inferior performance on both intrinsic (context-agnostic) and extrinsic (context-sensitive) tasks. In this paper, we first propose a novel Gaussian embedder which explicitly accounts for inner-word conceptual units (sememes) to represent word semantics more precisely; during learning, we propose Gaussian Distribution Attention over Gaussian representations to adaptively aggregate multiple sememe distributions into a word distribution, which guarantees the Gaussian linear combination property. Additionally, we propose a Gaussian contextualizer to utilize outer-word contexts in a sentence, producing contextualized Gaussian representations for context-sensitive tasks. Extensive experiments on intrinsic and extrinsic tasks demonstrate the effectiveness of the proposed approach, achieving state-of-the-art performance with near 5.00% relative improvement.

Zikun Hu, Yixin Cao, Lifu Huang, Tat-Seng Chua. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Fengbin Zhu, Wenqiang Lei, Youcheng Huang, Chao Wang, Shuo Zhang, Jiancheng Lv, Fuli Feng, Tat-Seng Chua. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

We introduce NExT-QA, a rigorously designed video question answering (VideoQA) benchmark to advance video understanding from describing to explaining the temporal actions. Based on the dataset, we set up multi-choice and open-ended QA tasks targeting causal action reasoning, temporal action reasoning, and common scene comprehension. Through extensive analysis of baselines and established VideoQA techniques, we find that top-performing methods excel at shallow scene descriptions but are weak in causal and temporal action reasoning. Furthermore, the models that are effective on multi-choice QA, when adapted to open-ended QA, still struggle in generalizing the answers. This raises doubt on the ability of these models to reason and highlights possibilities for improvement. With detailed results for different question types and heuristic observations for future works, we hope NExT-QA will guide the next generation of VQA research to go beyond superficial description towards a deeper understanding of videos. (The dataset and related resources are available at https://github.com/doc-doc/NExT-QA.git).

Many existing approaches for 3D point cloud semantic segmentation are fully supervised. These fully supervised approaches heavily rely on large amounts of labeled training data that are difficult to obtain and cannot segment new classes after training. To mitigate these limitations, we propose a novel attention-aware multi-prototype transductive few-shot point cloud semantic segmentation method to segment new classes given a few labeled examples. Specifically, each class is represented by multiple prototypes to model the complex data distribution of labeled points. Subsequently, we employ a transductive label propagation method to exploit the affinities between labeled multi-prototypes and unlabeled points, and among the unlabeled points. Furthermore, we design an attention-aware multi-level feature learning network to learn the discriminative features that capture the geometric dependencies and semantic correlations between points. Our proposed method shows significant and consistent improvements compared to baselines in different few-shot point cloud semantic segmentation settings (i.e. 2/3-way 1/5-shot) on two benchmark datasets. Our code is available at https://github.com/Na-Z/attMPTI.

Recommender systems usually amplify the biases in the data. The model learned from historical interactions with imbalanced item distribution will amplify the imbalance by over-recommending items from the majority groups. Addressing this issue is essential for a healthy ecosystem of recommendation in the long run. Existing work applies bias control to the ranking targets (e.g., calibration, fairness, and diversity), but ignores the true reason for bias amplification and trades off the recommendation accuracy.

Sequential fashion recommendation is of great significance in online fashion shopping, which accounts for an increasing portion of either fashion retailing or online e-commerce. The key to building an effective sequential fashion recommendation model lies in capturing two types of patterns: the personal fashion preference of users and the transitional relationships between adjacent items. The two types of patterns are usually related to user-item interaction and item-item transition modeling respectively. However, due to the large sets of users and items as well as the sparse historical interactions, it is difficult to train an effective and efficient sequential fashion recommendation model. To tackle these problems, we propose to leverage two types of global graph, i.e., the user-item interaction graph and item-item transition graph, to obtain enhanced user and item representations by incorporating higher-order connections over the graphs. In addition, we adopt the graph kernel of LightGCN [9] for the information propagation in both graphs and propose a new design for item-item transition graph. Extensive experiments on two established sequential fashion recommendation datasets validate the effectiveness and efficiency of our approach.

This companion paper supports the replication of the fashion trend forecasting experiments with the KERN (Knowledge Enhanced Recurrent Network) method that we presented in the ICMR 2020. We provide an artifact that allows the replication of the experiments using a Python implementation. The artifact is easy to deploy with simple installation, training and evaluation. We reproduce the experiments conducted in the original paper and obtain similar performance as previously reported. The replication results of the experiments support the main claims in the original paper.

This work targets at the problems of comprehensive video captioning and the generation of multiple descriptions from different perspectives, termed asMulti-Perspective Video Captioning. We build and release a dataset named VidOR-MPVC, the first dataset for multi-perspective video captioning, where each video is annotated with multiple descriptions from different perspectives. We also propose a novel model, dubbedperspective-aware captioner (PAC), which is capable of mining the various perspectives in a video and generating a description from each perspective. More specifically, a perspective generator is designed to perceive video content with perspective preferences, and followed by a language generator equipped with perspective-aware attention mechanism. As our new task expects to produce multiple descriptions for a video, existing evaluation metrics are fail to handle this situation. To address this problem, we devise the maximum matching scores based on existing metrics for an overall evaluation which aims to cover the aspects of semantic similarity, completeness and compactness. The experimental results demonstrate that our model is able to describe videos with multiple descriptions from different perspectives.

ACM Multimedia 2021 Video Relation Understanding Challenge is the third grand challenge which aims at exploring the relationship of subjects and objects appearing in videos for fine-grained and high-level video understanding. Given a video, the video relation detection model should output a serious of relation triplet subject, predicate, object and the corresponding trajectories of subject and object. The goal of this task is to promote research on developing video semantic understanding model, so as to perform complex inferences and mining of visual knowledge in videos. In this paper, we make a comprehensive and detailed introduction of this task, conclude the proposed algorithms in the last few years, and propose future direction for research in this task.

Video Visual Relation Detection (VidVRD) aims to semantically describe the dynamic interactions across visual concepts localized in a video in the form of subject, predicate, object. It can help to mitigate the semantic gap between vision and language in video understanding, thus receiving increasing attention in multimedia communities. Existing efforts primarily leverage the multimodal/spatio-temporal feature fusion to augment the representation of object trajectories as well as their interactions and formulate the prediction of predicates as a multi-class classification task. Despite their effectiveness, existing models ignore the severe long-tailed bias in VidVRD datasets. As a result, the models' prediction will be easily biased towards the popular head predicates (e.g., next-to and in-front-of), thus leading to poor generalizability.

The world has become multimodal. In addition to text, we have been sharing a huge amount of multimedia information in the form of images and videos on the Internet. The wide spread use of smart mobile devices has also changed the way we interact with the Internet. It is now natural for us to capture images and videos freely and use as part of a query, in addition to the traditional text and voices. These, along with the rapid advancements in multimedia, natural language processing, information retrieval, and conversation technologies, mean that it is time for us to explore multimodal conversation and its roles in search and recommendation. Multimodal conversation has the potential to help us to uncover and digest the huge amount of multimedia information and knowledge hidden within many systems. It also enables a natural 2-way interactions between humans and machines, with mutual benefits in enriching their respective knowledge. Finally, it opens up the possibilities of disrupting many existing applications and launching new innovative applications. This panel is timely and aims to explore this emerging trend, and discuss its potential benefits and pitfalls to society. The panel will also explore the limitations of current technologies and highlight future research directions towards developing a multimedia conversational system.

Understanding complex processes that give cities their form traditionally relied primarily on the analysis of various open data statistics in relation to e.g. neighbourhood demographics, economy and mobility. However, recent years have seen an unprecedented increase in the availability and use of city-related sensors, participatory data and social multimedia. As the valuable information about urban challenges is usually encoded across multiple modalities, such as visual (e.g. panoramic, satellite and user-contributed images), text (e.g. social media and participatory data) and open data statistics, extracting this information requires effective multimedia analysis tools. This Workshop will showcase the power of multimedia computing in addressing various urban challenges, ranging from event detection and analysis, location recommendation and crowdedness estimation to more efficient handling of citizen reports and modelling and improving city liveability. In addition, it will serve as an impulse for the multimedia community to intensify research on these interesting, challenging and truly multimodal problems.

The core problem of video visual relation detection (VidVRD) lies in accurately classifying the relation triplets, which comprise of the classes of subject and object entities, and the predicate classes of various relationships between them. Existing VidVRD approaches classify these three relation components in either independent or cascaded manner, thus fail to fully exploit the inter-dependency among them. In order to utilize this inter-dependency in tackling the challenges of visual relation recognition in videos, we propose a novel iterative relation inference approach for VidVRD. We derive our model from the viewpoint of joint relation classification which is light-weight yet effective, and propose a training approach to better learn the dependency knowledge from the likely correct triplet combinations. As such, the proposed inference approach is able to gradually refine each component based on its learnt dependency and the other two's predictions. Our ablation studies show that this iterative relation inference can empirically converge in a few steps and consistently boost the performance over baselines. Further, we incorporate it into a newly designed VidVRD architecture, named VidVRD-II (Iterative Inference), which generalizes well across different datasets. Experiments show that VidVRD-II achieves the start-of-the-art performance on both of ImageNet-VidVRD and VidOR benchmark datasets.

Recommending purely cold-start items is a long-standing and fundamental challenge in the recommender systems. Without any historical interaction on cold-start items, the collaborative filtering (CF) scheme fails to leverage collaborative signals to infer user preference on these items. To solve this problem, extensive studies have been conducted to incorporate side information of items (e.g. content features) into the CF scheme. Specifically, they employ modern neural network techniques (e.g., dropout, consistency constraint) to discover and exploit the coalition effect of content features and collaborative representations. However, we argue that these works less explore the mutual dependencies between content features and collaborative representations and lack sufficient theoretical supports, thus resulting in unsatisfactory performance on cold-start recommendation.

There is a soaring interest in the news recommendation research scenario due to the information overload. To accurately capture users' interests, we propose to model multi-modal features, in addition to the news titles that are widely used in existing works, for news recommendation. Besides, existing research pays little attention to the click decision-making process in designing multi-modal modeling modules. In this work, inspired by the fact that users make their click decisions mostly based on the visual impression they perceive when browsing news, we propose to capture such visual impression information with visual-semantic modeling for news recommendation. In this paper, we refer to visual impression as the region of the news displayed on the user interface of a news application, which delivers both content and layout information to users. Specifically, we devise the local impression modeling module to simultaneously attend to decomposed details in the impression when understanding the semantic meaning of news title, which could explicitly get close to the process of users reading news. In addition, we inspect the impression from a global view and take structural information, such as the arrangement of different fields and spatial position of different words on the impression, into the modeling of multiple modalities. To accommodate the research of visual impression-aware news recommendation, we extend the text-dominated news recommendation dataset MIND by adding snapshot impression images and will release it to nourish the research field. Extensive comparisons with the state-of-the-art news recommenders along with the in-depth analyses demonstrate the effectiveness of the proposed method and the promising capability of modeling visual impressions for the content-based recommenders.

Zhenghao Liu, Xiaoyuan Yi, Maosong Sun, Liner Yang, Tat-Seng Chua. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

When a graph neural network (GNN) made a prediction, one raises question about explainability: “Which fraction of the input graph is most influential to the model’s decision?” Producing an answer requires understanding the model’s inner workings in general and emphasizing the insights on the decision for the instance at hand. Nonetheless, most of current approaches focus only on one aspect: (1) local explainability, which explains each instance independently, thus hardly exhibits the class-wise patterns; and (2) global explainability, which systematizes the globally important patterns, but might be trivial in the local context. This dichotomy limits the flexibility and effectiveness of explainers greatly. A performant paradigm towards multi-grained explainability is until-now lacking and thus a focus of our work. In this work, we exploit the pre-training and fine-tuning idea to develop our explainer and generate multi-grained explanations. Specifically, the pre-training phase accounts for the contrastivity among different classes, so as to highlight the class-wise characteristics from a global view; afterwards, the fine-tuning phase adapts the explanations in the local context. Experiments on both synthetic and real-world datasets show the superiority of our explainer, in terms of AUC on explaining graph classification over the leading baselines. Our codes and datasets are available at https://github.com/Wuyxin/ReFine.

Graph Convolutional Network (GCN) is an emerging technique for information retrieval (IR) applications. While GCN assumes the homophily property of a graph, real-world graphs are never perfect: the local structure of a node may contain discrepancy, e.g., the labels of a node's neighbors could vary. This pushes us to consider the discrepancy of local structure in GCN modeling. Existing work approaches this issue by introducing an additional module such as graph attention, which is expected to learn the contribution of each neighbor. However, such module may not work reliably as expected, especially when there lacks supervision signal, e.g., when the labeled data is small. Moreover, existing methods focus on modeling the nodes in the training data, and never consider the local structure discrepancy of testing nodes.

The financial markets are moved by events such as the issuance of administrative orders. The participants in financial markets (e.g., traders) thus pay constant attention to financial news relevant to the financial asset (e.g., oil) of interest. Due to the large scale of news stream, it is time and labor intensive to manually identify influential events that can move the price of the financial asset, pushing the financial participants to embrace automatic financial event ranking, which has received relatively little scrutiny to date. In this work, we formulate the financial event ranking task, which aims to score financial news (document) according to its influence to the given asset (query). To solve this task, we propose a Hybrid News Ranking framework that, from the asset perspective, evaluates the influence of news articles by comparing their contents; and from the event perspective, accesses the influence over all query assets. Moreover, we resolve the dilemma between the essential requirement of sufficient labels for training the framework and the unaffordable cost of hiring domain experts for labeling the news. In particular, we design a cost-friendly system for news labeling that leverages the knowledge within published financial analyst reports. In this way, we construct three financial event ranking datasets. Extensive experiments on the datasets validate the effectiveness of the proposed framework and the rationality of solving financial event ranking through learning to rank.

Although conversational search has become a hot topic in both dialogue research and IR community, the real breakthrough has been limited by the scale and quality of datasets available. To address this fundamental obstacle, we introduce the Multimodal Multi-domain Conversational dataset (MMConv), a fully annotated collection of human-to-human role-playing dialogues spanning over multiple domains and tasks. The contribution is two-fold. First, beyond the task-oriented multimodal dialogues among user and agent pairs, dialogues are fully annotated with dialogue belief states and dialogue acts. More importantly, we create a relatively comprehensive environment for conducting multimodal conversational search with real user settings, structured venue database, annotated image repository as well as crowd-sourced knowledge database. A detailed description of the data collection procedure along with a summary of data structure and analysis is provided. Second, a set of benchmark results for dialogue state tracking, conversational recommendation, response generation as well as a unified model for multiple tasks are reported. We adopt the state-of-the-art methods for these tasks respectively to demonstrate the usability of the data, discuss limitations of current methods and set baselines for future studies.

Recommendation is a prevalent and critical service in information systems. To provide personalized suggestions to users, industry players embrace machine learning, more specifically, building predictive models based on the click behavior data. This is known as the Click-Through Rate (CTR) prediction, which has become the gold standard for building personalized recommendation service. However, we argue that there is a significant gap between clicks and user satisfaction --- it is common that a user is "cheated" to click an item by the attractive title/cover of the item. This will severely hurt user's trust on the system if the user finds the actual content of the clicked item disappointing. What's even worse, optimizing CTR models on such flawed data will result in the Matthew Effect, making the seemingly attractive but actually low-quality items be more frequently recommended.

We tackle the task of video moment retrieval (VMR), which aims to localize a specific moment in a video according to a textual query. Existing methods primarily model the matching relationship between query and moment by complex cross-modal interactions. Despite their effectiveness, current models mostly exploit dataset biases while ignoring the video content, thus leading to poor generalizability. We argue that the issue is caused by the hidden confounder in VMR, i.e., temporal location of moments, that spuriously correlates the model input and prediction. How to design robust matching models against the temporal location biases is crucial but, as far as we know, has not been studied yet for VMR.

Learning user representations based on historical behaviors lies at the core of modern recommender systems. Recent advances in sequential recommenders have convincingly demonstrated high capability in extracting effective user representations from the given behavior sequences. Despite significant progress, we argue that solely modeling the observational behaviors sequences may end up with a brittle and unstable system due to the noisy and sparse nature of user interactions logged. In this paper, we propose to learn accurate and robust user representations, which are required to be less sensitive to (attack on) noisy behaviors and trust more on the indispensable ones, by modeling counterfactual data distribution. Specifically, given an observed behavior sequence, the proposed CauseRec framework identifies dispensable and indispensable concepts at both the fine-grained item level and the abstract interest level. CauseRec conditionally samples user concept sequences from the counterfactual data distributions by replacing dispensable and indispensable concepts within the original concept sequence. With user representations obtained from the synthesized user sequences, CauseRec performs contrastive user representation learning by contrasting the counterfactual with the observational. We conduct extensive experiments on real-world public recommendation benchmarks and justify the effectiveness of CauseRec with multi-aspects model analysis. The results demonstrate that the proposed CauseRec outperforms state-of-the-art sequential recommenders by learning accurate and robust user representations.

The ubiquity of implicit feedback makes them the default choice to build online recommender systems. While the large volume of implicit feedback alleviates the data sparsity issue, the downside is that they are not as clean in reflecting the actual satisfaction of users. For example, in E-commerce, a large portion of clicks do not translate to purchases, and many purchases end up with negative reviews. As such, it is of critical importance to account for the inevitable noises in implicit feedback for recommender training. However, little work on recommendation has taken the noisy nature of implicit feedback into consideration.

Multi-domain dialogue state tracking (DST) is a critical component for monitoring user goals during the course of an interaction. Existing approaches have relied on dialogue history indiscriminately or updated on the most recent turns incrementally. However, in spite of modeling it based on fixed ontology or open vocabulary, the former setting violates the interactive and progressing nature of dialogue, while the later easily gets affected by the error accumulation conundrum. Here, we propose a Recursive Inference mechanism (ReInf) to resolve DST in multi-domain scenarios that call for more robust and accurate tracking capability. Specifically, our agent reversely reviews the dialogue history until the agent has pinpointed sufficient turns confidently for slot value prediction. It also recursively factors in potential dependencies among domains and slots to further solve the co-reference and value sharing problems. The quantitative and qualitative experimental results on the MultiWOZ 2.1 corpus demonstrate that the proposed ReInf not only outperforms the state-of-the-art methods, but also achieves reasonable turn reference and interpretable slot co-reference.

Knowledge graph (KG) plays an increasingly important role in recommender systems. A recent technical trend is to develop end-to-end models founded on graph neural networks (GNNs). However, existing GNN-based models are coarse-grained in relational modeling, failing to (1) identify user-item relation at a fine-grained level of intents, and (2) exploit relation dependencies to preserve the semantics of long-range connectivity.

Abstract Tracking dialogue states to better interpret user goals and feed downstream policy learning is a bottleneck in dialogue management. Common practice has been to treat it as a problem of classifying dialogue content into a set of pre-defined slot-value pairs, or generating values for different slots given the dialogue history. Both have limitations on considering dependencies that occur on dialogues, and are lacking of reasoning capabilities. This paper proposes to track dialogue states gradually with reasoning over dialogue turns with the help of the back-end data. Empirical results demonstrate that our method outperforms the state-of-the-art methods in terms of joint belief accuracy for MultiWOZ 2.1, a large-scale human--human dialogue dataset across multiple domains.

Food recognition has captured numerous research attention for its importance for health-related applications. The existing approaches mostly focus on the categorization of food according to dish names, while ignoring the underlying ingredient composition. In reality, two dishes with the same name do not necessarily share the exact list of ingredients. Therefore, the dishes under the same food category are not mandatorily equal in nutrition content. Nevertheless, due to limited datasets available with ingredient labels, the problem of ingredient recognition is often overlooked. Furthermore, as the number of ingredients is expected to be much less than the number of food categories, ingredient recognition is more tractable in the real-world scenario. This paper provides an insightful analysis of three compelling issues in ingredient recognition. These issues involve recognition in either image-level or region level, pooling in either single or multiple image scales, learning in either single or multi-task manner. The analysis is conducted on a large food dataset, Vireo Food-251, contributed by this paper. The dataset is composed of 169,673 images with 251 popular Chinese food and 406 ingredients. The dataset includes adequate challenges in scale and complexity to reveal the limit of the current approaches in ingredient recognition.

Recent efforts show that neural networks are vulnerable to small but intentional perturbations on input features in visual classification tasks. Due to the additional consideration of connections between examples (e.g., articles with citation link tend to be in the same class), graph neural networks could be more sensitive to the perturbations, since the perturbations from connected examples exacerbate the impact on a target example. Adversarial Training (AT), a dynamic regularization technique, can resist the worst-case perturbations on input features and is a promising choice to improve model robustness and generalization. However, existing AT methods focus on standard classification, being less effective when training models on graph since it does not model the impact from connected examples. In this work, we explore adversarial training on graph, aiming to improve the robustness and generalization of models learned on graph. We propose Graph Adversarial Training (GraphAT), which takes the impact from connected examples into account when learning to construct and resist perturbations. We give a general formulation of GraphAT, which can be seen as a dynamic regularization scheme based on the graph structure. To demonstrate the utility of GraphAT, we employ it on a state-of-the-art graph neural network model - Graph Convolutional Network (GCN). We conduct experiments on two citation graphs (Citeseer and Cora) and a knowledge graph (NELL), verifying the effectiveness of GraphAT which outperforms normal training on GCN by 4.51 percent in node classification accuracy. Codes are available via: https://github.com/fulifeng/GraphAT.

Most existing recommender systems leverage user behavior data of one type only, such as the purchase behavior in E-commerce that is directly related to the business Key Performance Indicator (KPI) of conversion rate. Besides the key behavioral data, we argue that other forms of user behaviors also provide valuable signal, such as views, clicks, adding a product to shopping carts and so on. They should be taken into account properly to provide quality recommendation for users. In this work, we contribute a new solution named short for Neural Multi-Task Recommendation (NMTR) for learning recommender systems from user multi-behavior data. We develop a neural network model to capture the complicated and multi-type interactions between users and items. In particular, our model accounts for the cascading relationship among different types of behaviors (e.g., a user must click on a product before purchasing it). To fully exploit the signal in the data of multiple types of behaviors, we perform a joint optimization based on the multi-task learning framework, where the optimization on a behavior is treated as a task. Extensive experiments on two real-world datasets demonstrate that NMTR significantly outperforms state-of-the-art recommender systems that are designed to learn from both single-behavior data and multi-behavior data. Further analysis shows that modeling multiple behaviors is particularly useful for providing recommendation for sparse users that have very few interactions.

Modeling the structure of culinary recipes is the core of recipe representation learning. Current approaches mostly focus on extracting the workflow graph from recipes based on text descriptions. Process images, which constitute an important part of cooking recipes, has rarely been investigated in recipe structure modeling. We study this recipe structure problem from a multi-modal learning perspective, by proposing aprerequisite treeto represent recipes with cooking images at a step-level granularity. We propose a simple-yet-effective two-stage framework to automatically construct the prerequisite tree for a recipe by (1) utilizing a trained classifier to detect pairwise prerequisite relations that fuses multi-modal features as input; then (2) applying different strategies (greedy method, maximum weight, and beam search) to build the tree structure. Experiments on the MM-ReS dataset demonstrates the advantages of introducing process images for recipe structure modeling. Also, compared with neural methods which require large numbers of training data, we show that our two-stage pipeline can achieve promising results using only 400 labeled prerequisite trees as training data.

With superhuman-level performance of face recognition, we are more concerned about the recognition of fine-grained attributes, such as emotion, age, and gender. However, given that the label space is extremely large and follows a long-tail distribution, it is quite expensive to collect sufficient samples for fine-grained attributes. This results in imbalanced training samples and inferior attribute recognition models. To this end, we propose the use of arbitrary attribute combinations, without human effort, to synthesize face images. In particular, to bridge the semantic gap between high-level attribute label space and low-level face image, we propose a novel neural-network-based approach that maps the target attribute labels to an embedding vector, which can be fed into a pretrained image decoder to synthesize a new face image. Furthermore, to regularize the attribute for image synthesis, we propose to use a perceptual loss to make the new image explicitly faithful to target attributes. Experimental results show that our approach can generate photorealistic face images from attribute labels, and more importantly, by serving as augmented training samples, these images can significantly boost the performance of attribute recognition model. The code is open-sourced at this link.

Static recommendation methods like collaborative filtering suffer from the inherent limitation of performing real-time personalization for cold-start users. Online recommendation, e.g., multi-armed bandit approach, addresses this limitation by interactively exploring user preference online and pursuing the exploration-exploitation (EE) trade-off. However, existing bandit-based methods model recommendation actions homogeneously. Specifically, they only consider the items as the arms, being incapable of handling the item attributes , which naturally provide interpretable information of user’s current demands and can effectively filter out undesired items. In this work, we consider the conversational recommendation for cold-start users, where a system can both ask the attributes from and recommend items to a user interactively. This important scenario was studied in a recent work [54]. However, it employs a hand-crafted function to decide when to ask attributes or make recommendations. Such separate modeling of attributes and items makes the effectiveness of the system highly rely on the choice of the hand-crafted function, thus introducing fragility to the system. To address this limitation, we seamlessly unify attributes and items in the same arm space and achieve their EE trade-offs automatically using the framework of Thompson Sampling. Our Conversational Thompson Sampling (ConTS) model holistically solves all questions in conversational recommendation by choosing the arm with the maximal reward to play. Extensive experiments on three benchmark datasets show that ConTS outperforms the state-of-the-art methods Conversational UCB (ConUCB) [54] and Estimation—Action—Reflection model [27] in both metrics of success rate and average number of conversation turns.

Recognizing ingredients for a given dish image is at the core of automatic dietary assessment, attracting increasing attention from both industry and academia. Nevertheless, the task is challenging due to the difficulty of collecting and labeling sufficient training data. On one hand, there are hundred thousands of food ingredients in the world, ranging from the common to rare. Collecting training samples for all of the ingredient categories is difficult. On the other hand, as the ingredient appearances exhibit huge visual variance during the food preparation, it requires to collect the training samples under different cooking and cutting methods for robust recognition. Since obtaining sufficient fully annotated training data is not easy, a more practical way of scaling up the recognition is to develop models that are capable of recognizing unseen ingredients. Therefore, in this paper, we target the problem of ingredient recognition with zero training samples. More specifically, we introduce multi-relational GCN (graph convolutional network) that integrates ingredient hierarchy, attribute as well as co-occurrence for zero-shot ingredient recognition. Extensive experiments on both Chinese and Japanese food datasets are performed to demonstrate the superior performance of multi-relational GCN and shed light on zero-shot ingredients recognition.

Existing domain adaptation methods on visual sentiment classification typically are investigated under the single-source scenario, where the knowledge learned from a source domain of sufficient labeled data is transferred to the target domain of loosely labeled or unlabeled data. However, in practice, data from a single source domain usually have a limited volume and can hardly cover the characteristics of the target domain. In this paper, we propose a novel multi-source domain adaptation (MDA) method, termed Multi-source Sentiment Generative Adversarial Network (MSGAN), for visual sentiment classification. To handle data from multiple source domains, it learns to find a unified sentiment latent space where data from both the source and target domains share a similar distribution. This is achieved via cycle consistent adversarial learning in an end-to-end manner. Extensive experiments conducted on four benchmark datasets demonstrate that MSGAN significantly outperforms the state-of-the-art MDA approaches for visual sentiment classification.

With the rapid expansion of digital music formats, it's indispensable to recommend users with their favorite music. For music recommendation, users' personality and emotion greatly affect their music preference, respectively in a long-term and short-term manner, while rich social media data provides effective feedback on these information. In this paper, aiming at music recommendation on social media platforms, we propose a Personality and Emotion Integrated Attentive model (PEIA), which fully utilizes social media data to comprehensively model users' long-term taste (personality) and short-term preference (emotion). Specifically, it takes full advantage of personality-oriented user features, emotion-oriented user features and music features of multi-faceted attributes. Hierarchical attention is employed to distinguish the important factors when incorporating the latent representations of users' personality and emotion. Extensive experiments on a large real-world dataset of 171,254 users demonstrate the effectiveness of our PEIA model which achieves an NDCG of 0.5369, outperforming the state-of-the-art methods. We also perform detailed parameter analysis and feature contribution analysis, which further verify our scheme and demonstrate the significance of co-modeling of user personality and emotion in music recommendation.

Event detection is a crucial and challenging sub-task of event extraction, which suffers from a severe ambiguity issue of trigger words. Existing works mainly focus on using textual context information, while there naturally exist many images accompanied by news articles that are yet to be explored. We believe that images not only reflect the core events of the text, but are also helpful for the disambiguation of trigger words. In this paper, we first contribute an image dataset supplement to ED benchmarks (i.e., ACE2005) for training and evaluation. We then propose a novel Dual Recurrent Multimodal Model, DRMM, to conduct deep interactions between images and sentences for modality features aggregation. DRMM utilizes pre-trained BERT and ResNet to encode sentences and images, and employs an alternating dual attention to select informative features for mutual enhancements. Our superior performance compared to six state-of-art baselines as well as further ablation studies demonstrate the significance of image modality and effectiveness of the proposed architecture. The code and image dataset are avaliable at https://github.com/shuaiwa16/image-enhanced-event-extraction.

We study the problem of attacking video recognition models in the black-box setting, where the model information is unknown and the adversary can only make queries to detect the predicted top-1 class and its probability. Compared with the black-box attack on images, attacking videos is more challenging as the computation cost for searching the adversarial perturbations on a video is much higher due to its high dimensionality. To overcome this challenge, we propose a heuristic black-box attack model that generates adversarial perturbations only on the selected frames and regions. More specifically, a heuristic-based algorithm is proposed to measure the importance of each frame in the video towards generating the adversarial examples. Based on the frames' importance, the proposed algorithm heuristically searches a subset of frames where the generated adversarial example has strong adversarial attack ability while keeps the perturbations lower than the given bound. Besides, to further boost the attack efficiency, we propose to generate the perturbations only on the salient regions of the selected frames. In this way, the generated perturbations are sparse in both temporal and spatial domains. Experimental results of attacking two mainstream video recognition methods on the UCF-101 dataset and the HMDB-51 dataset demonstrate that the proposed heuristic black-box adversarial attack method can significantly reduce the computation cost and lead to more than 28% reduction in query numbers for the untargeted attack on both datasets.

Online Social Networks (OSNs) evolve through two pervasive behaviors: follow and unfollow, which respectively signify relationship creation and relationship dissolution. Researches on social network evolution mainly focus on the follow behavior, while the unfollow behavior has largely been ignored. Mining unfollow behavior is challenging because user's decision on unfollow is not only affected by the simple combination of user's attributes like informativeness and reciprocity, but also affected by the complex interaction among them. Meanwhile, prior datasets seldom contain sufficient records for inferring such complex interaction. To address these issues, we first construct a large-scale real-world Weibo1 dataset, which records detailed post content and relationship dynamics of 1.8 million Chinese users. Next, we define user's attributes as two categories: spatial attributes (e.g., social role of user) and temporal attributes (e.g., post content of user). Leveraging the constructed dataset, we systematically study how the interaction effects between user's spatial and temporal attributes contribute to the unfollow behavior. Afterwards, we propose a novel unified model with heterogeneous information (UMHI) for unfollow prediction. Specifically, our UMHI model: 1) captures user's spatial attributes through social network structure; 2) infers user's temporal attributes through user-posted content and unfollow history; and 3) models the interaction between spatial and temporal attributes by the nonlinear MLP layers. Comprehensive evaluations on the constructed dataset demonstrate that the proposed UMHI model outperforms baseline methods by 16.44 on average in terms of precision. In addition, factor analyses verify that both spatial attributes and temporal attributes are essential for mining unfollow behavior.

The curse of knowledge can impede communication between experts and laymen.We propose a new task of expertise style transfer and contribute a manually annotated dataset with the goal of alleviating such cognitive biases.Solving this task not only simplifies the professional language, but also improves the accuracy and expertise level of laymen descriptions using simple words.This is a challenging task, unaddressed in previous work, as it requires the models to have expert intelligence in order to modify text with a deep understanding of domain knowledge and structures.We establish the benchmark performance of five stateof-the-art models for style transfer and text simplification.The results demonstrate a significant gap between machine and human performance.We also discuss the challenges of automatic evaluation, to provide insights into future research directions.

This paper proposes the problem of Deep Question Generation (DQG), which aims to generate complex questions that require reasoning over multiple pieces of information of the input passage.In order to capture the global structure of the document and facilitate reasoning, we propose a novel framework which first constructs a semantic-level graph for the input document and then encodes the semantic graph by introducing an attention-based GGNN (Att-GGNN).Afterwards, we fuse the document-level and graphlevel representations to perform joint training of content selection and question decoding.On the HotpotQA deep-question centric dataset, our model greatly improves performance over questions requiring reasoning over multiple facts, leading to state-of-theart performance.The code is publicly available at https://github.com/WING-NUS/ SG-Deep-Question-Generation.

This paper proposes a Hyperbolic Visual Embedding Learning Network for zero-shot recognition. The network learns image embeddings in hyperbolic space, which is capable of preserving the hierarchical structure of semantic classes in low dimensions. Comparing with existing zeroshot learning approaches, the network is more robust because the embedding feature in hyperbolic space better represents class hierarchy and thereby avoid misleading resulted from unrelated siblings. Our network outperforms exiting baselines under hierarchical evaluation with an extremely challenging setting, i.e., learning only from 1,000 categories to recognize 20,841 unseen categories. While under flat evaluation, it has competitive performance as state-of-the-art methods but with five times lower embedding dimensions. Our code is publicly available.

The performance of existing point cloud-based 3D object detection methods heavily relies on large-scale high-quality 3D annotations. However, such annotations are often tedious and expensive to collect. Semi-supervised learning is a good alternative to mitigate the data annotation issue, but has remained largely unexplored in 3D object detection. Inspired by the recent success of self-ensembling technique in semi-supervised image classification task, we propose SESS, a self-ensembling semi-supervised 3D object detection framework. Specifically, we design a thorough perturbation scheme to enhance generalization of the network on unlabeled and new unseen data. Furthermore, we propose three consistency losses to enforce the consistency between two sets of predicted 3D object proposals, to facilitate the learning of structure and semantic invariances of objects. Extensive experiments conducted on SUN RGB-D and ScanNet datasets demonstrate the effectiveness of SESS in both inductive and transductive semi-supervised 3D object detection. Our SESS achieves competitive performance compared to the state-of-the-art fully-supervised method by using only 50% labeled data. Our code is available at https://github.com/Na-Z/sess.

In existing sophisticated text-to-SQL models, schema linking is often considered as a simple, minor component, belying its importance.By providing a schema linking corpus based on the Spider text-to-SQL dataset, we systematically study the role of schema linking.We also build a simple BERT-based baseline, called Schema-Linking SQL (SLSQL) to perform a data-driven study.We find when schema linking is done well, SLSQL demonstrates good performance on Spider despite its structural simplicity.Many remaining errors are attributable to corpus noise.This suggests schema linking is the crux for the current textto-SQL task.Our analytic studies provide insights on the characteristics of schema linking for future developments of text-to-SQL tasks. 1

Entity alignment (EA) aims at building a unified Knowledge Graph (KG) of rich content by linking the equivalent entities from various KGs.GNN-based EA methods present promising performance by modeling the KG structure defined by relation triples.However, attribute triples can also provide crucial alignment signal but have not been well explored yet.In this paper, we propose to utilize an attributed value encoder and partition the KG into subgraphs to model the various types of attribute triples efficiently.Besides, the performances of current EA methods are overestimated because of the name-bias of existing EA datasets.To make an objective evaluation, we propose a hard experimental setting where we select equivalent entity pairs with very different names as the test set.Under both the regular and hard settings, our method achieves significant improvements (5.10% on average Hits@1 in DBP15k) over 12 baselines in crosslingual and monolingual datasets.Ablation studies on different subgraphs and a case study about attribute types further demonstrate the effectiveness of our method.Source code and data can be found at https://github.com/ thunlp/explore-and-evaluate.

In this paper, we present the PS2-Net - a locally and globally aware deep learning framework for semantic segmentation on 3D scene-level point clouds. In order to deeply incorporate local structures and global context to support 3D scene segmentation, our network is built on four repeatedly stacked encoders, where each encoder has two basic components: EdgeConv that captures local structures and NetVLAD that models global context. Different from existing state-of-the-art methods for point-based scene semantic segmentation that either violate or do not achieve permutation invariance, our PS2-Net is designed to be permutation invariant which is an essential property of any deep network used to process unordered point clouds. We further provide theoretical proof to guarantee the permutation invariance property of our network. We perform extensive experiments on two large-scale 3D indoor scene datasets and demonstrate that our PS2-Net is able to achieve state-of-the-art performances as compared to existing approaches.

Traditional recommendation systems estimate user preference on items from past interaction history, thus suffering from the limitations of obtaining fine-grained and dynamic user preference. Conversational recommendation system (CRS) brings revolutions to those limitations by enabling the system to directly ask users about their preferred attributes on items. However, existing CRS methods do not make full use of such advantage --- they only use the attribute feedback in rather implicit ways such as updating the latent user representation. In this paper, we propose Conversational Path Reasoning (CPR), a generic framework that models conversational recommendation as an interactive path reasoning problem on a graph. It walks through the attribute vertices by following user feedback, utilizing the user preferred attributes in an explicit way. By leveraging on the graph structure, CPR is able to prune off many irrelevant candidate attributes, leading to a better chance of hitting user-preferred attributes. To demonstrate how CPR works, we propose a simple yet effective instantiation named SCPR (Simple CPR). We perform empirical studies on the multi-round conversational recommendation scenario, the most realistic CRS setting so far that considers multiple rounds of asking attributes and recommending items. Through extensive experiments on two datasets Yelp and LastFM, we validate the effectiveness of our SCPR, which significantly outperforms the state-of-the-art CRS methods EAR and CRM. In particular, we find that the more attributes there are, the more advantages our method can achieve.

Fashion trend forecasting is a crucial task for both academia andindustry. Although some efforts have been devoted to tackling this challenging task, they only studied limited fashion elements with highly seasonal or simple patterns, which could hardly reveal thereal fashion trends. Towards insightful fashion trend forecasting,this work focuses on investigating fine-grained fashion element trends for specific user groups. We first contribute a large-scale fashion trend dataset (FIT) collected from Instagram with extracted time series fashion element records and user information. Furthermore, to effectively model the time series data of fashion elements with rather complex patterns, we propose a Knowledge Enhanced Recurrent Network model (KERN) which takes advantage of the capability of deep recurrent neural networks in modeling time series data. Moreover, it leverages internal and external knowledgein fashion domain that affects the time-series patterns of fashion element trends. Such incorporation of domain knowledge further enhances the deep learning model in capturing the patterns of specific fashion elements and predicting the future trends. Extensive experiments demonstrate that the proposed KERN model can effectively capture the complicated patterns of objective fashion elements, therefore making preferable fashion trend forecast.

The ability of recommending cold items (that have no behavior history) is a core strength of multimedia recommendation compared with behavior-only collaborative filtering. To learn effective item representation, a key challenge lies in the discrepancy between training and testing, since the cold items only exist in the testing data. This means that the signal used to represent an item varies during training and testing --- in the training stage, we can represent an item with both collaborative embedding and content embedding; whereas in the testing stage, we represent a cold item with content embedding only. Nevertheless, existing learning frameworks omit this critical discrepancy, resulting in suboptimal item representation for multimedia recommendation.

Definitive embeddings remain a fundamental challenge of computational musicology for symbolic music in deep learning today. Analogous to natural language, music can be modeled as a sequence of tokens. This motivates the majority of existing solutions to explore the utilization of word embedding models to build music embeddings. However, music differs from natural languages in two key aspects: (1) musical token is multi-faceted -- it comprises of pitch, rhythm and dynamics information; and (2) musical context is two-dimensional -- each musical token is dependent on both melodic and harmonic contexts. In this work, we provide a comprehensive solution by proposing a novel framework named PiRhDy that integrates pitch, rhythm, and dynamics information seamlessly. PiRhDy adopts a hierarchical strategy which can be decomposed into two steps: (1) token (i.e., note event) modeling, which separately represents pitch, rhythm, and dynamics and integrates them into a single token embedding; and (2) context modeling, which utilizes melodic and harmonic knowledge to train the token embedding. A thorough study was made on each component and sub-strategy of PiRhDy.We further validate our embeddings in three downstream tasks -- melody completion, accompaniment suggestion, and genre classification. Results indicate a significant advancement of the neural approach towards symbolic music as well as PiRhDy's potential as a pretrained tool for a broad range of symbolic music applications.

Anomaly detection in surveillance videos, as a special case of video-based action recognition, has been of increasing interest in multimedia community and public security. Action recognition in videos faces some challenges, such as cluttered background, illumination conditions. Besides these above difficulties, detecting anomaly in surveillance videos has several unique problems to be solved. For example, the lack of sufficient training samples is one of the main challenges for detecting anomalies in surveillance videos. In this paper, we propose to utilize transfer learning to leverage the good results from action recognition for anomaly detection in surveillance videos. More specially, we explore some techniques based on action recognition models from the following aspects: training samples, temporal modules for action recognition, network backbones. We draw some conclusions. First, more training samples from surveillance videos lead to higher classification accuracy. Second, stronger temporal modules designed for recognizing action and deeper networks do not achieve better results. This conclusion is reasonable since deeper networks tend to over-fitting, especially for the small-scale training set. Besides, to distinguish the hard examples from normal activities, we separately train a neural network to classify the hard category and normal events. Then we fuse the binary network and previous network to generate the final prediction for general anomaly detection. On the benchmarks of CitySCENE, our framework achieves promising performance and obtains the first prize for general anomaly detection and the second prize for specific anomaly detection.

Visually-aware food recommendation recommends food items based on their visual features. Existing methods typically use the pre-extracted visual features from food classification models, which mainly encode the visual content with limited semantic information, such as the classes and ingredients. Therefore, such features may not cover the personalized visual preferences of users, termed collaborative information, e.g. users may attend to different colors and textures of food based on their preferred ingredients and cooking methods. To address this problem, this paper presents a heterogeneous multi-task learning framework, termed privileged-channel infused network (PiNet). It learns the visual features that contain both the semantic and collaborative information by training the image encoder to simultaneously fulfill the ingredient prediction and food recommendation tasks. However, the heterogeneity between the two tasks may lead to different visual information in need and different directions in model parameter optimization. To handle these challenges, PiNet first employs a dual-gating module (DGM) to enable the encoding and passing of different visual information from the image encoder to individual tasks. Secondly, PiNet adopts a two-phase training strategy and two prior knowledge incorporation methods to ensure an effective model training. Experimental results from two real-world datasets show that the visual features generated by PiNet better attend to the informative image regions, yielding superior performance.

Understanding food recipe requires anticipating the implicit causal effects of cooking actions, such that the recipe can be converted into a graph describing the temporal workflow of the recipe. This is a non-trivial task that involves common-sense reasoning. However, existing efforts rely on hand-crafted features to extract the workflow graph from recipes due to the lack of large-scale labeled datasets. Moreover, they fail to utilize the cooking images, which constitute an important part of food recipes. In this paper, we build MM-ReS, the first large-scale dataset for cooking workflow construction, consisting of 9,850 recipes with human-labeled workflow graphs. Cooking steps are multi-modal, featuring both text instructions and cooking images. We then propose a neural encoder-decoder model that utilizes both visual and textual information to construct the cooking workflow, which achieved over 20% performance gain over existing hand-crafted baselines.

Video visual relation detection (VidVRD) aims at obtaining not only the trajectories of objects but also the dynamic visual relations between them. It provides abundant information for video understanding and can serve as a bridge between vision and language. Compared with visual relation detection on image, VidVRD requires one more step at last called visual relation association which associates relation segments across time dimension into video relations. This step plays an important role in the task but is less studied. Nevertheless, visual relation association is a difficult task as the association process is easily affected by inaccurate tracklet detection and relation prediction in the former steps. In this paper, we propose a novel relation association method called Multiple Hypothesis Association (MHA). It maintains multiple possible relation hypothesis during the association process in order to tolerate and handle the inaccurate or missing problem in the former steps and generate more accurate video relations. Our experiments on the benchmark datasets (Imagenet-VidVRD and VidOR) show that our method outperforms the state-of-the-art methods.

Reorganizing implicit feedback of users as a user-item interaction graph facilitates the applications of graph convolutional networks (GCNs) in recommendation tasks. In the interaction graph, edges between user and item nodes function as the main element of GCNs to perform information propagation and generate informative representations. Nevertheless, an underlying challenge lies in the quality of interaction graph, since observed interactions with less-interested items occur in implicit feedback (say, a user views micro-videos accidentally). This means that the neighborhoods involved with such false-positive edges will be influenced negatively and the signal on user preference can be severely contaminated. However, existing GCN-based recommender models leave such challenge under-explored, resulting in suboptimal representations and performance.

Photo-realistic free-viewpoint rendering of real-world scenes using classical computer graphics techniques is challenging, because it requires the difficult step of capturing detailed appearance and geometry models. Recent studies have demonstrated promising results by learning scene representations that implicitly encodes both geometry and appearance without 3D supervision. However, existing approaches in practice often show blurry renderings caused by the limited network capacity or the difficulty in finding accurate intersections of camera rays with the scene geometry. Synthesizing high-resolution imagery from these representations often requires time-consuming optical ray marching. In this work, we introduce Neural Sparse Voxel Fields (NSVF), a new neural scene representation for fast and high-quality free-viewpoint rendering. The NSVF defines a series of voxel-bounded implicit fields organized in a sparse voxel octree to model local properties in each cell. We progressively learn the underlying voxel structures with a differentiable ray-marching operation from only a set of posed RGB images. With the sparse voxel octree structure, rendering novel views at inference time can be accelerated by skipping the voxels without relevant scene content. Our method is over 10 times faster than the state-of-the-art while achieving higher quality results. Furthermore, by utilizing an explicit sparse voxel representation, our method can be easily applied to scene editing and scene composition. we also demonstrate various kinds of challenging tasks, including multi-object learning, free-viewpoint rendering of a moving human, and large-scale scene rendering.

Text classification in low-resource languages (eg Thai) is of great practical value for some information retrieval applications (eg sentiment-analysis-based restaurant recommendation). Due to lacking large-scale corpus for learning comprehensive text representation, bilingual text classification which borrows the linguistics knowledge from a rich-resource language becomes a promising solution. Despite the success of bilingual methods, they largely ignore another source of semantic information---the writing system. Noting that most low-resource languages are phonographic languages, we argue that a logographic language (eg Chinese) can provide helpful information for improving some phonographic languages' text classification, since a logographic character (ie logogram) could represent a sememe or a whole concept, not only a phoneme or a sound. In this paper, by using a phonographic labeled corpus and its machine-translated logographic corpus both, we devise a framework to explore the central theme of utilizing logograms as a "semantic detection assistant''. Specifically, from a logographic labeled corpus, we first devise a statistical-significance-based module to pick out informative text pieces. To represent them and further reduce the effects of translation errors, our approach is equipped with Gaussian embedding whose covariances serve as reliable signals of translation errors. For a test document, all seeds' Gaussian representations are used to convolute the document and produce a logographic embedding, before being fused with its phonographic embedding for final prediction. Extensive experiments validate the effectiveness of our approach and further investigations show its generalizability and robustness.

This half-day workshop explores challenges and potential research directions about Information Retrieval (IR) in finance. The focus will be on stimulating discussions around the accessing, searching, filtering, and analyzing financial documents in banking, insurance, and investment, such as the financial statements, analyst reports, filling forms, and news articles. We welcome theoretical, experimental, and methodological studies that aim to advance techniques of managing and understanding financial documents, as well as emphasize the applicability in practical applications. The workshop aims to bring together a diverse set of researchers and practitioners interested in investigating relevant topics. Besides, to facilitate developing and testing some relevant techniques, we hold a data challenge on quantifying analyst reports and news articles for the prediction of commodity prices.

Recommender systems have demonstrated great success in information seeking. However, traditional recommender systems work in a static way, estimating user preferences on items from past interaction history. This prevents recommender systems from capturing dynamic and fine-grained preferences of users. Conversational recommender systems bring a revolution to existing recommender systems. They are able to communicate with users through natural languages during which they can explicitly ask whether a user likes an attribute or not. With the preferred attributes, a recommender system can conduct more accurate and personalized recommendations.

Fashion outfit recommendation has attracted increasing attentions from online shopping services and fashion communities.Distinct from other scenarios (e.g., social networking or content sharing) which recommend a single item (e.g., a friend or picture) to a user, outfit recommendation predicts user preference on a set of well-matched fashion items. Hence, performing high-quality personalized outfit recommendation should satisfy two requirements -- 1) the nice compatibility of fashion items and 2) the consistence with user preference. However, present works focus mainly on one of the requirements and only consider either user-outfit or outfit-item relationships, thereby easily leading to suboptimal representations and limiting the performance.

Learning informative representations of users and items from the interaction data is of crucial importance to collaborative filtering (CF). Present embedding functions exploit user-item relationships to enrich the representations, evolving from a single user-item instance to the holistic interaction graph. Nevertheless, they largely model the relationships in a uniform manner, while neglecting the diversity of user intents on adopting the items, which could be to pass time, for interest, or shopping for others like families. Such uniform approach to model user interests easily results in suboptimal representations, failing to model diverse relationships and disentangle user intents in representations.

The rapid growth of user-generated videos on the Internet has intensified the need for text-based video retrieval systems. Traditional methods mainly favor the concept-based paradigm on retrieval with simple queries, which are usually ineffective for complex queries that carry far more complex semantics. Recently, embedding-based paradigm has emerged as a popular approach. It aims to map the queries and videos into a shared embedding space where semantically-similar texts and videos are much closer to each other. Despite its simplicity, it forgoes the exploitation of the syntactic structure of text queries, making it suboptimal to model the complex queries.

Recommender systems are embracing conversational technologies to obtain user preferences dynamically, and to overcome inherent limitations of their static models. A successful Conversational Recommender System (CRS) requires proper handling of interactions between conversation and recommendation. We argue that three fundamental problems need to be solved: 1) what questions to ask regarding item attributes, 2) when to recommend items, and 3) how to adapt to the users' online feedback. To the best of our knowledge, there lacks a unified framework that addresses these problems. In this work, we fill this missing interaction framework gap by proposing a new CRS framework named Estimation"Action" Reflection, or EAR, which consists of three stages to better converse with users. (1) Estimation, which builds predictive models to estimate user preference on both items and item attributes; (2) Action, which learns a dialogue policy to determine whether to ask attributes or recommend items, based on Estimation stage and conversation history; and (3) Reflection, which updates the recommender model when a user rejects the recommendations made by the Action stage. We present two conversation scenarios on binary and enumerated questions, and conduct extensive experiments on two datasets from Yelp and LastFM, for each scenario, respectively. Our experiments demonstrate significant improvements over the state-of-the-art method CRM [32], corresponding to fewer conversation turns and a higher level of recommendation hits.

Recommendation methods construct predictive models to estimate the likelihood of a user-item interaction. Previous models largely follow a general supervised learning paradigm - treating each interaction as a separate data instance and building a supervised learning model upon the information isolated island. Such paradigm, however, overlook relations among data instances, hence easily resulting in suboptimal performance especially for sparse scenarios. Moreover, due to the black-box nature, most models hardly exhibit the reasons behind a prediction, making the recommendation process opaque to understand.

Properly handling missing data is a fundamental challenge in recommendation. Most present works perform negative sampling from unobserved data to supply the training of recommender models with negative signals. Nevertheless, existing negative sampling strategies, either static or adaptive ones, are insufficient to yield high-quality negative samples — both informative to model training and reflective of user real needs.

In the past few years, scene contextual information has been increasingly used for action understanding with promising results. However, unsupervised video action clustering using context has been less explored, and existing clustering methods cannot achieve satisfactory performances. In this paper, we propose a novel unsupervised video action clustering method by using the motion-scene interaction constraint (MSIC). The proposed method takes the unique static scene and dynamic motion characteristics of video action into account, and develops a contextual interaction constraint model under a self-representation subspace clustering framework. First, the complementarity of multi-view subspace representation in each context is explored by single-view and multi-view constraints. Afterward, the context-constrained affinity matrix is calculated and the MSIC is introduced to mutually regularize the disagreement of subspace representation in scene and motion. Finally, by jointly constraining the complementarity of multi-views and the consistency of multi-contexts, an overall objective function is constructed to guarantee the video action clustering result. The experiments on four video benchmark datasets (Weizmann, KTH, UCFsports, and Olympic) demonstrate that the proposed method outperforms the state-of-the-art methods.

With the prevalence of multimedia content on the Web, developing recommender solutions that can effectively leverage the rich signal in multimedia data is in urgent need. Owing to the success of deep neural networks in representation learning, recent advances on multimedia recommendation has largely focused on exploring deep learning methods to improve the recommendation accuracy. To date, however, there has been little effort to investigate the robustness of multimedia representation and its impact on the performance of multimedia recommendation. In this paper, we shed light on the robustness of multimedia recommender system. Using the state-of-the-art recommendation framework and deep image features, we demonstrate that the overall system is not robust, such that a small (but purposeful) perturbation on the input image will severely decrease the recommendation accuracy. This implies the possible weakness of multimedia recommender system in predicting user preference, and more importantly, the potential of improvement by enhancing its robustness. To this end, we propose a novel solution named Adversarial Multimedia Recommendation (AMR), which can lead to a more robust multimedia recommender model by using adversarial learning. The idea is to train the model to defend an adversary, which adds perturbations to the target image with the purpose of decreasing the model's accuracy. We conduct experiments on two representative multimedia recommendation tasks, namely, image recommendation and visually-aware product recommendation. Extensive results verify the positive effect of adversarial learning and demonstrate the effectiveness of our AMR method. Source codes are available in https://github.com/duxy-me/AMR.

Food recommender systems play an important role in assisting users to identify the desired food to eat. Deciding what food to eat is a complex and multi-faceted process, which is influenced by many factors such as the ingredients, appearance of the recipe, the user's personal preference on food, and various contexts like what had been eaten in the past meals. This work formulates the food recommendation problem as predicting user preference on recipes based on three key factors that determine a user's choice on food, namely, 1) the user's (and other users') history; 2) the ingredients of a recipe; and 3) the descriptive image of a recipe. To address this challenging problem, this work develops a dedicated neural network-based solution Hierarchical Attention based Food Recommendation (HAFR) which is capable of: 1) capturing the collaborative filtering effect like what similar users tend to eat; 2) inferring a user's preference at the ingredient level; and 3) learning user preference from the recipe's visual images. To evaluate our proposed method, this work constructs a large-scale dataset consisting of millions of ratings from AllRecipes.com. Extensive experiments show that our method outperforms several competing recommender solutions like Factorization Machine and Visual Bayesian Personalized Ranking with an average improvement of 12%, offering promising results in predicting user preference on food.

Matrix factorization (MF) has been widely used to discover the low-rank structure and to predict the missing entries of data matrix. In many real-world learning systems, the data matrix can be very high dimensional but sparse. This poses an imbalanced learning problem since the scale of missing entries is usually much larger than that of the observed entries, but they cannot be ignored due to the valuable negative signal. For efficiency concern, existing work typically applies a uniform weight on missing entries to allow a fast learning algorithm. However, this simplification will decrease modeling fidelity, resulting in suboptimal performance for downstream applications. In this paper, we weight the missing data nonuniformly, and more generically, we allow any weighting strategy on the missing data. To address the efficiency challenge, we propose a fast learning method, for which the time complexity is determined by the number of observed entries in the data matrix rather than the matrix size. The key idea is twofold: 1) we apply truncated singular value decomposition on the weight matrix to get a more compact representation of the weights and 2) we learn MF parameters with elementwise alternating least squares (eALS) and memorize the key intermediate variables to avoid repeating computations that are unnecessary. We conduct extensive experiments on two recommendation benchmarks, demonstrating the correctness, efficiency, and effectiveness of our fast eALS method.

Online Social Network Sites have become a primary platform for brands and organizations to engage their audience by sharing image and video posts on their timelines. Different from traditional advertising, these posts are not restricted to the products or logo but include visual elements that express more in general the values and attributes of the brand, called brand associations. Since marketers are increasingly spending time in discovering and re-posting user generated posts that reflect the brand attributes, there is an increasing demand for such discovery systems. The goal of these systems is to assist brand experts in filtering through online collections of new user media to discover actionable posts, which match the brand value and have the potential to engage the consumers. Driven by this real-life application, we define and formulate a new task of content discovery for brands and propose a framework that learns to rank posts for brands from their historical timeline. We design a Personalized Content Discovery (PCD) framework to address the three challenges of high inter-brand similarity, sparsity of brand--post interactions, and diversification of timeline. To learn fine-grained brand representation and to generate explanations for the ranking, we automatically learn visual elements of posts from the timeline of brands and from a set of brand attributes in the domain of marketing. To test our framework we use two large-scale Instagram datasets that contain a total of more than 1.5 million image and video posts from the historical timeline of hundreds of brands from multiple verticals such as food and fashion. Extensive experiments indicate that our model can effectively learn fine-grained brand representations and outperform the closest state-of-the-art solutions.

Nowadays, video games play a very important role in human life and no longer purely associated with escapism or entertainment. In fact, gaming has become an essential part of our daily routines, which give rise to the exponential growth of various online game platforms. By participating in such platforms, individuals generate a multitude of game data points, which, for example, can be further used for automatic user profiling and recommendation applications. However, the literature on automatic learning from the game data is relatively sparse, which had inspired us to tackle the problem of player profiling in this first preliminary study. Specifically, in this work, we approach the task of player gender prediction based on various types of game data. Our initial experimental results inspire further research on user profiling in the game domain.

Incorporating knowledge graph into recommender systems has attracted increasing attention in recent years. By exploring the interlinks within a knowledge graph, the connectivity between users and items can be discovered as paths, which provide rich and complementary information to user-item interactions. Such connectivity not only reveals the semantics of entities and relations, but also helps to comprehend a user’s interest. However, existing efforts have not fully explored this connectivity to infer user preferences, especially in terms of modeling the sequential dependencies within and holistic semantics of a path.In this paper, we contribute a new model named Knowledgeaware Path Recurrent Network (KPRN) to exploit knowledge graph for recommendation. KPRN can generate path representations by composing the semantics of both entities and relations. By leveraging the sequential dependencies within a path, we allow effective reasoning on paths to infer the underlying rationale of a user-item interaction. Furthermore, we design a new weighted pooling operation to discriminate the strengths of different paths in connecting a user with an item, endowing our model with a certain level of explainability. We conduct extensive experiments on two datasets about movie and music, demonstrating significant improvements over state-of-the-art solutions Collaborative Knowledge Base Embedding and Neural Factorization Machine.

Identifying mix-and-match relationships between fashion items is an urgent task in a fashion e-commerce recommender system. It will significantly enhance user experience and satisfaction. However, due to the challenges of inferring the rich yet complicated set of compatibility patterns in a large e-commerce corpus of fashion items, this task is still underexplored. Inspired by the recent advances in multirelational knowledge representation learning and deep neural networks, this paper proposes a novel Translation-based Neural Fashion Compatibility Modeling (TransNFCM) framework, which jointly optimizes fashion item embeddings and category-specific complementary relations in a unified space via an end-to-end learning manner. TransNFCM places items in a unified embedding space where a category-specific relation (category-comp-category) is modeled as a vector translation operating on the embeddings of compatible items from the corresponding categories. By this way, we not only capture the specific notion of compatibility conditioned on a specific pair of complementary categories, but also preserve the global notion of compatibility. We also design a deep fashion item encoder which exploits the complementary characteristic of visual and textual features to represent the fashion products. To the best of our knowledge, this is the first work that uses category-specific complementary relations to model the category-aware compatibility between items in a translation-based embedding space. Extensive experiments demonstrate the effectiveness of TransNFCM over the state-of-the-arts on two real-world datasets.

Entity alignment typically suffers from the issues of structural heterogeneity and limited seed alignments.In this paper, we propose a novel Multi-channel Graph Neural Network model (MuGNN) to learn alignment-oriented knowledge graph (KG) embeddings by robustly encoding two KGs via multiple channels.Each channel encodes KGs via different relation weighting schemes with respect to self-attention towards KG completion and cross-KG attention for pruning exclusive entities respectively, which are further combined via pooling techniques.Moreover, we also infer and transfer rule knowledge for completing two KGs consistently.MuGNN is expected to reconcile the structural differences of two KGs, and thus make better use of seed alignments.Extensive experiments on five publicly available datasets demonstrate our superior performance (5% Hits@1 up on average).

10.18653/v1/P19-1128

Recommendation methods construct predictive models to estimate the likelihood of a user-item interaction. Previous models largely follow a general supervised learning paradigm --- treating each interaction as a separate data instance and performing prediction based on the ''information isolated island''. Such methods, however, overlook the relations among data instances, which may result in suboptimal performance especially for sparse scenarios. Moreover, the models built on a separate data instance only can hardly exhibit the reasons behind a recommendation, making the recommendation process opaque to understand. In this tutorial, we revisit the recommendation problem from the perspective of graph learning. Common data sources for recommendation can be organized into graphs, such as user-item interactions (bipartite graphs), social networks, item knowledge graphs (heterogeneous graphs), among others. Such a graph-based organization connects the isolated data instances, bringing benefits to exploiting high-order connectivities that encode meaningful patterns for collaborative filtering, content-based filtering, social influence modeling and knowledge-aware reasoning. Together with the recent success of graph neural networks (GNNs), graph-based models have exhibited the potential to be the technologies for next-generation recommendation systems. This tutorial provides a review on graph-based learning methods for recommendation, with special focus on recent developments of GNNs and knowledge graph-enhanced recommendation. By introducing this emerging and promising topic in this tutorial, we expect the audience to get deep understanding and accurate insight on the spaces, stimulate more ideas and discussions, and promote developments of technologies.

Meta-learning has been proposed as a framework to address the challenging few-shot learning setting. The key idea is to leverage a large number of similar few-shot tasks in order to learn how to adapt a base-learner to a new task for which only a few labeled samples are available. As deep neural networks (DNNs) tend to overfit using a few samples only, meta-learning typically uses shallow neural networks (SNNs), thus limiting its effectiveness. In this paper we propose a novel few-shot learning method called meta-transfer learning (MTL) which learns to adapt a deep NN for few shot learning tasks. Specifically, "meta" refers to training multiple tasks, and "transfer" is achieved by learning scaling and shifting functions of DNN weights for each task. In addition, we introduce the hard task (HT) meta-batch scheme as an effective learning curriculum for MTL. We conduct experiments using (5-class, 1-shot) and (5-class, 5-shot) recognition tasks on two challenging few-shot learning benchmarks: miniImageNet and Fewshot-CIFAR100. Extensive comparisons to related works validate that our meta-transfer learning approach trained with the proposed HT meta-batch scheme achieves top performance. An ablation study also shows that both components contribute to fast convergence and high accuracy.

Yixin Cao, Zikun Hu, Tat-seng Chua, Zhiyuan Liu, Heng Ji. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Wenqiang Lei, Weiwen Xu, Ai Ti Aw, Yuanxin Xiang, Tat Seng Chua. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Chengjiang Li, Yixin Cao, Lei Hou, Jiaxin Shi, Juanzi Li, Tat-Seng Chua. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Most existing recommender systems leverage user behavior data of one type, such as the purchase behavior data in E-commerce. We argue that other types of user behavior data also provide valuable signal, such as views, clicks, and so on. In this work, we contribute a new solution named NMTR (short for Neural Multi-Task Recommendation) for learning recommender systems from user multi-behavior data. In particular, our model accounts for the cascading relationship among different types of behaviors (e.g., a user must click on a product before purchasing it). We perform a joint optimization based on the multi-task learning framework, where the optimization on a behavior is treated as a task. Extensive experiments on the real-world dataset demonstrate that NMTR significantly outperforms state-of-the-art recommender systems that are designed to learn from both single-behavior data and multi-behavior data.

This paper contributes a new machine learning solution for stock movement prediction, which aims to predict whether the price of a stock will be up or down in the near future. The key novelty is that we propose to employ adversarial training to improve the generalization of a neural network prediction model. The rationality of adversarial training here is that the input features to stock prediction are typically based on stock price, which is essentially a stochastic variable and continuously changed with time by nature. As such, normal training with static price-based features (e.g. the close price) can easily overfit the data, being insufficient to obtain reliable models. To address this problem, we propose to add perturbations to simulate the stochasticity of price variable, and train the model to work well under small yet intentional perturbations. Extensive experiments on two real-world stock data show that our method outperforms the state-of-the-art solution [Xu and Cohen, 2018] with 3.11% relative improvements on average w.r.t. accuracy, validating the usefulness of adversarial training for stock prediction task.

To provide more accurate, diverse, and explainable recommendation, it is compulsory to go beyond modeling user-item interactions and take side information into account. Traditional methods like factorization machine (FM) cast it as a supervised learning problem, which assumes each interaction as an independent instance with side information encoded. Due to the overlook of the relations among instances or items (e.g., the director of a movie is also an actor of another movie), these methods are insufficient to distill the collaborative signal from the collective behaviors of users. In this work, we investigate the utility of knowledge graph (KG), which breaks down the independent interaction assumption by linking items with their attributes. We argue that in such a hybrid structure of KG and user-item graph, high-order relations --- which connect two items with one or multiple linked attributes --- are an essential factor for successful recommendation. We propose a new method named Knowledge Graph Attention Network (KGAT) which explicitly models the high-order connectivities in KG in an end-to-end fashion. It recursively propagates the embeddings from a node's neighbors (which can be users, items, or attributes) to refine the node's embedding, and employs an attention mechanism to discriminate the importance of the neighbors. Our KGAT is conceptually advantageous to existing KG-based recommendation methods, which either exploit high-order relations by extracting paths or implicitly modeling them with regularization. Empirical results on three public benchmarks show that KGAT significantly outperforms state-of-the-art methods like Neural FM and RippleNet. Further studies verify the efficacy of embedding propagation for high-order relation modeling and the interpretability benefits brought by the attention mechanism. We release the codes and datasets at https://github.com/xiangwang1223/knowledge_graph_attention_network.

AI as a concept has been around since the 1950's. With the recent advancements in machine learning technology, and the availability of big data and large computing processing power, the scene is set for AI to be used in many more systems and applications which will profoundly impact society. The current deep learning based AI systems are mostly in black box form and are often non-explainable. Though it has high performance, it is also known to make occasional fatal mistakes. This has limited the applications of AI, especially in mission critical problems such as decision support, command and control, and other life-critical operations. This talk focuses on explainable AI, which holds promise in helping humans to better understand and interpret the decisions made by black-box AI models. Current research efforts towards explainable multimedia AI center on two parts of solution. The first part focuses on better understanding of multimedia content, especially video. This includes dense annotation of video content from not just object recognition, but also relation inference. The relation includes both correlation and causality relations, as well as common sense knowledge. The dense annotation enables us to transform the level of representation of video towards that of language, in the form of relation triplets and relation graphs, and permits in-depth research on flexible descriptions, question-answering and knowledge inference of video content. A large scale video dataset has been created to support this line of research. The second direction focuses on the development of explainable AI models, which are just beginning. Existing works focus on either the intrinsic approach, which designs self-explanatory models, or post-hoc approach, which constructs a second model to interpret the target model. Both approaches have limitations on trade-offs between interpretability and accuracy, and the lack of guarantees about the explanation quality. In addition, there are issues of quality, fairness, robustness and privacy in model interpretation. In this talk, I present current state-of-the arts approaches in explainable multimedia AI, along with our preliminary research on relation inference in videos, as well as leveraging prior domain knowledge, information theoretic principles, and adversarial algorithms to achieving interpretability. I will also discuss future research towards quality, fairness and robustness of interpretable AI.

Understanding the objects and relations between them is indispensable to fine-grained video content analysis, which is widely studied in recent research works in multimedia and computer vision. However, existing works are limited to evaluating with either small datasets or indirect metrics, such as the performance over images. The underlying reason is that the construction of a large-scale video dataset with dense annotation is tricky and costly. In this paper, we address several main issues in annotating objects and relations in user-generated videos, and propose an annotation pipeline that can be executed at a modest cost. As a result, we present a new dataset, named VidOR, consisting of 10k videos (84 hours) together with dense annotations that localize 80 categories of objects and 50 categories of predicates in each video. We have made the training and validation set public and extendable for more tasks to facilitate future research on video object and relation recognition.

Restaurant dishes represent a significant portion of food that people consume in their daily life. While people are becoming health-conscious in their food intake, convenient restaurant food tracking becomes an essential task in wellness and fitness applications. Given the huge number of dishes (food categories) involved, it becomes extremely challenging for traditional food photo classification to be feasible in both algorithm design and training data availability. In this work, we present a demo that runs on restaurant dish images in a city of millions of residents and tens of thousand restaurants. We propose a rank-loss based convolutional neural network to optimize the image features representation. Context information such as GPS location of the recognition request is also used to further improve the performance. Our experimental results are highly promising. We have shown in our demo that the proposed algorithm is near ready to be deployed in real-world applications.

Mixed dish is a food category that contains different dishes mixed in one plate, and is popular in Eastern and Southeast Asia. Recognizing individual dishes in a mixed dish image is important for health related applications, e.g. calculating the nutrition values. However, most existing methods that focus on single dish classification are not applicable to mixed-dish recognition. The new challenge in recognizing mixed-dish images are the complex ingredient combination and severe overlap among different dishes. In order to tackle these problems, we propose a novel approach called contextual relation networks (CR-Nets) that encodes the implicit and explicit contextual relations among multiple dishes using region-level features and label-level co-occurrence, respectively. This is inspired by the intuition that people are likely to choose dishes with common eating habits, e.g., with multiple nutrition but without repeating ingredients. In addition, we collect a large-scale dataset of mixed-dish images that contain $9,254$ mixed-dish images from $6$ school canteens in Singapore. Extensive experiments on both our dataset and a smaller-scale public dataset validate that our CR-Nets can achieve top performance for localizing the dishes and recognizing their food categories.

Recent years have seen unprecedented research on using artificial intelligence to understand the subjective attributes of images and videos. These attributes are not objective properties of the content but are highly dependent on the perception of the viewers. Subjective attributes are extremely valuable in many applications where images are tailored to the needs of a large group, which consists of many individuals with inherently different ideas and preferences. For instance, marketing experts choose images to establish specific associations in the consumers' minds, while psychologists look for pictures with adequate emotions for therapy. Unfortunately, most of the existing frameworks either focus on objective attributes or rely on large scale datasets of annotated images, making them costly and unable to clearly measure multiple interpretations of a single input. Meanwhile, we can see that users or organizations often interact with images in a multitude of real-life applications, such as the sharing of photographs by brands on social media or the re-posting of image microblogs by users. We argue that these aggregated interactions can serve as auxiliary information to infer image interpretations. To this end, we propose a probabilistic learning framework capable of transferring such subjective information to the image-level labels based on a known aggregated distribution. We use our framework to rank images by subjective attributes from the domain knowledge of social media marketing and personality psychology. Extensive studies and visualizations show that using auxiliary information is a viable line of research for the multimedia community to perform subjective attributes prediction.

Fashion knowledge plays a pivotal role in helping people in their dressing. In this paper, we present a novel system to automatically harvest fashion knowledge from social media. It unifies three tasks of occasion, person and clothing discovery from multiple modalities of images, texts and metadata. A contextualized fashion concept learning model is applied to leverage the rich contextual information for improving the fashion concept learning performance. At the same time, to counter the label noise within training data, we employ a weak label modeling method to further boost the performance. We build a website to demonstrate the quality of fashion knowledge extracted by our system.

Fashion knowledge helps people to dress properly and addresses not only physiological needs of users, but also the demands of social activities and conventions. It usually involves three mutually related aspects of: occasion, person and clothing. However, there are few works focusing on extracting such knowledge, which will greatly benefit many downstream applications, such as fashion recommendation. In this paper, we propose a novel method to automatically harvest fashion knowledge from social media. We unify three tasks of occasion, person and clothing discovery from multiple modalities of images, texts and metadata. For person detection and analysis, we use the off-the-shelf tools due to their flexibility and satisfactory performance. For clothing recognition and occasion prediction, we unify the two tasks by using a contextualized fashion concept learning module, which captures the dependencies and correlations among different fashion concepts. To alleviate the heavy burden of human annotations, we introduce a weak label modeling module which can effectively exploit machine-labeled data, a complementary of clean data. In experiments, we contribute a benchmark dataset and conduct extensive experiments from both quantitative and qualitative perspectives. The results demonstrate the effectiveness of our model in fashion concept prediction, and the usefulness of extracted knowledge with comprehensive analysis.

Food recognition for user-uploaded images is crucial in visual diet tracking, an emerging application linking multimedia and healthcare domains. However, it is challenging due to the various visual appearances of food images. This is caused by different conditions when taking the photos, such as angles, distances, light conditions, food containers, and background scenes. To alleviate such a semantic gap, this paper presents a cross-modal alignment and transfer network (ATNet), which is motivated by the paradigm of learning using privileged information (LUPI). It additionally utilizes the ingredients in food images as an "intelligent teacher" in the training stage to facilitate cross-modal information passing. Specifically, ATNet first uses a pair of synchronized autoencoders to build the base image and ingredient channels for information flow. Subsequently, the information passing is enabled through a two-stage cross-modal interaction. The first stage of interaction adopts a two-step method, called partial heterogeneous transfer, to 1) alleviate the intrinsic heterogeneity between images and ingredients and 2) align them in a shared space to make their carried information about food classes interact. In the second stage, ATNet learns to map the visual embeddings of images to the ingredient channel for food recognition from the view of "teacher''. This leads a refined recognition by a multi-view fusion. Experiments on two real-world datasets show that ATNet can be incorporated with any state-of-the-art CNN models to consistently improve their performance.

ACM Multimedia 2019 Video Relation Understanding Challenge is the first grand challenge aiming at pushing video content analysis at the relational and structural level. This year, the challenge asks the participants to explore and develop innovative algorithms to detect object entities and their relations based on a large-scale user-generated video dataset. The tasks will advance the foundation of future visual systems that are able to perform complex inferences. This paper presents an overview of the grand challenge, including background, detailed descriptions of the three proposed tasks, the corresponding datasets for training, validation and testing, and the evaluation process.

Multimedia research has now moved beyond laboratory experiments and is rapidly being deployed in real-life applications including advertisements, social interaction, search, security, automated driving, and healthcare. Hence, the developed algorithms now have a direct impact on the individuals using the abovementioned services and the society as a whole. While there is a huge potential to benefit the society using such technologies, there is also an urgent need to identify the checks and balances to ensure that the impact of such technologies is ethical and positive. This panel will bring together an array of experts who have experience collecting large-scale datasets, building multimedia algorithms, and deploying them in practical applications, as well as, a lawyer whose eyes have been on the fundamental rights at stake. They will lead a discussion on the ethics and lawfulness of dataset creation, licensing, privacy of individuals represented in the datasets, algorithmic transparency, algorithmic bias, explainability, and the implications of application deployment. Through an interactive process engaging the audience, the panel hopes to: increase the awareness of such concepts in the multimedia research community; initiate a discussion on community guidelines all for setting the future direction of conducting multimedia research in a lawful and ethical manner.

Personalized recommendation plays a central role in many online content sharing platforms. To provide quality micro-video recommendation service, it is of crucial importance to consider the interactions between users and items (i.e. micro-videos) as well as the item contents from various modalities (e.g. visual, acoustic, and textual). Existing works on multimedia recommendation largely exploit multi-modal contents to enrich item representations, while less effort is made to leverage information interchange between users and items to enhance user representations and further capture user's fine-grained preferences on different modalities. In this paper, we propose to exploit user-item interactions to guide the representation learning in each modality, and further personalized micro-video recommendation. We design a Multi-modal Graph Convolution Network (MMGCN) framework built upon the message-passing idea of graph neural networks, which can yield modal-specific representations of users and micro-videos to better capture user preferences. Specifically, we construct a user-item bipartite graph in each modality, and enrich the representation of each node with the topological structure and features of its neighbors. Through extensive experiments on three publicly available datasets, Tiktok, Kwai, and MovieLens, we demonstrate that our proposed model is able to significantly outperform state-of-the-art multi-modal recommendation methods.

Few-shot classification (FSC) is challenging due to the scarcity of labeled training data (e.g. only one labeled data point per class). Meta-learning has shown to achieve promising results by learning to initialize a classification model for FSC. In this paper we propose a novel semi-supervised meta-learning method called learning to self-train (LST) that leverages unlabeled data and specifically meta-learns how to cherry-pick and label such unsupervised data to further improve performance. To this end, we train the LST model through a large number of semi-supervised few-shot tasks. On each task, we train a few-shot model to predict pseudo labels for unlabeled data, and then iterate the self-training steps on labeled and pseudo-labeled data with each step followed by fine-tuning. We additionally learn a soft weighting network (SWN) to optimize the self-training weights of pseudo labels so that better ones can contribute more to gradient descent optimization. We evaluate our LST method on two ImageNet benchmarks for semi-supervised few-shot classification and achieve large improvements over the state-of-the-art.

Learning vector representations (aka. embeddings) of users and items lies at the core of modern recommender systems. Ranging from early matrix factorization to recently emerged deep learning based methods, existing efforts typically obtain a user's (or an item's) embedding by mapping from pre-existing features that describe the user (or the item), such as ID and attributes. We argue that an inherent drawback of such methods is that, the collaborative signal, which is latent in user-item interactions, is not encoded in the embedding process. As such, the resultant embeddings may not be sufficient to capture the collaborative filtering effect.

Understanding the mix-and-match relationships of fashion items receives increasing attention in fashion industry. Existing methods have primarily utilized the visual content to learn the visual compatibility and performed matching in a latent space. Despite their effectiveness, these methods work like a black box and cannot reveal the reasons that two items match well. The rich attributes associated with fashion items, e.g.,off-shoulder dress and black skinny jean, which describe the semantics of items in a human-interpretable way, have largely been ignored.