Source-free domain adaptation (SFDA) involves training a model on source domain and then applying it to a related target domain without access to the source data and labels during adaptation. The complexity of scene information and lack of the source domain make SFDA a difficult task. Recent studies have shown promising results, but many approaches to domain adaptation concentrate on domain shift and neglect the effects of negative transfer, which may impede enhancements of model performance during adaptation. In this paper, addressing this issue, we propose a novel framework of Attention Residual Fusion Network (ARFNet) based on contrast learning for SFDA to alleviate negative transfer and domain shift during the progress of adaptation, in which attention residual fusion, global-local attention contrast, and dynamic centroid evaluation are exploited. Concretely, the attention mechanism is first exploited to capture the discriminative region of the target object. Then, in each block, attention features are decomposed into spatial-wise and channel-wise attentions. The spatial-wise attentions are aggregated with original semantic features to achieve the cross-layer attention residual fusion progressively while the channel-wise attentions are exploited for self-distillation. During adaptation progress, we contrast global and local representations to improve the perceptual capabilities of different categories, which enables the model to discriminate variations between inner-class and intra-class. Finally, a dynamic centroid evaluation strategy is exploited to evaluate the trustworthy centroids and labels for self-supervised self-distillation, which aims to accurately approximate the center of the source domain and pseudo-labels to mitigate domain shift. To validate the efficacy of our methods, we execute comprehensive experiments on five benchmarks of varying scales, i.e., Office-31, Office-Home, VisDA-C, DomainNet-126, Cub-Paintings. Experimental outcomes indicate that our method surpasses other techniques, attaining superior performance across SFDA benchmarks.Code is available at https://github.com/RoryShao/ARFNet.git.

For recommender systems in internet platforms, search activities provide additional insights into user interest through query-click interactions with items, and are thus widely used for enhancing personalized recommendation. However, these interacted items have not only transferable features that match users’ interests and are beneficial to the recommendation domain, but also have features related to users’ unique intents in the search domain. Such a domain gap of item features is neglected by most current search-enhanced recommendation methods. They directly incorporate these search behaviors into recommendation, and thus introduce partial negative transfer. Tackling this problem is challenging due to the lack of explicit supervision signals to disentangle features matching search-specific intent or general interest. To address this, we propose ClardRec, a c ounterfactual l e a rning-driven r epresentation d isentanglement framework for search-enhanced recommendation, based on the common belief that a user would click an item under a query not solely because of the item-query match but also due to the item’s query-independent general features (e.g., color or style) that interest the user. These general features exclude the reflection of search-specific intents contained in queries, ensuring a pure match to users’ underlying interests to complement recommendation. We perform the disentanglement based on a counterfactual thinking idea, how would user preferences and query match change for items if we removed their query-related features in search. Specifically, we leverage search queries to construct counterfactual signals to disentangle item representations, isolating only query-independent general features. These representations subsequently enable feature augmentation and data augmentation for the recommendation scenario. Comprehensive experiments on real datasets demonstrate that ClardRec is effective in both collaborative filtering and sequential recommendation scenarios. The source code is available at https://github.com/JJCui96/ClardRec .

Explainable recommender systems are designed to elucidate the explanation behind each recommendation, enabling users to comprehend the underlying logic. Previous works perform rating prediction and explanation generation in a multi-task manner. However, these works suffer from incoherence between predicted ratings and explanations. To address the issue, we propose a novel framework that employs a large language model (LLM) to generate a rating, transforms it into a rating vector, and finally generates an explanation based on the rating vector and user-item information. Moreover, we propose utilizing publicly available LLMs and pre-trained sentiment analysis models to automatically evaluate the coherence without human annotations. Extensive experimental results on three datasets of explainable recommendation show that the proposed framework is effective, outperforming state-of-the-art baselines with improvements of 7.3% in explainability and 4.4% in text quality.

While the mining of modalities is the focus of most multimodal recommendation methods, we believe that how to fully utilize both collaborative and multimodal information is pivotal in e-commerce scenarios where, as clarified in this work, the user behaviors are rarely determined entirely by multimodal features. In order to combine the two distinct types of information, some additional challenges are encountered: 1) Modality erasure: Vanilla graph convolution, which proves rather useful in collaborative filtering, however erases multimodal information; 2) Modality forgetting: Multimodal information tends to be gradually forgotten as the recommendation loss essentially facilitates the learning of collaborative information. To this end, we propose a novel approach named STAIR, which employs a novel stepwise graph convolution to enable a co-existence of collaborative and multimodal information in e-commerce recommendation. Besides, it starts with the raw multimodal features as an initialization, and the forgetting problem can be significantly alleviated through constrained embedding updates. As a result, STAIR achieves state-of-the-art recommendation performance on three public e-commerce datasets with minimal computational and memory costs.

Xiaodong Wu, Minhao Wang, Yichen Liu, Xiaoming Shi, He Yan, Lu Xiangju, Junmin Zhu, Wei Zhang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

The drug recommendation task aims to provide effective and safe prescription decision support for clinical treatment based on patients' past Electronic Health Records (EHR). However, the prevalent phenomenon of missing views in multi-source heterogeneous EHR data may cause performance degradation. This is due to the lack of sufficient information and increased learning difficulties, which limit the practical effectiveness of drug recommendation models in medical applications. In this paper, we emphasize the problems of incompleteness in practical drug recommendation and propose the Incomplete Multi-View Drug Recommendation model via Multi-Level Representation Learning and Curriculum Learning named IMDR. In particular, IMDR employs a Multi-Level Representation Learning architecture equipped with a Medical Code-Level Drug Knowledge Infusion Module and a Visit-Level Cross-View Information Module for patient representation learning to overcome the information loss caused by incomplete data. And then, a Gaussian-guided curriculum learning strategy is proposed to assist the learning process of IMDR with a novel difficulty measure to achieve effective progressive learning under missing medical views. Systematic evaluation on two large-scale real-world medical datasets, MIMIC-III and MIMIC-IV, demonstrates that IMDR reduces the Drug-Drug Interaction (DDI) rate by 2.97% compared to existing state-of-the-art drug recommendation baselines, while achieving significant improvements of 3.29% and 1.97% in Jaccard similarity scores and F1 score, respectively. Furthermore, compared to advanced incomplete multi-view learning (IML) models, IMDR's advantages in Jaccard similarity scores and F1 score further expand to 4.03% and 2.41%.

Graph Neural Networks (GNNs) have demonstrated their superiority in collaborative filtering, where the user-item (U-I) interaction bipartite graph serves as the fundamental data format. However, when graph-structured side information (e.g., multimodal similarity graphs or social networks) is integrated into the U-I bipartite graph, existing graph collaborative filtering methods fall short of achieving satisfactory performance. We quantitatively analyze this problem from a spectral perspective. Recall that a bipartite graph possesses a full spectrum within the range of [-1, 1], with the highest frequency exactly achievable at -1 and the lowest frequency at 1; however, we observe as more side information is incorporated, the highest frequency of the augmented adjacency matrix progressively shifts rightward. This spectrum shift phenomenon has caused previous approaches built for the full spectrum [-1, 1] to assign mismatched importance to different frequencies. To this end, we propose Spectrum Shift Correction (dubbed SSC), incorporating shifting and scaling factors to enable spectral GNNs to adapt to the shifted spectrum. Unlike previous paradigms of leveraging side information, which necessitate tailored designs for diverse data types, SSC directly connects traditional graph collaborative filtering with any graph-structured side information. Experiments on social and multimodal recommendation demonstrate the effectiveness of SSC, achieving relative improvements of up to 23% without incurring any additional computational overhead. Our code is available at https://github.com/yhhe2004/SSC-KDD.

In this paper, we analyze the feature-based knowledge distillation for recommendation from the frequency perspective. By defining knowledge as different frequency components of the features, we theoretically demonstrate that regular feature-based knowledge distillation is equivalent to equally minimizing losses on all knowledge and further analyze how this equal loss weight allocation method leads to important knowledge being overlooked. In light of this, we propose to emphasize important knowledge by redistributing knowledge weights. Furthermore, we propose FreqD, a lightweight knowledge reweighting method, to avoid the computational cost of calculating losses on each knowledge. Extensive experiments demonstrate that FreqD consistently and significantly outperforms state-of-the-art knowledge distillation methods for recommender systems. Our code is available at https://github.com/woriazzc/KDs.

Source-free domain adaptation (SFDA) aims to address the challenge of adapting to a target domain without accessing the source domain directly. However, due to the inaccessibility of source domain data, deterministic invariable features cannot be obtained. Current mainstream methods primarily focus on evaluating invariant features in the target domain that closely resemble those in the source domain, subsequently aligning the target domain with the source domain. However, these methods are susceptible to hard samples and influenced by domain bias. In this paper, we propose a Consistent Assistant Domains Transformer for SFDA, abbreviated as CADTrans, which solves the issue by constructing invariable feature representations of domain consistency. Concretely, we develop an assistant domain module for CADTrans to obtain diversified representations from the intermediate aggregated global attentions, which addresses the limitation of existing methods in adequately representing diversity. Based on assistant and target domains, invariable feature representations are obtained by multiple consistent strategies, which can be used to distinguish easy and hard samples. Finally, to align the hard samples to the corresponding easy samples, we construct a conditional multi-kernel max mean discrepancy (CMK-MMD) strategy to distinguish between samples of the same category and those of different categories. Extensive experiments are conducted on various benchmarks such as Office-31, Office-Home, VISDA-C, and DomainNet-126, proving the significant performance improvements achieved by our proposed approaches. Code is available at https://github.com/RoryShao/CADTrans.git.

This paper explores explaining session-based recommendation (SR) by path reasoning. Current SR models emphasize accuracy but lack explainability, while traditional path reasoning prioritizes knowledge graph exploration, ignoring sequential patterns present in the session history. Therefore, we propose a generalized hierarchical reinforcement learning framework for SR, which improves the explainability of existing SR models via Path Reasoning, namely PR4SR. Considering the different importance of items to the session, we design the session-level agent to select the items in the session as the starting nodes for path reasoning and the path-level agent to perform path reasoning. In particular, we design a multi-target reward mechanism to adapt to the skip behaviors of sequential patterns in SR and introduce path midpoint reward to enhance the exploration efficiency and accuracy in knowledge graphs. To improve the knowledge graph’s completeness and diversify the paths of explanation, we incorporate extracted feature information from images into the knowledge graph. We instantiate PR4SR in five state-of-the-art SR models (i.e., GRU4REC, NARM, GCSAN, SR-GNN, SASRec) and compare it with other explainable SR frameworks to demonstrate the effectiveness of PR4SR for recommendation and explanation tasks through extensive experiments with these approaches on four datasets.

A transparent decision-making process is essential for developing reliable and trustworthy recommender systems. For sequential recommendation, it means that the model can identify key items that account for its recommendation results. However, achieving both interpretability and recommendation performance simultaneously is challenging, especially for models that take the entire sequence of items as input without screening. In this paper, we propose an interpretable framework (named PTSR) that enables a pattern-wise transparent decision-making process without extra features. It breaks the sequence of items into multi-level patterns that serve as atomic units throughout the recommendation process. The contribution of each pattern to the outcome is quantified in the probability space. With a carefully designed score correction mechanism, the pattern contribution can be implicitly learned in the absence of ground-truth key patterns. The final recommended items are those that most key patterns strongly endorse. Extensive experiments on five public datasets demonstrate remarkable recommendation performance, while statistical analysis and case studies validate the model interpretability.

Feature-based knowledge distillation has been applied to compress modern recommendation models, usually with projectors that align student (small) recommendation models' dimensions with teacher dimensions. However, existing studies have only focused on making the projected features (i.e., student features after projectors) similar to teacher features, overlooking investigating whether the user preference can be transferred to student features (i.e., student features before projectors) in this manner. In this paper, we find that due to the lack of restrictions on projectors, the process of transferring user preferences will likely be interfered with. We refer to this phenomenon as preference inconsistency. It greatly wastes the power of feature-based knowledge distillation. To mitigate preference inconsistency, we propose PCKD, which consists of two regularization terms for projectors. We also propose a hybrid method that combines the two regularization terms. We focus on items with high preference scores and significantly mitigate preference inconsistency, improving the performance of feature-based knowledge distillation. Extensive experiments on three public datasets and three backbones demonstrate the effectiveness of PCKD.

The adversarial robustness of a neural network mainly relies on two factors: model capacity and antiperturbation ability. In this article, we study the antiperturbation ability of the network from the feature maps of convolutional layers. Our theoretical analysis discovers that larger convolutional feature maps before average pooling can contribute to better resistance to perturbations, but the conclusion is not true for max pooling. It brings new inspiration to the design of robust neural networks and urges us to apply these findings to improve existing architectures. The proposed modifications are very simple and only require upsampling the inputs or slightly modifying the stride configurations of downsampling operators. We verify our approaches on several benchmark neural network architectures, including AlexNet, VGG, RestNet18, and PreActResNet18. Nontrivial improvements in terms of both natural accuracy and adversarial robustness can be achieved under various attack and defense mechanisms. The code is available at https://github.com/MTandHJ/rcm.

Knowledge Tracing (KT) is a crucial task in computer-aided education and intelligent tutoring systems, predicting students’ performance on new questions from their responses to prior ones. An accurate KT model can capture a student’s mastery level of different knowledge topics, as reflected in their predicted performance on different questions. This helps improve the learning efficiency by suggesting appropriate new questions that complement students’ knowledge states. However, current KT models have significant drawbacks that they neglect the imbalanced discrimination of historical responses. A significant proportion of question responses provide limited information for discerning students’ knowledge mastery, such as those that demonstrate uniform performance across different students. Optimizing the prediction of these cases may increase overall KT accuracy, but also negatively impact the model’s ability to trace personalized knowledge states, especially causing a deceptive surge of performance. Towards this end, we propose a framework to reweight the contribution of different responses based on their discrimination in training. Additionally, we introduce an adaptive predictive score fusion technique to maintain accuracy on less discriminative responses, achieving proper balance between student knowledge mastery and question difficulty. Experimental results demonstrate that our framework enhances the performance of three mainstream KT methods on three widely used datasets.

Cognitive diagnosis assessment is a fundamental and crucial task for student learning. It models the student-exercise interaction, and discovers the students' proficiency levels on each knowledge attribute. In real-world intelligent education systems, generalization and interpretability of cognitive diagnosis methods are of equal importance. However, most existing methods can hardly make the best of both worlds due to the complicated student-exercise interaction. To this end, this paper proposes a symbolic cognitive diagnosis~(SCD) framework to simultaneously enhance generalization and interpretability. The SCD framework incorporates the symbolic tree to explicably represent the complicated student-exercise interaction function, and utilizes gradient-based optimization methods to effectively learn the student and exercise parameters. Meanwhile, the accompanying challenge is that we need to tunnel the discrete symbolic representation and continuous parameter optimization. To address this challenge, we propose to hybridly optimize the representation and parameters in an alternating manner. To fulfill SCD, it alternately learns the symbolic tree by derivative-free genetic programming and learns the student and exercise parameters via gradient-based Adam. The extensive experimental results on various real-world datasets show the superiority of SCD on both generalization and interpretability. The ablation study verifies the efficacy of each ingredient in SCD, and the case study explicitly showcases how the interpretable ability of SCD works.

The key challenge in semantic search is to create models that are both accurate and efficient in pinpointing relevant sentences for queries.While BERT-style bi-encoders excel in efficiency with pre-computed embeddings, they often miss subtle nuances in search tasks.Conversely, GPT-style LLMs with crossencoder designs capture these nuances but are computationally intensive, hindering realtime applications.In this paper, we present D2LLMs-Decomposed and Distilled LLMs for semantic search-that combines the best of both worlds.We decompose a cross-encoder into an efficient bi-encoder integrated with Pooling by Multihead Attention and an Interaction Emulation Module, achieving nuanced understanding and pre-computability.Knowledge from the LLM is distilled into this model using contrastive, rank, and feature imitation techniques.Our experiments show that D2LLM surpasses five leading baselines in terms of all metrics across three tasks, particularly improving NLI task performance by at least 6.45%.

Long-term time series forecasting (LTSF) has been widely applied in finance, traffic prediction, and other domains. Recently, patch-based transformers have emerged as a promising approach, segmenting data into sub-level patches that serve as input tokens. However, existing methods mostly rely on predetermined patch lengths, necessitating expert knowledge and posing challenges in capturing diverse characteristics across various scales. Moreover, time series data exhibit diverse variations and fluctuations across different temporal scales, which traditional approaches struggle to model effectively. In this paper, we propose a dynamic tokenizer with a dynamic sparse learning algorithm to capture diverse receptive fields and sparse patterns of time series data. In order to build hierarchical receptive fields, we develop a multi-scale Transformer model, coupled with multi-scale sequence extraction, capable of capturing multi-resolution features. Additionally, we introduce a group-aware rotary position encoding technique to enhance intra- and inter-group position awareness among representations across different temporal scales. Our proposed model, named DRFormer, is evaluated on various real-world datasets, and experimental results demonstrate its superiority compared to existing methods. Our code is available at: https://github.com/ruixindingECNU/DRFormer.

Numerous self-supervised learning paradigms, such as contrastive learning and masked image modeling, have been proposed to acquire powerful and general representations from unlabeled data. However, these models are commonly pretrained within their specific framework alone, failing to consider the complementary nature of visual representations. To tackle this issue, we introduce Comprehensive Distillation with Multiple Self-supervised Teachers (DMT) for pretrained model compression, which leverages the strengths of multiple off-the-shelf self-supervised models. Our experimental results on prominent benchmark datasets exhibit that the proposed method significantly surpasses state-of-the-art competitors while retaining favorable efficiency metrics. On classification tasks, our DMT framework utilizing three different self-supervised ViT-Base teachers enhances the performance of both small/tiny models and the base model itself. For dense tasks, DMT elevates the AP/mIoU of standard SSL models on MS-COCO and ADE20K datasets by 4.0%.

Knowledge tracing (KT) plays a crucial role in computer-aided education and intelligent tutoring systems, aiming to assess students' knowledge proficiency by predicting their future performance on new questions based on their past response records. While existing deep learning knowledge tracing (DLKT) methods have significantly improved prediction accuracy and achieved state-of-the-art results, they often suffer from a lack of interpretability. To address this limitation, current approaches have explored incorporating psychological influences to achieve more explainable predictions, but they tend to overlook the potential influences of historical responses. In fact, understanding how models make predictions based on response influences can enhance the transparency and trustworthiness of the knowledge tracing process, presenting an opportunity for a new paradigm of interpretable KT. However, measuring unobservable response influences is challenging. In this paper, we resort to counterfactual reasoning that intervenes in each response to answer what if a student had answered a question incorrectly that he/she actually answered correctly, and vice versa. Based on this, we propose RCKT, a novel response influence-based counterfactual knowledge tracing framework. RCKT generates response influences by comparing prediction outcomes from factual sequences and constructed counterfactual sequences after interventions. Additionally, we introduce maximization and inference techniques to leverage accumulated influences from different past responses, further improving the model's performance and credibility. Extensive experimental results demonstrate that our RCKT method outperforms state-of-the-art knowledge tracing methods on four datasets against six baselines, and provides credible interpretations of response influences. The source code is available at https://github.com/JJCui96IRCKT.

Knowledge tracing (KT) is a crucial task in intelligent education, focusing on predicting students' performance on given questions to trace their evolving knowledge. The advancement of deep learning in this field has led to deep-learning knowledge tracing (DLKT) models that prioritize high predictive accuracy. However, many existing DLKT methods overlook the fundamental goal of tracking students' dynamical knowledge mastery. These models do not explicitly model knowledge mastery tracing processes or yield unreasonable results that educators find difficulty to comprehend and apply in real teaching scenarios. In response, our research conducts a preliminary analysis of mainstream KT approaches to highlight and explain such unreasonableness. We introduce GRKT, a graph-based reasonable knowledge tracing method to address these issues. By leveraging graph neural networks, our approach delves into the mutual influences of knowledge concepts, offering a more accurate representation of how the knowledge mastery evolves throughout the learning process. Additionally, we propose a fine-grained and psychological three-stage modeling process as knowledge retrieval, memory strengthening, and knowledge learning/forgetting, to conduct a more reasonable knowledge tracing process. Comprehensive experiments demonstrate that GRKT outperforms eleven baselines across three datasets, not only enhancing predictive accuracy but also generating more reasonable knowledge tracing results. This makes our model a promising advancement for practical implementation in educational settings. The source code is available at https://github.com/JJCui96/GRKT.

Cognitive diagnosis is a vital upstream task in intelligent education systems. It models the student-exercise interaction, aiming to infer the students' proficiency levels on each knowledge concept. This paper observes that most existing methods can hardly effectively capture the homogeneous influence due to its inherent complexity. That is to say, although students exhibit similar performance on given exercises, their proficiency levels inferred by these methods vary significantly, resulting in shortcomings in interpretability and efficacy. Given the complexity of homogeneous influence, a hypergraph could be a choice due to its flexibility and capability of modeling high-order similarity which aligns with the nature of homogeneous influence. However, before incorporating hypergraph, one at first needs to address the challenges of distorted homogeneous influence, sparsity of response logs, and over-smoothing. To this end, this paper proposes a hypergraph cognitive diagnosis model (HyperCDM) to address these challenges and effectively capture the homogeneous influence. Specifically, to avoid distortion, HyperCDM employs a divide-and-conquer strategy to learn student, exercise and knowledge representations in their own hypergraphs respectively, and interconnects them via a feature-based interaction function. To construct hypergraphs based on sparse response logs, the auto-encoder is utilized to preprocess response logs and K-means is applied to cluster students. To mitigate over-smoothing, momentum hypergraph convolution networks are designed to partially keep previous representations during the message propagation. Extensive experiments on both offline and online real-world datasets show that HyperCDM achieves state-of-the-art performance in terms of interpretability and capturing homogeneous influence effectively, and is competitive in generalization. The ablation study verifies the efficacy of each component, and the case study explicitly showcases the homogeneous influence captured by HyperCDM.

Embedding plays a key role in modern recommender systems because they are virtual representations of real-world entities and the foundation for subsequent decision-making models. In this paper, we propose a novel embedding update mechanism, Structure-aware Embedding Evolution (SEvo for short), to encourage related nodes to evolve similarly at each step. Unlike GNN (Graph Neural Network) that typically serves as an intermediate module, SEvo is able to directly inject graph structural information into embedding with minimal computational overhead during training. The convergence properties of SEvo along with its potential variants are theoretically analyzed to justify the validity of the designs. Moreover, SEvo can be seamlessly integrated into existing optimizers for state-of-the-art performance. Particularly SEvo-enhanced AdamW with moment estimate correction demonstrates consistent improvements across a spectrum of models and datasets, suggesting a novel technical route to effectively utilize graph structural information beyond explicit GNN modules.

Vision transformers (ViT) usually extract features via forwarding all the tokens in the self-attention layers from top to toe. In this paper, we introduce dynamic token-pass vision transformers (DoViT) for semantic segmentation, which can adaptively reduce the inference cost for images with different complexity. DoViT gradually stops partial easy tokens from self-attention calculation and keeps the hard tokens forwarding until meeting the stopping criteria. We employ lightweight auxiliary heads to make the token-pass decision and divide the tokens into keeping/stopping parts. With a token separate calculation, the self-attention layers are speeded up with sparse tokens and still work friendly with hardware. A token reconstruction module is built to collect and reset the grouped tokens to their original position in the sequence, which is necessary to predict correct semantic masks. We conduct extensive experiments on two common semantic segmentation tasks, and demonstrate that our method greatly reduces about 40% ∼ 60% FLOPs and the drop of mIoU is within 0.8% for various segmentation transformers. The throughput and inference speed of ViT-L/B are increased to more than 2× on Cityscapes. Code is available at https://github.com/FLHonker/DoViT-code.

Rule-based models, e.g., decision trees, are widely used in scenarios demanding high model interpretability for their transparent inner structures and good model expressivity. However, rule-based models are hard to optimize, especially on large data sets, due to their discrete parameters and structures. Ensemble methods and fuzzy/soft rules are commonly used to improve performance, but they sacrifice the model interpretability. To obtain both good scalability and interpretability, we propose a new classifier, named Rule-based Representation Learner (RRL), that automatically learns interpretable non-fuzzy rules for data representation and classification. To train the non-differentiable RRL effectively, we project it to a continuous space and propose a novel training method, called Gradient Grafting, that can directly optimize the discrete model using gradient descent. A novel design of logical activation functions is also devised to increase the scalability of RRL and enable it to discretize the continuous features end-to-end. Exhaustive experiments on ten small and four large data sets show that RRL outperforms the competitive interpretable approaches and can be easily adjusted to obtain a trade-off between classification accuracy and model complexity for different scenarios.

Sequential recommendation is a critical but challenging task in capturing users’ potential preferences due to inherent biases in the data. Existing debiasing recommendation methods aim to eliminate biases from historical interaction data collected by recommender systems and have shown promising results. However, there is another significant bias that hinders the improvement of sequential recommendation models: dynamic item tendency bias. This bias arises because a period might have some unique tendencies consisting of items interacted with by users with the same intent, leading to a dynamic tendency distribution that biases the model training towards these tendencies. To address this issue, we propose a causal approach to model dynamic item tendency bias in sequential recommendation. We first extract tendencies on carefully designed item-item graphs through community detection. We then use causal intervention to conduct deconfounded training to capture true user preferences and introduce the beneficial item tendency bias to the inference process through optimal transport techniques. Experimental results on four real-world datasets demonstrate that our proposed method consistently outperforms state-of-the-art debiasing recommendation methods, confirming that our model is effective in reducing dynamic item tendency bias and dealing with tendency drifts.

Personalized review-based rating prediction aims at leveraging existing reviews to model user interests and item characteristics for rating prediction. Most of the existing studies mainly encounter two issues. First, the rich knowledge contained in the fine-grained aspects of each review and the knowledge graph is rarely considered to complement the pure text for better modeling user-item interactions. Second, the power of pre-trained language models is not carefully studied for personalized review-based rating prediction. To address these issues, we propose an approach named Knowledge-aware Collaborative Filtering with Pre-trained Language Model (KCF-PLM). For the first issue, to utilize rich knowledge, KCF-PLM develops a transformer network to model the interactions of the extracted aspects w.r.t. a user-item pair. For the second issue, to better represent users and items, KCF-PLM takes all the historical reviews of a user or an item as input to pre-trained language models. Moreover, KCF-PLM integrates the transformer network and the pre-trained language models through representation propagation on the knowledge graph and user-item guided attention of the aspect representations. Thus KCF-PLM combines review text, aspect, knowledge graph, and pre-trained language models together for review-based rating prediction. We conduct comprehensive experiments on several public datasets, demonstrating the effectiveness of KCF-PLM.

Graph Convolutional Networks (GCNs) have been widely applied to collaborative filtering, where each layer typically contains neighborhood aggregation and feature transformation. Recent studies have found that feature transformation contributes little to the final recommendation performance. They however eliminated it directly without further exploration, leading to a degradation of model expressive power. In this paper, we show that this problem arises from inconsistent information propagation process, in which the dominance of feature transformation prevents features from being properly smoothed by neighborhood aggregation. To this end, we present StableGCN to decouple and reconcile this contradictory process in an orderly rather than intertwined manner. The coarse-grained node features are first refined by an elaborate extractor, and then smoothed by a specific kind of GCN concerning feature denoising. Consequently, feature transformation and neighborhood aggregation can support each other without sacrificing expressive power. Extensive experiments on six public datasets demonstrate the effectiveness and state-of-the-art performance of StableGCN.

Knowledge distillation (KD) is a promising teacher-student learning paradigm that transfers information from a cumbersome teacher to a student network. To avoid the training cost of a large teacher network, the recent studies propose to distill knowledge from the student itself, called Self-KD. However, due to the limitations of the performance and capacity of the student, the soft-labels or features distilled by the student barely provide reliable guidance. Moreover, most of the Self-KD algorithms are specific to classification tasks based on soft-labels, and not suitable for semantic segmentation. To alleviate these contradictions, we revisit the label and feature distillation problem in segmentation, and propose Self-Decoupling and Ensemble Distillation for Efficient Segmentation (SDES). Specifically, we design a decoupled prediction ensemble distillation (DPED) algorithm that generates reliable soft-labels with multiple expert decoders, and a decoupled feature ensemble distillation (DFED) mechanism to utilize more important channel-wise feature maps for encoder learning. The extensive experiments on three public segmentation datasets demonstrate the superiority of our approach and the efficacy of each component in the framework through the ablation study.

Relevance modeling aims to locate desirable items for corresponding queries, which is crucial for search engines to ensure user experience. Although most conventional approaches address this problem by assessing the semantic similarity between the query and item, pure semantic matching is not everything. In reality, auxiliary query-item interactions extracted from user historical behavior data of the search log could provide hints to reveal users' search intents further. Drawing inspiration from this, we devise a novel Behavior Augmented Relevance Learning model for Alipay Search (BARL-ASe) that leverages neighbor queries of target item and neighbor items of target query to complement target query-item semantic matching. Specifically, our model builds multi-level co-attention for distilling coarse-grained and fine-grained semantic representations from both neighbor and target views. The model subsequently employs neighbor-target self-supervised learning to improve the accuracy and robustness of BARL-ASe by strengthening representation and logit learning. Furthermore, we discuss how to deal with the long-tail query-item matching of the mini apps search scenario of Alipay practically. Experiments on real-world industry data and online A/B testing demonstrate our proposal achieves promising performance with low latency.

Data-free knowledge distillation (DFKD) is a promising approach for addressing issues related to model compression, security privacy, and transmission restrictions. Although the existing methods exploiting DFKD have achieved inspiring achievements in coarse-grained classification, in practical applications involving fine-grained classification tasks that require more detailed distinctions between similar categories, sub-optimal results are obtained. To address this issue, we propose an approach called DFKD-FGVC that extends DFKD to fine-grained visual categorization (FGVC) tasks. Our approach utilizes an adversarial distillation framework with attention generator, mixed high-order attention distillation, and semantic feature contrast learning. Specifically, we introduce a spatial-wise attention mechanism to the generator to synthesize fine-grained images with more details of discriminative parts. We also utilize the mixed high-order attention mechanism to capture complex interactions among parts and the subtle differences among discriminative features of the fine-grained categories, paying attention to both local features and semantic context relationships. Moreover, we leverage the teacher and student models of the distillation framework to contrast high-level semantic feature maps in the hyperspace, comparing variances of different categories. We evaluate our approach on three widely-used FGVC benchmarks (Aircraft, Cars196, and CUB200) and demonstrate its superior performance. Code is available at https://github.com/RoryShao/DFKD-FGVC.git

Neuro-Rule Networks (NRNs) emerge as a promising neuro-symbolic method, enjoyed by the ability to equate fully-connected neural networks with logic rules. To support learning logic rules consisting of boolean variables, converting input features into binary representations is required. Different from discrete features that could be directly transformed by one-hot encodings, continuous features need to be binarized based on some numerical intervals. Existing studies usually select the bound values of intervals based on empirical strategies (e.g., equal-width interval). However, it is not optimal since the bounds are fixed and cannot be optimized to accommodate the ultimate training target. In this paper, we propose AutoInt, an approach that automatically binarizes continuous features and enables the intervals to be optimized with NRNs in an end-to-end fashion. Specifically, AutoInt automatically selects an interval for a given continuous feature in a soft manner to enable a differentiable learning procedure of interval-related parameters. Moreover, it introduces an additional soft K-means clustering loss to make the interval centres approach the original feature value distribution, thus reducing the risk of overfitting intervals. We conduct comprehensive experiments on public datasets and demonstrate the effectiveness of AutoInt in boosting the performance of NRNs.

Learning from positive and unlabeled data is known as positive-unlabeled (PU) learning in literature and has attracted much attention in recent years. One common approach in PU learning is to sample a set of pseudo-negatives from the unlabeled data using ad-hoc thresholds so that conventional supervised methods can be applied with both positive and negative samples. Owing to the label uncertainty among the unlabeled data, errors of misclassifying unlabeled positive samples as negative samples inevitably appear and may even accumulate during the training processes. Those errors often lead to performance degradation and model instability. To mitigate the impact of label uncertainty and improve the robustness of learning with positive and unlabeled data, we propose a new robust PU learning method with a training strategy motivated by the nature of human learning: easy cases should be learned first. Similar intuition has been utilized in curriculum learning to only use easier cases in the early stage of training before introducing more complex cases. Specifically, we utilize a novel ''hardness'' measure to distinguish unlabeled samples with a high chance of being negative from unlabeled samples with large label noise. An iterative training strategy is then implemented to fine-tune the selection of negative samples during the training process in an iterative manner to include more ''easy'' samples in the early stage of training. Extensive experimental validations over a wide range of learning tasks show that this approach can effectively improve the accuracy and stability of learning with positive and unlabeled data. Our code is available at https://github.com/woriazzc/Robust-PU.

Deep neural networks have achieved great success in sequential recommendation systems. While maintaining high competence in user modeling and next-item recommendation, these models have long been plagued by the numerous parameters and computation, which inhibit them to be deployed on resource-constrained mobile devices. Model quantization, as one of the main paradigms for compression techniques, converts float parameters to low-bit values to reduce parameter redundancy and accelerate inference. To avoid drastic performance degradation, it usually requests a fine-tuning phase with an original dataset. However, the training set of user-item interactions is not always available due to transmission limits or privacy concerns. In this paper, we propose a novel framework to quantize sequential recommenders without access to any real private data. A generator is employed in the framework to synthesize fake sequence samples to feed the quantized sequential recommendation model and minimize the gap with a full-precision sequential recommendation model. The generator and the quantized model are optimized with a min-max game — alternating discrepancy estimation and knowledge transfer. Moreover, we devise a two-level discrepancy modeling strategy to transfer information between the quantized model and the full-precision model. The extensive experiments of various recommendation networks on three public datasets demonstrate the effectiveness of the proposed framework.

Session-based recommendation aims at predicting the next item that a user is more likely to interact with by a target behavior type. Most of the existing session-based recommendation methods focus on developing powerful representation learning approaches to model items' sequential correlations, whereas they usually encounter the following limitations. Firstly, they only utilize sessions that belong to the target behavior type, neglecting the potential of leveraging other behavior types as auxiliary information for modeling user preference. Secondly, they separately model item-to-item relations for each session, overlooking to globally characterize the relations across different sessions for better item representations. To overcome these limitations, we first build a Multi-Relational Item Graph (MRIG) involving target and auxiliary behavior types over all sessions. Consequently, a novel Graph Neural Network (GNN) based model is devised to encode MRIG's item-to-item relations into target and auxiliary session-based representations, and adaptively fuse them to represent user interests. To facilitate model training, we further incorporate link prediction into multi-relational item graph modeling, acting as a simple but relevant task to session-based recommendation. The extensive experiments on real-world datasets demonstrate the superiority of the model over diverse and competitive baselines, validating its main components' significant contributions.

The proliferation of location-aware social networks (LSNs) has facilitated the research of user mobility modeling and check-in prediction, thereby benefiting various downstream applications such as precision marketing and urban management. Most of the existing studies only focus on predicting the spatial aspect of check-ins, whereas the joint inference of the spatial and temporal aspects more fits the real application scenarios. Moreover, although social relations have been extensively studied in a recommender system, only a few efforts have been observed in the next check-in location prediction, leaving room for further improvement. In this article, we study the next check-in inference problem, which demands the joint inference of the next check-in location (Where) and time (When) for a target user (Who). We devise a model named ARNPP-GAT, which combines an attention-based recurrent neural point process with a graph attention networks. The core technical insight of ARNPP-GAT is to integrate user long-term representation learning, short-term behavior modeling, and temporal point process into a unified architecture. Specifically, ARNPP-GAT first leverages graph attention networks to learn the long-term representation of users by encoding their social relations. More importantly, the ARNPP endows the model with the capability of characterizing the effects of past check-in events and performing multitask learning to yield the next check-in time and location prediction. Empirical results on two real-world data sets demonstrate that ARNPP-GAT is superior compared with several competitors, validating the contributions of multitask learning and social relation modeling.

In this article, we investigate the problem of social link inference in a target location-aware social network (LSN), which aims at predicting the unobserved links between users within the network. This problem is critical for downstream applications, including network completion and friend recommendation. In addition to the network structures commonly used in general link prediction, the studies tailored for social link inference in an LSN leverage user trajectories from the spatial aspect. However, the temporal factor lying in user trajectories is largely overlooked by most of the prior studies, limiting the capabilities of capturing the temporal relevance between users. Moreover, effective user matching by fusing different views, i.e., social, spatial, and temporal factors, remains unresolved, which hinders the potential improvement of link inference. To this end, this article devises a novel multiview matching network (MVMN) by regarding each of the three factors as one view of any target user pair. MVMN enjoys the flexibility and completeness of modeling each factor by developing its suitable matching module: 1) location matching module; 2) time-series matching module; and 3) relation matching module. Each module learns a view-specific representation for matching, and MVMN fuses them for final link inference. Extensive experiments on two real-world data sets demonstrate the superiority of our approach against several competitive baselines for link prediction and sequence matching, validating the contribution of its key components.

Knowledge tracing, the goal of which is predicting students’ future performance given their past question response sequences to trace their knowledge states, is pivotal for computer-aided education and intelligent tutoring systems. Although many technical efforts have been devoted to modeling students based on their question-response sequences, fine-grained interaction modeling between question-response pairs within each sequence is underexplored. This causes question-response representations less contextualized and further limits student modeling. To address this issue, we first conduct a data analysis and reveal the existence of complex cross effects between different question-response pairs within a sequence. Consequently, we propose MRT-KT, a multi-relational transformer for knowledge tracing, to enable fine-grained interaction modeling between question-response pairs. It introduces a novel relation encoding scheme based on knowledge concepts and student performance. Comprehensive experimental results show that MRT-KT outperforms state-of-the-art knowledge tracing methods on four widely-used datasets, validating the effectiveness of considering fine-grained interaction for knowledge tracing.

As a popular deep neural networks (DNN) compression technique, knowledge distillation (KD) has attracted increasing attentions recently. Existing KD methods usually utilize one kind of knowledge in an intermediate layer of DNN for classification tasks to transfer useful information from cumbersome teacher networks to compact student networks. However, this paradigm is not very suitable for semantic segmentation, a comprehensive vision task based on both pixel-level and contextual information, since it cannot provide rich information for distillation. In this paper, we propose a novel multi-knowledge aggregation and transfer (MKAT) framework to comprehensively distill knowledge within an intermediate layer for semantic segmentation. Specifically, the proposed framework consists of three parts: Independent Transformers and Encoders module (ITE), Auxiliary Prediction Branch (APB), and Mutual Label Calibration (MLC) mechanism, which can take advantage of abundant knowledge from intermediate features. To demonstrate the effectiveness of our proposed approach, we conduct extensive experiments on three segmentation datasets: Pascal VOC, Cityscapes, and CamVid, showing that MKAT outperforms the other KD methods.

Entity linking aims to link ambiguous mentions to their corresponding entities in a knowledge base, which is significant and fundamental for various downstream applications, e.g., knowledge base completion, question answering, and information extraction. While great efforts have been devoted to this task, most of these studies follow the assumption that large-scale labeled data is available. However, when the labeled data is insufficient for specific domains due to labor-intensive annotation work, the performance of existing algorithms will suffer an intolerable decline. In this paper, we endeavor to solve the problem of few-shot entity linking, which only requires a minimal amount of in-domain labeled data and is more practical in real situations. Specifically, we firstly propose a novel weak supervision strategy to generate non-trivial synthetic entity-mention pairs based on mention rewriting. Since the quality of the synthetic data has a critical impact on effective model training, we further design a meta-learning mechanism to assign different weights to each synthetic entity-mention pair automatically. Through this way, we can profoundly exploit rich and precious semantic information to derive a well-trained entity linking model under the few-shot setting. The experiments on real-world datasets show that the proposed method can extensively improve the state-of-the-art few-shot entity linking model and achieve impressive performance when only a small amount of labeled data is available. Moreover, we also demonstrate the outstanding ability of the model's transferability. Our code and models will be open-sourced.

With the rapid growth of service requirements, the rising concern of resource allocation fairness (e.g., the actual gained Quality of Service(QoS)) leads to the popularity of studies for fair scheduling mechanisms in service systems. The Generalized Processor Sharing (GPS) mechanism has been widely utilized as the fair reference base in the resource allocation due to its fairness and flexible configuration (i.e. scheduling based on weights). The urgent objective of GPS is to accurately evaluate the QoS of each service. Due to the inherent interconnected feature of the GPS mechanism, it is challenging to accurately evaluate individual service’ obtained QoS metrics by analytical-based methods with the growing number of services. Besides, considering the burstiness of the service requests, self-similar process is utilized to delineate the arrival of the service. Therefore, we propose a deep learning based approach, terms as DLPE_GPS, to accurately compute the QoS metrics of individual services in multi-queue GPS under self-similar request traffic. Specifically, DLPE_GPS firstly leverages knowledge-driven features to characterize each service (i.e., arrival rates, weight assigned, etc), and then the representation of each service is computed by a well designed multi-head attention mechanism considering mutual effect be-tween different service subsystems. After that, the knowledge-driven features and intermediate capacities are fused for QoS evaluation. Finally, we conduct complex experiments to show the effectiveness of the proposed method in terms of various aspects.

Recommender System (RS) is ubiquitous on today’s Internet to provide multifaceted personalized information services. While an enormous success has been made in pushing forward high-accuracy recommendations, the other side of the coin — the recommendation explainability — needs to be better handled for pursuing persuasiveness, especially for the era of deep learning based recommendation. A few research efforts investigate interpretable recommendation from the feature and result levels. Compared with them, model-level explanation, which unfolds the reasoning process of recommendation through transparent models, still remains underexplored and deserves more attention.

Traditional Chinese Medicine (TCM) has the longest clinical history in Asia and contributes a lot to health maintenance worldwide. An essential step during the TCM diagnostic process is syndrome induction, which comprehensively analyzes the symptoms and generates an overall summary of the symptoms. Given a set of symptoms, the existing herb recommenders aim to generate the corresponding herbs as a treatment by inducing the implicit syndrome representations based on TCM prescriptions. As different symptoms have various importance during the comprehensive consideration, we argue that treating the co-occurred symptoms equally to do syndrome induction in the previous studies will lead to the coarse-grained syndrome representation. In this paper, we bring the attention mechanism to model the syndrome induction process. Given a set of symptoms, we leverage an attention network to discriminate the symptom importance and adaptively fuse the symptom embeddings. Besides, we introduce a TCM knowledge graph to enrich the input corpus and improve the quality of representation learning. Further, we build a KG-enhanced Multi-Graph Neural Network architecture, which performs the attentive propagation to combine node feature and graph structural information. Extensive experimental results on two TCM data sets show that our proposed model has the outstanding performance over the state-of-the-arts.

Short text clustering has far-reaching effects on semantic analysis, showing its importance for multiple applications such as corpus summarization and information retrieval. However, it inevitably encounters the severe sparsity of short text representations, making the previous clustering approaches still far from satisfactory. In this paper, we present a novel attentive representation learning model for shot text clustering, wherein cluster-level attention is proposed to capture the correlations between text representations and cluster representations. Relying on this, the representation learning and clustering for short texts are seamlessly integrated into a unified model. To further ensure robust model training for short texts, we apply adversarial training to the unsupervised clustering setting, by injecting perturbations into the cluster representations. The model parameters and perturbations are optimized alternately through a minimax game. Extensive experiments on four real-world short text datasets demonstrate the superiority of the proposed model over several strong competitors, verifying that robust adversarial training yields substantial performance gains.

In this article, we study the problem of guaranteed display ads (GDAs) allocation, which requires proactively allocate display ads to different impressions to fulfill their impression demands indicated in the contracts. Existing methods for this problem either assume the impressions that are static or solely consider a specific ad's benefits. Thus, it is hard to generalize to the industrial production scenario where the impressions are dynamical and large-scale, and the overall allocation optimality of all the considered GDAs is required. To bridge this gap, we formulate this problem as a sequential decision-making problem in the scope of multiagent reinforcement learning (MARL), by assigning an allocation agent to each ad and coordinating all the agents for allocating GDAs. The inputs are the states (e.g., the demands of the ad and the remaining time steps for displaying the ads) of each ad and the impressions at different time steps, and the outputs are the display ratios of each ad for each impression. Specifically, we propose a novel hierarchical MARL (HMARL) method that creates hierarchies over the agent policies to handle a large number of ads and the dynamics of impressions. HMARL contains: 1) a manager policy to navigate the agent to choose an appropriate subpolicy and 2) a set of subpolicies that let the agents perform diverse conditioning on their states. Extensive experiments on three real-world data sets from the Tencent advertising platform with tens of millions of records demonstrate significant improvements of HMARL over state-of-the-art approaches.

Sequential product recommendation, aiming at predicting the products that a target user will interact with soon, has become a hotspot topic. Most of the sequential recommendation models focus on learning from users’ interacted product sequences in a purely data-driven manner. However, they largely overlook the knowledgeable substitutable and complementary relations between products. To address this issue, we propose a novel Substitutable and Complementary Graph-based Sequential Product Recommendation model, namely, SCG-SPRe. The innovations of SCG-SPRe lie in its two main modules: (1) The module of interactive graph neural networks jointly encodes the high-order product correlations in the substitutable graph and the complementary graph into two types of relation-specific product representations. (2) The module of kernel-enhanced transformer networks adaptively fuses multiple temporal kernels to characterize the unique temporal patterns between a candidate product to be recommended and any interacted product in a target behavior sequence. Thanks to the seamless integration of the two modules, SCG-SPRe obtains candidate-dependent user representations for different candidate products to compute the corresponding ranking scores. We conduct extensive experiments on three public datasets, demonstrating SCG-SPRe is superior to competitive sequential recommendation baselines and validating the benefits of explicitly modeling the product-product relations.

In community-based question answering (CQA) platforms, automatic answer ranking for a given question is critical for finding potentially popular answers in early times. The mainstream approaches learn to generate answer ranking scores based on the matching degree between question and answer representations as well as the influence of respondents. However, they encounter two main limitations: (1) Correlations between answers in the same question are often overlooked. (2) Question and respondent representations are built independently of specific answers before affecting answer representations. To address the limitations, we devise a novel graph-based tri-attention network, namely GTAN, which has two innovations. First, GTAN proposes to construct a graph for each question and learn answer correlations from each graph through graph neural networks (GNNs). Second, based on the representations learned from GNNs, an alternating tri-attention method is developed to alternatively build target-aware respondent representations, answer-specific question representations, and context-aware answer representations by attention computation. GTAN finally integrates the above representations to generate answer ranking scores. Experiments on three real-world CQA datasets demonstrate GTAN significantly outperforms state-of-the-art answer ranking methods, validating the rationality of the network architecture.

Sequential Recommendation aims to recommend items that a target user will interact with in the near future based on the historically interacted items. While modeling temporal dynamics is crucial for sequential recommendation, most of the existing studies concentrate solely on the user side while overlooking the sequential patterns existing in the counterpart, i.e., the item side. Although a few studies investigate the dynamics involved in the dual sides, the complex user-item interactions are not fully exploited from a global perspective to derive dynamic user and item representations. In this paper, we devise a novel Dynamic Representation Learning model for Sequential Recommendation (DRL-SRe). To better model the user-item interactions for characterizing the dynamics from both sides, the proposed model builds a global user-item interaction graph for each time slice and exploit time-sliced graph neural networks to learn user and item representations. Moreover, to enable the model to capture fine-grained temporal information, we propose an auxiliary temporal prediction task over consecutive time slices based on temporal point process. Comprehensive experiments on three public real-world datasets demonstrate DRL-SRe outperforms the state-of-the-art sequential recommendation models with a large margin.

Unsupervised Domain Adaptation (UDA) can tackle the challenge that convolutional neural network (CNN)-based approaches for semantic segmentation heavily rely on the pixel-level annotated data, which is labor-intensive. However, existing UDA approaches in this regard inevitably require the full access to source datasets to reduce the gap between the source and target domains during model adaptation, which are impractical in the real scenarios where the source datasets are private, and thus cannot be released along with the well-trained source models. To cope with this issue, we propose a source-free domain adaptation framework for semantic segmentation, namely SFDA, in which only a well-trained source model and an unlabeled target domain dataset are available for adaptation. SFDA not only enables to recover and preserve the source domain knowledge from the source model via knowledge transfer during model adaptation, but also distills valuable information from the target domain for self-supervised learning. The pixel-and patch-level optimization objectives tailored for semantic segmentation are seamlessly integrated in the framework. The extensive experimental results on numerous benchmark datasets highlight the effectiveness of our framework against the existing UDA approaches relying on source data.

Model quantization is a promising approach to compress deep neural networks and accelerate inference, making it possible to be deployed on mobile and edge devices. To retain the high performance of full-precision models, most existing quantization methods focus on fine-tuning quantized model by assuming training datasets are accessible. However, this assumption sometimes is not satisfied in real situations due to data privacy and security issues, thereby making these quantization methods not applicable. To achieve zero-short model quantization without accessing training data, a tiny number of quantization methods adopt either post-training quantization or batch normalization statistics-guided data generation for fine-tuning. However, both of them inevitably suffer from low performance, since the former is a little too empirical and lacks training support for ultra-low precision quantization, while the latter could not fully restore the peculiarities of original data and is often low efficient for diverse data generation. To address the above issues, we propose a zero-shot adversarial quantization (ZAQ) framework, facilitating effective discrepancy estimation and knowledge transfer from a full-precision model to its quantized model. This is achieved by a novel two-level discrepancy modeling to drive a generator to synthesize informative and diverse data examples to optimize the quantized model in an adversarial learning fashion. We conduct extensive experiments on three fundamental vision tasks, demonstrating the superiority of ZAQ over the strong zero-shot baselines and validating the effectiveness of its main components. Code is available at https://git.io/Jqc0y.

Rule-based models, e.g., decision trees, are widely used in scenarios demanding high model interpretability for their transparent inner structures and good model expressivity. However, rule-based models are hard to optimize, especially on large data sets, due to their discrete parameters and structures. Ensemble methods and fuzzy/soft rules are commonly used to improve performance, but they sacrifice the model interpretability. To obtain both good scalability and interpretability, we propose a new classifier, named Rule-based Representation Learner (RRL), that automatically learns interpretable non-fuzzy rules for data representation and classification. To train the non-differentiable RRL effectively, we project it to a continuous space and propose a novel training method, called Gradient Grafting, that can directly optimize the discrete model using gradient descent. An improved design of logical activation functions is also devised to increase the scalability of RRL and enable it to discretize the continuous features end-to-end. Exhaustive experiments on nine small and four large data sets show that RRL outperforms the competitive interpretable approaches and can be easily adjusted to obtain a trade-off between classification accuracy and model complexity for different scenarios. Our code is available at: https://github.com/12wang3/rrl.

Models with transparent inner structure and high classification performance are required to reduce potential risk and provide trust for users in domains like health care, finance, security, etc. However, existing models are hard to simultaneously satisfy the above two properties. In this paper, we propose a new hierarchical rule-based model for classification tasks, named Concept Rule Sets (CRS), which has both a strong expressive ability and a transparent inner structure. To address the challenge of efficiently learning the non-differentiable CRS model, we propose a novel neural network architecture, Multilayer Logical Perceptron (MLLP), which is a continuous version of CRS. Using MLLP and the Random Binarization (RB) method we proposed, we can search the discrete solution of CRS in continuous space using gradient descent and ensure the discrete CRS acts almost the same as the corresponding continuous MLLP. Experiments on 12 public data sets show that CRS outperforms the state-of-the-art approaches and the complexity of the learned CRS is close to the simple decision tree.

Domain-adapted sentiment classification refers to training on a labeled source domain to well infer document-level sentiment on an unlabeled target domain. Most existing relevant models involve a feature extractor and a sentiment classifier, where the feature extractor works towards learning domain-invariant features from both domains, and the sentiment classifier is trained only on the source domain to guide the feature extractor. As such, they lack a mechanism to use sentiment polarity lying in the target domain. To improve domain-adapted sentiment classification by learning sentiment from the target domain as well, we devise a novel deep adversarial mutual learning approach involving two groups of feature extractors, domain discriminators, sentiment classifiers, and label probers. The domain discriminators enable the feature extractors to obtain domain-invariant features. Meanwhile, the label prober in each group explores document sentiment polarity of the target domain through the sentiment prediction generated by the classifier in the peer group, and guides the learning of the feature extractor in its own group. The proposed approach achieves the mutual learning of the two groups in an end-to-end manner. Experiments on multiple public datasets indicate our method obtains the state-of-the-art performance, validating the effectiveness of mutual learning through label probers.

Personalized image caption, a natural extension of the standard image caption task, requires to generate brief image descriptions tailored for users' writing style and traits, and is more practical to meet users' real demands. Only a few recent studies shed light on this crucial task and learn static user representations to capture their long-term literal-preference. However, it is insufficient to achieve satisfactory performance due to the intrinsic existence of not only long-term user literal-preference, but also short-term literal-preference which is associated with users' recent states. To bridge this gap, we develop a novel multimodal hierarchical transformer network (MHTN) for personalized image caption in this paper. It learns short-term user literal-preference based on users' recent captions through a short-term user encoder at the low level. And at the high level, the multimodal encoder integrates target image representations with short-term literal-preference, as well as long-term literal-preference learned from user IDs. These two encoders enjoy the advantages of the powerful transformer networks. Extensive experiments on two real datasets show the effectiveness of considering two types of user literal-preference simultaneously and better performance over the state-of-the-art models.

Herb recommendation plays a crucial role in the therapeutic process of Traditional Chinese Medicine (TCM), which aims to recommend a set of herbs to treat the symptoms of a patient. While several machine learning methods have been developed for herb recommendation, they are limited in modeling only the interactions between herbs and symptoms, and ignoring the intermediate process of syndrome induction. When performing TCM diagnostics, an experienced doctor typically induces syndromes from the patient's symptoms and then suggests herbs based on the induced syndromes. As such, we believe the induction of syndromes - an overall description of the symptoms - is important for herb recommendation and should be properly handled. However, due to the ambiguity and complexity of syndrome induction, most prescriptions lack the explicit ground truth of syndromes. In this paper, we propose a new method that takes the implicit syndrome induction process into account for herb recommendation. Specifically, given a set of symptoms to treat, we aim to generate an overall syndrome representation by effectively fusing the embeddings of all the symptoms in the set, so as to mimic how a doctor induces the syndromes. Towards symptom embedding learning, we additionally construct a symptom-symptom graph from the input prescriptions for capturing the relations (cooccurred patterns) between symptoms; we then build graph convolution networks (GCNs) on both symptom-symptom and symptom-herb graphs to learn symptom embedding. Similarly, we construct a herb-herb graph and build GCNs on both herbherb and symptom-herb graphs to learn herb embedding, which is finally interacted with the syndrome representation to predict the scores of herbs. The advantage of such a Multi-Graph GCN architecture is that more comprehensive representations can be obtained for symptoms and herbs. We conduct extensive experiments on a public TCM dataset, demonstrating significant improvements over state-of-the-art herb recommendation methods. Further studies justify the effectiveness of our design of syndrome representation and multiple graphs.

Social Media Popularity (SMP) prediction focuses on predicting the social impact of a given post from a specific user in social media, which is crucial for online advertising, social recommendation, and demand prediction. In this paper, we present HyFea, our winning solution to the Social Media Prediction (SMP) Challenge for multimedia grand challenge of ACM Multimedia 2020. To address the multi-modality and personality issues of this challenge, HyFea carefully considers multiple feature types and adopts a tree-based ensembling method, i.e., CatBoost, which is shown to perform well in prediction. Specifically, HyFea involves the features related to Image, Category, Space-Time, User Profile, Tag, and Others. We conduct several experiments on the Social Media Prediction Dataset (SMPD), verifying the positive contributions of each type of features.

Sequential recommendation and group recommendation are two important branches in the field of recommender system. While considerable efforts have been devoted to these two branches in an independent way, we combine them by proposing the novel sequential group recommendation problem which enables modeling group dynamic representations and is crucial for achieving better group recommendation performance. The major challenge of the problem is how to effectively learn dynamic group representations based on the sequential user-item interactions of group members in the past time frames. To address this, we devise a Group-aware Long- and Short-term Graph Representation Learning approach, namely GLS-GRL, for sequential group recommendation. Specifically, for a target group, we construct a group-aware long-term graph to capture user-item interactions and item-item co-occurrence in the whole history, and a group-aware short-term graph to contain the same information regarding only the current time frame. Based on the graphs, GLS-GRL performs graph representation learning to obtain long-term and short-term user representations, and further adaptively fuse them to gain integrated user representations. Finally, group representations are obtained by a constrained user-interacted attention mechanism which encodes the correlations between group members. Comprehensive experiments demonstrate that GLS-GRL achieves better performance than several strong alternatives coming from sequential recommendation and group recommendation methods, validating the effectiveness of the core components in GLS-GRL.

Session-based target behavior prediction aims to predict the next item to be interacted with specific behavior types (e.g., clicking). Although existing methods for session-based behavior prediction leverage powerful representation learning approaches to encode items’ sequential relevance in a low-dimensional space, they suffer from several limitations. Firstly, they focus on only utilizing the same type of user behavior for prediction, but ignore the potential of taking other behavior data as auxiliary information. This is particularly crucial when the target behavior is sparse but important (e.g., buying or sharing an item). Secondly, item-to-item relations are modeled separately and locally in one behavior sequence, and they lack a principled way to globally encode these relations more effectively. To overcome these limitations, we propose a novel Multi-relational Graph Neural Network model for Session-based target behavior Prediction, namely MGNN-SPred for short. Specifically, we build a Multi-Relational Item Graph (MRIG) based on all behavior sequences from all sessions, involving target and auxiliary behavior types. Based on MRIG, MGNN-SPred learns global item-to-item relations and further obtains user preferences w.r.t. current target and auxiliary behavior sequences, respectively. In the end, MGNN-SPred leverages a gating mechanism to adaptively fuse user representations for predicting next item interacted with target behavior. The extensive experiments on two real-world datasets demonstrate the superiority of MGNN-SPred by comparing with state-of-the-art session-based prediction methods, validating the benefits of leveraging auxiliary behavior and learning item-to-item relations over MRIG.

In modern pavement management systems, pavement roughness is an important indicator of pavement performance, and it reflects the smoothness of pavement surface. International Roughness Index (IRI) is the de-facto metric to quantitatively analyze the roughness of pavement surface. The pavement with high IRI not only reduces the lifetime of vehicles, but also raises the risk of car accidents. Accurate prediction of IRI becomes a key task for the pavement management system, and it helps the transportation department refurbish the pavement in time. However, existing models are proposed on top of small datasets, and have poor performance. Besides, they only consider cross-sectional features of the pavements without any time-series information. In order to better capture the latent relationship between the cross-sectional and time-series features, we propose a novel feature fusion LSTM-BPNN model. LSTM-BPNN first learns the cross-sectional and time-series features with two neural networks separately, then it fuses both features via an attention mechanism. Experimental results on a high-quality real-world dataset clearly demonstrate that the new model outperforms existing considerable alternatives.

Mortality prediction is one of the essential tasks in medical data mining and is significant for inferring clinical outcomes. With a large number of medical notes collected from hospitals, there is an urgent need for developing effective models for predicting mortality based on them. In contrast to structured electronic health records, medical notes are unstructured texts written by experienced caregivers and contain more complicated information about patients, posing more challenges for modeling. Most previous studies rely on tedious hand-crafted features or generating indirect features based on some statistical models such as topic modeling, which might incur information loss for later model training. Recently, some deep models have been proposed to unify the stages of feature construction and model training. However, domain concept knowledge has been neglected, which is important to gain a better understanding of medical notes. To address the above issues, we propose novel Knowledge-aware Deep Dual Networks (K-DDN) for the text-based mortality prediction task. Specifically, a simple deep dual network is first proposed to fuse the representations of medical knowledge and raw text for prediction. Afterward, we incorporate a co-attention mechanism into the basic model, guiding the knowledge and text representation learning with the help of each other. Experimental results on two publicly real-world datasets show the proposed deep dual networks outperform state-of-the-art methods and the co-attention mechanism can further improve the performance.

Existing representation learning methods based on graph neural networks and their variants rely on the aggregation of neighborhood information, which makes it sensitive to noises in the graph, e.g. erroneous links between nodes, incorrect/missing node features. In this paper, we propose Graph Denoising Policy Network (short for GDPNet) to learn robust representations from noisy graph data through reinforcement learning. GDPNet first selects signal neighborhoods for each node, and then aggregates the information from the selected neighborhoods to learn node representations for the down-stream tasks. Specifically, in the signal neighborhood selection phase, GDPNet optimizes the neighborhood for each target node by formulating the process of removing noisy neighborhoods as a Markov decision process and learning a policy with task-specific rewards received from the representation learning phase. In the representation learning phase, GDPNet aggregates features from signal neighbors to generate node representations for down-stream tasks, and provides task-specific rewards to the signal neighbor selection phase. These two phases are jointly trained to select optimal sets of neighbors for target nodes with maximum cumulative task-specific rewards, and to learn robust representations for nodes. Experimental results on node classification task demonstrate the effectiveness of GDNet, outperforming the state-of-the-art graph representation learning methods on several well-studied datasets.

This paper presents our solution for the Hulu Content-Based Video Relevance Prediction (CBVRP) challenge, which focuses on cold-start videos as candidates. The keys to success of this prediction scenario are to learn effective user and video representations. To this end, we develop a multi-view multi-level deep interest network (MMDIN), which involves a multi-level deep interest network to learn user and video representations in a single-view, and a late fusion technique to integrate their multi-view representations corresponding to different types of video features. Through the above manner, the cold-start video prediction could be handled well with representations through their past interaction behaviors with videos and video representations based on their multiple types of content profiles.

Recently, the Visual Question Answering (VQA) task has gained increasing attention in artificial intelligence. Existing VQA methods mainly adopt the visual attention mechanism to associate the input question with corresponding image regions for effective question answering. The free-form region based and the detection-based visual attention mechanisms are mostly investigated, with the former ones attending free-form image regions and the latter ones attending pre-specified detection-box regions. We argue that the two attention mechanisms are able to provide complementary information and should be effectively integrated to better solve the VQA problem. In this paper, we propose a novel deep neural network for VQA that integrates both attention mechanisms. Our proposed framework effectively fuses features from free-form image regions, detection boxes, and question representations via a multi-modal multiplicative feature embedding scheme to jointly attend question-related free-form image regions and detection boxes for more accurate question answering. The proposed method is extensively evaluated on two publicly available datasets, COCO-QA and VQA, and outperforms state-of-the-art approaches. Source code is available at https://github.com/lupantech/dual-mfa-vqa.

In addition to only considering stocks' price series, utilizing short and instant texts from social medias like Twitter has potential to yield better stock market prediction. While some previous approaches have explored this direction, their results are still far from satisfactory due to their reliance on performance of sentiment analysis and limited capabilities of learning direct relations between target stock trends and their daily social texts. To bridge this gap, we propose a novel Cross-modal attention based Hybrid Recurrent Neural Network (CH-RNN), which is inspired by the recent proposed DA-RNN model. Specifically, CH-RNN consists of two essential modules. One adopts DA-RNN to gain stock trend representations for different stocks. The other utilizes recurrent neural network to model daily aggregated social texts. These two modules interact seamlessly by the following two manners: 1) daily representations of target stock trends from the first module are leveraged to select trend-related social texts through a cross-modal attention mechanism, and 2) representations of text sequences and trend series are further integrated. The comprehensive experiments on the real dataset we build demonstrate the effectiveness of CH-RNN and benefit of considering social texts.

Popularity prediction of user generated textual content is critical for prioritizing information in the web, which alleviates heavy information overload for ordinary readers. Most previous studies model each content instance separately for prediction and thus overlook the sequential correlations between instances of a specific user. In this paper, we go deeper into this problem based on the two observations for each user, i.e., sequential content correlation and sequential popularity correlation. We propose a novel deep sequential model called User Memory-augmented recurrent Attention Network (UMAN). This model encodes the two correlations by updating external user memories which is further leveraged for target text representation learning and popularity prediction. The experimental results on several real-world datasets validate the benefits of considering these correlations and demonstrate UMAN achieves best performance among several strong competitors.

Recently, Visual Question Answering (VQA) has emerged as one of the most significant tasks in multimodal learning as it requires understanding both visual and textual modalities. Existing methods mainly rely on extracting image and question features to learn their joint feature embedding via multimodal fusion or attention mechanism. Some recent studies utilize external VQA-independent models to detect candidate entities or attributes in images, which serve as semantic knowledge complementary to the VQA task. However, these candidate entities or attributes might be unrelated to the VQA task and have limited semantic capacities. To better utilize semantic knowledge in images, we propose a novel framework to learn visual relation facts for VQA. Specifically, we build up a Relation-VQA (R-VQA) dataset based on the Visual Genome dataset via a semantic similarity module, in which each data consists of an image, a corresponding question, a correct answer and a supporting relation fact. A well-defined relation detector is then adopted to predict visual question-related relation facts. We further propose a multi-step attention model composed of visual attention and semantic attention sequentially to extract related visual knowledge and semantic knowledge. We conduct comprehensive experiments on the two benchmark datasets, demonstrating that our model achieves state-of-the-art performance and verifying the benefit of considering visual relation facts.

Dynamic treatment recommendation systems based on large-scale electronic health records (EHRs) become a key to successfully improve practical clinical outcomes. Prior relevant studies recommend treatments either use supervised learning (e.g. matching the indicator signal which denotes doctor prescriptions), or reinforcement learning (e.g. maximizing evaluation signal which indicates cumulative reward from survival rates). However, none of these studies have considered to combine the benefits of supervised learning and reinforcement learning. In this paper, we propose Supervised Reinforcement Learning with Recurrent Neural Network (SRL-RNN), which fuses them into a synergistic learning framework. Specifically, SRL-RNN applies an off-policy actor-critic framework to handle complex relations among multiple medications, diseases and individual characteristics. The "actor'' in the framework is adjusted by both the indicator signal and evaluation signal to ensure effective prescription and low mortality. RNN is further utilized to solve the Partially-Observed Markov Decision Process (POMDP) problem due to lack of fully observed states in real world applications. Experiments on the publicly real-world dataset, i.e., MIMIC-3, illustrate that our model can reduce the estimated mortality, while providing promising accuracy in matching doctors' prescriptions.

Short text stream clustering has become an increasingly important problem due to the explosive growth of short text in diverse social medias. In this paper, we propose a model-based short text stream clustering algorithm (MStream) which can deal with the concept drift problem and sparsity problem naturally. The MStream algorithm can achieve state-of-the-art performance with only one pass of the stream, and can have even better performance when we allow multiple iterations of each batch. We further propose an improved algorithm of MStream with forgetting rules called MStreamF, which can efficiently delete outdated documents by deleting clusters of outdated batches. Our extensive experimental study shows that MStream and MStreamF can achieve better performance than three baselines on several real datasets.

Community-acquired pneumonia (CAP) is a major death cause for children, requiring an early administration of appropriate antibiotics to cure it. To achieve this, accurate detection of pathogenic microorganism is crucial, especially for reducing the abuse of antibiotics. Conventional gold standard detection methods are mainly etiology based, incurring high cost and labor intensity. Although recently electronic health records (EHRs) become prevalent and widely used, their power for automatically determining pathogenic microorganism has not been investigated. In this paper, we formulate a new problem for automatically detecting pathogenic microorganism of CAP by considering patient biomedical features from EHRs, including time-varying body temperatures and common laboratory measurements. We further develop a Patient Attention based Recurrent Neural Network (PA-RNN) model to fuse different patient features for detection. We conduct experiments on a real dataset, demonstrating utilizing electronic health records yields promising performance and PA-RNN outperforms several alternatives.

Popularity prediction for the growing social images has opened unprecedented opportunities for wide commercial applications, such as precision advertising and recommender system. While a few studies have explored this significant task, little research has addressed its unstructured properties of both visual and textual modalities, and further considered to learn effective representation from multi-modalities for popularity prediction. To this end, we propose a model named User-guided Hierarchical Attention Network (UHAN) with two novel user-guided attention mechanisms to hierarchically attend both visual and textual modalities. It is capable of not only learning effective representation for each modality, but also fusing them to obtain an integrated multi-modal representation under the guidance of user embedding. As no benchmark dataset exists, we extend a publicly available social image dataset by adding the descriptions of images. The comprehensive experiments have demonstrated the rationality of our proposed UHAN and its better performance than several strong alternatives.

Popularity prediction, aiming at predicting target items' total interactions with users, is a very significant type of problem and has attracted a lot of attention in recent years. It can benefit a lot of real applications, such as cold-start recommendation[8] and online advertising [4]. The Social Media Prediction Task-1 (SMP-T1) of the ACM Multimedia 2017 Grand Challenge is designed to predict popularity of photos published by users in social media.

The availability of massive geo-annotated social media data sheds light on studying human mobility patterns. Among them, periodic pattern, \ie an individual visiting a geographical region with some specific time interval, has been recognized as one of the most important. Mining periodic patterns has a variety of applications, such as location prediction, anomaly detection, and location- and time-aware recommendation. However, it is a challenging task: the regions of a person and the periods of each region are both unknown. The interdependency between them makes the task even harder. Hence, existing methods are far from satisfactory for detecting periodic patterns from the low-sampling and noisy social media data.

Personalized review-based rating prediction, a newly emerged research problem, aims at inferring users’ ratings over their unrated items using existing reviews and corresponding ratings. While some researchers proposed to learn topic factor from review text to obtain interpretability for rating prediction, they often overlooked the fact that the learned topic factors are limited to review text and cannot fully reveal the complicated relations between reviews and ratings. Moreover, topic modeling based solutions for this problem usually utilize Gibbs sampling algorithms to learn topics and word distributions, resulting in non-negligible computational overload. To address the above challenges, we propose an integrated topic and latent factor model (ITLFM), which combines topic and latent factors in a linear way to make them complement each other for better accuracies in rating prediction tasks. In addition, ITLFM models review text through an additive topic model to reveal user's and item's topic factors simultaneously. To ensure high learning efficiency, we design a hybrid stochastic learning algorithm for ITLFM. We evaluate ITLFM on several standard benchmarks and compare with representative approaches. The experimental results demonstrate that the proposed ITLFM method is computationally efficient and accurate, as well as scalable for large scale applications.

In location-based social networks (LBSNs), new successive point-of-interest (POI) recommendation is a newly formulated task which tries to regard the POI a user currently visits as his POI-related query and recommend new POIs the user has not visited before. While carefully designed methods are proposed to solve this problem, they ignore the essence of the task which involves retrieval and recommendation problem simultaneously and fail to employ the social relations or temporal information adequately to improve the results.

What is happening around the world? When and where? Mining the geo-tagged Twitter stream makes it possible to answer the above questions in real-time. Although a single tweet can be short and noisy, proper aggregations of tweets can provide meaningful results. In this paper, we focus on hierarchical spatio-temporal hashtag clustering techniques. Our system has the following features: (1) Exploring events (hashtag clusters) with different space granularity. Users can zoom in and out on maps to find out what is happening in a particular area. (2) Exploring events with different time granularity. Users can choose to see what is happening today or in the past week. (3) Efficient single-pass algorithm for event identification, which provides human-readable hashtag clusters. (4) Efficient event ranking which aims to find burst events and localized events given a particular region and time frame. To support aggregation with different space and time granularity, we propose a data structure called STREAMCUBE, which is an extension of the data cube structure from the database community with spatial and temporal hierarchy. To achieve high scalability, we propose a divide-and-conquer method to construct the STREAMCUBE. To support flexible event ranking with different weights, we proposed a top-k based index. Different efficient methods are used to speed up event similarity computations. Finally, we have conducted extensive experiments on a real twitter data. Experimental results show that our framework can provide meaningful results with high scalability.

Event-based social networks (EBSNs), in which organizers publish events to attract other users in local city to attend offline, emerge in recent years and grow rapidly. Due to the large volume of events in EBSNs, event recommendation is essential. A few recent works focus on this task, while almost all the methods need that each event to be recommended should have been registered by some users to attend. Thus they ignore two essential characteristics of events in EBSNs: (1) a large number of new events will be published every day which means many events have few participants in the beginning, (2) events have life cycles which means outdated events should not be recommended. Overall, event recommendation in EBSNs inevitably faces the cold-start problem.

Groups play an essential role in many social websites which promote users' interactions and accelerate the diffusion of information. Recommending groups that users are really interested to join is significant for both users and social media. While traditional group recommendation problem has been extensively studied, we focus on a new type of the problem, i.e., event-based group recommendation. Unlike the other forms of groups, users join this type of groups mainly for participating offline events organized by group members or inviting other users to attend events sponsored by them. These characteristics determine that previously proposed approaches for group recommendation cannot be adapted to the new problem easily as they ignore the geographical influence and other explicit features of groups and users.