The performance of graph collaborative filtering (GCF) models could be affected by noisy user-item interactions. Existing studies on data denoising either ignore the nature of noise in implicit feedback or seldom consider the long-tail distribution of historical interaction data. For the first challenge, we analyze the role of noise from a causal perspective: noise is an unobservable confounder. Therefore, we use the instrumental variable for causal intervention without requiring confounder observation. For the second challenge, we consider degree distribution of nodes in the course of causal intervention. And then we propose a model named causal graph collaborative filtering (CausalGCF) to denoise implicit feedback for GCF. Specifically, we design a degree augmentation strategy as the instrumental variable. First, we divide nodes into head and tail nodes according to their degree. Then, we purify the interactions of the head nodes and enrich those of the tail nodes based on similarity. We perform degree augmentation strategy from the user and item sides to obtain two different graph structures, which are trained together with self-supervised learning. Empirical studies on four real and four synthetic datasets demonstrate the effectiveness of CausalGCF, which is more robust against noisy interactions in implicit feedback than the baselines.

Due to the inherent challenges in acquiring explicit feedback, graph collaborative filtering (GCF) models often resort to implicit feedback. However, there exists noise in implicit feedback that may not accurately reflect users’ preferences. This noise will be amplified by the aggregating and propagating operations of GCF, thereby affecting the performance of GCF. Existing noise mitigation methods attempt to filter noisy samples from implicit feedback data, yet they face limitations such as dependency on side information for sample selection, neglect of false-negative noise, and disregard for the impact of previous selections on current iteration. To overcome these challenges, we propose adynamicgraphlearning framework todenoise implicit feedback for GCF (DGLD). DGLD comprises a graph learning module and a reinforcement learning module. The graph learning module evaluates the confidence degrees of interactions by leveraging user–item cosine similarity and global user preferences, updating user–item interaction graph without relying on side information. Meanwhile, the reinforcement learning module employs a policy network to select “false-positive denoising” and “false-negative denoising” actions based on performance of the recommendation model and state of interaction graph obtained from previous iterations. These modules work synergistically to dynamically denoise implicit feedback. Experimental results on three benchmark datasets underscore the superiority of our approach over state-of-the-art general and denoising recommendation models.

Cold-start recommendation is a well-known problem in practical application scenarios. Generating reliable recommendations can be challenging when interactions are typically sparse. To mitigate the cold-start problem, some methods incorporate auxiliary information about users and items, and others adopt meta-learning to improve recommendation accuracy. However, these approaches overlook the fact that items are interdependent and likely to be related or similar. Moreover, user preference distributions in the meta-training and meta-testing phases are different in the cold-start scenario. To address these problems, we present a novel strategy called Causal Meta-learning with Multi-view Graphs (CausalMMG). Specifically, we first construct multi-view item-item graphs to explore the correlations and similarities between items from multiple perspectives. A multi-view item representer is then used to learn item representations, exploiting graph convolution neural networks to capture the structure of these different item–item graphs. We then resort to the structural causal models of causal inference and further develop a causality-enhanced bi-level adaptive meta-learner to eliminate bias caused by the different distributions of user preferences. Moreover, the meta-learner learns the user preferences for items in different orders through hierarchical and task-level adaptations. Finally, we evaluate CausalMMG on several real-world datasets, demonstrating its effectiveness in various scenarios. The results show that the proposed CausalMMG is significantly superior to competitive baseline methods for cold-start recommendation on all datasets, highlighting the importance of incorporating the multiple relationships between items and modeling different user preference distributions in recommender systems.

Interactive recommender systems have garnered widespread attention due to their ability to dynamically update recommendation strategies based on user feedback, enhancing the user's interactive experience. To maximize long-term user satisfaction, existing research has incorporated reinforcement learning into interactive recommender systems and combined it with meta-learning to form a meta-reinforcement learning framework that further addresses the cold-start problem in interactive recommendation. However, on one hand, there are latent confounders affecting user feedback; on the other hand, since training samples are observed rather than experimentally obtained, selection bias and exposure bias exist in the interactive data. Most existing studies remove biases using the method of Inverse Propensity Score, which often utilizes fixed propensity scores and neglects the latent confounders affecting user feedback. In this paper, we propose an unbiased interactive recommender system (UIRS) based on a meta-reinforcement learning framework. To eliminate the impact of latent confounders in the state encoding process, we design a user preference representer consisting of three interconnected gated recurrent units. Additionally, we use the item recommendation probabilities output from the policy network as propensity scores and design the objective functions based on these scores, to eliminate biases while addressing latent confounders. Extensive experiments conducted on three benchmark datasets demonstrate that our proposed UIRS model achieves significant improvements over existing state-of-the-art baseline models.

Conversational recommender systems (CRS) have garnered significant attention in academia and industry because of their ability to capture user preferences via system questions and user responses. Typically, in a CRS, reinforcement learning (RL) is utilized to determine the optimal timing for requesting attribute information or suggesting items. However, existing methods consider user-preferred attributes independently and ignore that attributes may be of different importance to the same user, in the attribute and item selection phases, which limits the accuracy and interpretability of CRS. Inspired by this, we propose deep conversational path reasoning (DeepCPR), which involves constructing a reasoning path on a graph with a series of user-favored attributes. It utilizes the attention mechanism to thoroughly examine the connections between these attributes and provide improved explanations for which attributes to inquire about or which items to recommend. In DeepCPR, two deep-learning-based modules are proposed to realize attribute and item selection. In the first module, the sequence of attributes confirmed by the user in conversation is encoded with a gated graph neural network to obtain the user’s long-term preference using a self-attention mechanism for the selection of candidate attributes. In the second module, a self-attention approach with more appropriate strategies is developed to dynamically select candidate items. In addition, to achieve fine-grained user preference modeling, a recurrent neural network is employed to aggregate the sequence of attributes that interact with the users. Numerous experimental evaluations conducted on four real CRS datasets show that the proposed method significantly outperforms existing advanced methods in terms of conversational recommendations.

Zero-shot image recognition attempts to simulate the zero-shot learning mechanism of humans and recognizes the images of novel classes. It is crucial to learn transferable knowledge from seen classes and generalize it to unseen classes for image recognition in zero-shot learning (ZSL). Most existing ZSL methods extract visual features with pretrained backbone networks and learn transferable knowledge with the extracted visual features. However, the backbone networks are not pretrained for a special task, and the extracted visual features usually contain some distractive information for the ZSL task, which causes some discriminative information to be ignored or weakened and degrades the quality of knowledge learned from seen classes. Moreover, since visual samples of unseen classes are not obtainable, domain shift is another challenging problem. In this paper, we propose visual feature enhancement to learn more discriminative visual features via a graph convolutional network (GCN) and an attention mechanism for improving the quality of the learned transferable knowledge. Different from previous works, we explore the correlations between different latent visual patterns of an image and introduce GCN to enhance visual features. On the other hand, we take advantage of different learning mechanisms of GCN and MLP and propose dual classifier learning for improving the generalization and inference capabilities of our model. In end-to-end model training, the module of visual feature enhancement and the module of dual classifier learning are beneficial to each other via joint optimization. Finally, we perform extensive experiments in the ZSL setting and GZSL setting. The extensive experimental results verify the effectiveness and superiority of our method.

Data uncertainty is inherent in many real-world applications such as sensor data monitoring and mobile tracking. Mining sequential patterns from uncertain/inaccurate data, such as sensor readings and GPS trajectories, is important to discover hidden knowledge in such applications. This paper addres ses the problem of pattern matching with periodical wildcards for uncertain sequences. We present a dynamic programming approach, called CoDP, to compute the exact probability that a pattern q is a subsequence of an uncertain sequence s, and this approach can be further applied to substring matching for uncertain sequences. The efficiency and effectiveness of our algorithm have been verified through extensive experiments on both real and synthetic data.

Recently, document recommendation has become a very hot research area in online services. Since rating information is usually sparse with exploding growth of the numbers of users and items, conventional collaborative filtering-based methods degrade significantly in recommendation performance. To address this sparseness problem, auxiliary information such as item content information may be utilized. Convolution matrix factorization (ConvMF) is an appealing method, which tightly combines the rating and item content information. Although ConvMF captures contextual information of item content by utilizing convolutional neural network (CNN), the latent representation may not be effective when the rating information is very sparse. To address this problem, we generalize recent advances in supervised CNN and propose a novel recommendation model called supervised convolution matrix factorization (Super-ConvMF), which effectively combines the rating information, item content information and tag information into a unified recommendation framework. Experiments on three real-world datasets, two datasets come from MovieLens and the other one is from Amazon, show our model outperforms the state-of-the-art competitors in terms of the whole range of sparseness.