Data Poisoning Attack against Recommender System Using Incomplete and Perturbed Data

In an era of information explosion, recommender systems are vital tools to deliver personalized recommendations for users. The key of recommender systems is to forecast users' future behaviors based on previous user-item interactions. Due to their strong expressive power of capturing high-order connectivities in user-item interaction data, recent years have witnessed a rising interest in leveraging Graph Neural Networks (GNNs) to boost the prediction performance of recommender systems. Nonetheless, classic Matrix Factorization (MF) and Deep Neural Network (DNN) approaches still play an important role in real-world large-scale recommender systems due to their scalability advantages. Despite the existence of GNN-acceleration solutions, it remains an open question whether GNN-based recommender systems can scale as efficiently as classic MF and DNN methods. In this paper, we propose a Linear-Time Graph Neural Network (LTGNN) to scale up GNN-based recommender systems to achieve comparable scalability as classic MF approaches while maintaining GNNs' powerful expressiveness for superior prediction accuracy. Extensive experiments and ablation studies are presented to validate the effectiveness and scalability of the proposed algorithm. Our implementation based on PyTorch is available.

Recommendation systems get ever-increasing importance due to their applications in both academia and industry. The most popular type of these systems, known as collaborative filtering algorithms, employ user-item interactions to perform the recommendation tasks. With growth of additional information sources other than the rating (or purchase) history of users on items, such as item descriptions and social media information, further extensions of these systems have been proposed, known as hybrid recommendation algorithms. Hybrid recommenders use both user-item interaction data and their contextual information. In this work, we propose new hybrid recommender algorithms by considering the relationship between content features. This relationship is embedded into the hybrid recommenders to improve their accuracy. We first introduce a novel method to extract the content feature relationship matrix, and then the collaborative filtering recommender is modified such that this relationship matrix can be effectively integrated within the algorithm. The proposed algorithm can better deal with the cold-start problem than the state-of-art algorithms. We also propose a novel content-based hybrid recommender system. Our experiments on a benchmark movie dataset show that the proposed approach significantly improves the accuracy of the system, while resulting in satisfactory performance in terms of novelty and diversity of the recommendation lists.

Learning instrumental variable representation for debiasing in recommender systems.

Interaction-knowledge semantic alignment for recommendation

Recommender systems face longstanding challenges in gaining users’ trust due to the unreliable information caused by profile injection or human misbehavior. Traditional solutions to those challenges focus on leveraging users’ social relationships for inferring the user preference, i.e., recommending items according to the preference by user’s trusted friends; or adding random noise to the input to improve the robustness of the recommender systems. However, such approaches cannot defend the real-world noises like fake ratings. The recommender model is generally built upon all the user-item interactions, which incorporates the information from fake ratings or spammer groups, that neglects the reliability of the ratings. To address the above challenges, we propose an adversarial training approach in this work. In details, our approach includes two components: a predictor that infers the user preference; and a discriminator that enforces cohort rating patterns. In particular, the predictor applies an encoder-decoder structure to learn the shared latent information from sparse users’ ratings and trust relationships; the discriminator enforces the predictor to provide ratings as coherent with the cohort rating patterns. Our extensive experiments on three real-world datasets show the advantages of our approach over several competitive baselines.

The Recommender System (RS) has obtained a pivotal role in e-commerce. To improve the performance of RS, review text information has been extensively utilized. However, it is still a challenge for RS to extract the most informative feature from a tremendous amount of reviews. Another significant issue is the modeling of user–item interaction, which is rarely considered to capture high- and low-order interactions simultaneously. In this paper, we design a multi-level attention mechanism to learn the usefulness of reviews and the significance of words by Deep Neural Networks (DNN). In addition, we develop a hybrid prediction structure that integrates Factorization Machine (FM) and DNN to model low-order user–item interactions as in FM and capture the high-order interactions as in DNN. Based on these two designs, we build a Multi-level Attentional and Hybrid-prediction-based Recommender (MAHR) model for recommendation. Extensive experiments on Amazon and Yelp datasets showed that our approach provides more accurate recommendations than the state-of-the-art recommendation approaches. Furthermore, the verification experiments and explainability study, including the visualization of attention modules and the review-usefulness prediction test, also validated the reasonability of our multi-level attention mechanism and hybrid prediction.

A Social-aware Gaussian Pre-trained model for effective cold-start recommendation

Recommender models are hard to evaluate, particularly under offline setting. In this article, we provide a comprehensive and critical analysis of the data leakage issue in recommender system offline evaluation. Data leakage is caused by not observing global timeline in evaluating recommenders e.g., train/test data split does not follow global timeline. As a result, a model learns from the user-item interactions that are not expected to be available at the prediction time. We first show the temporal dynamics of user-item interactions along global timeline, then explain why data leakage exists for collaborative filtering models. Through carefully designed experiments, we show that all models indeed recommend future items that are not available at the time point of a test instance, as the result of data leakage. The experiments are conducted with four widely used baseline models—BPR, NeuMF, SASRec, and LightGCN, on four popular offline datasets—MovieLens-25M, Yelp, Amazon-music, and Amazon-electronic, adopting leave-last-one-out data split. 1 We further show that data leakage does impact models’ recommendation accuracy. Their relative performance orders thus become unpredictable with different amount of leaked future data in training. To evaluate recommendation systems in a realistic manner in offline setting, we propose a timeline scheme, which calls for a revisit of the recommendation model design.

In recent years, multidimensional data is becoming increasingly popular in recommender systems. For example, social tagging systems encourage users to employ user-defined keywords to help manage content in a personalized way. Recommender systems built upon social tagging systems utilize social tagging data to improve recommendation accuracy. Context-aware recommender systems incorporate context information (e.g. time, location, weather, etc.) into recommendation models to make recommendations more personalized. User profiles play a crucial role in recommender systems because user profiles provide the information about users’ information needs based on which personalized recommendations can be generated. Existing user profiling techniques for collaborative filtering (CF) recommender systems based on multidimensional data mostly analyze data through splitting the multidimensional relations into lower-dimensional relations. However, this leads to the loss of multidimensionality in user-item interactions. A different type of methods for addressing multidimensional data is factorization techniques such as tensor factorization (TF), which discover holistic latent relations between all dimensions in multidimensional data. However, TF models are not constructed to analyze any particular dimension of multidimensional data. But for many application domains, certain dimensions are of particular interest such as the user dimension in recommender systems. However, user profiles are not explicitly reflected in TF models. In this paper, we propose a unified approach to profile any particular dimension in multidimensional data based on partially-decomposed Tucker models (TM). We prove that the proposed profiling approach is a compatible higher-order extension of the matrix factorization based approaches. Moreover, we prove that the proposed TM-based profiles can be updated incrementally with new data without reconstructing the entire profiles. We integrate the proposed TF-based user/item profiling approach into neighborhood-based collaborative filtering recommenders for making top-N item recommendations. Extensive experiments have been conducted using real-world social tagging datasets to demonstrate that the CF recommenders integrated with the proposed user/item profiles outperform state-of-the-art CF recommendation approaches in terms of recommendation accuracy.

Prompt-based and weak-modality enhanced multimodal recommendation

Recommender Systems (RS) tend to recommend more popular items instead of the relevant long-tail items. Mitigating such popularity bias is crucial to ensure that less popular but relevant items are part of the recommendation list shown to the user. In this work, we study the phenomenon of popularity bias in session-based RS (SRS) obtained via deep learning (DL) models. We observe that DL models trained on the historical user-item interactions in session logs (having long-tailed item-click distributions) tend to amplify popularity bias. To understand the source of this bias amplification, we consider potential sources of bias at two distinct stages in the modeling process: i. the data-generation stage (user-item interactions captured as session logs), ii. the DL model training stage. We highlight that the popularity of an item has a causal effect on i. user-item interactions via conformity bias, as well as ii. item ranking from DL models via biased training process due to class (target item) imbalance. While most existing approaches in literature address only one of these effects, we consider a comprehensive causal inference framework that identifies and mitigates the effects at both stages. Through extensive empirical evaluation on simulated and real-world datasets, we show that our approach improves upon several strong baselines from literature for popularity bias and long-tailed classification. Ablation studies show the advantage of our comprehensive causal analysis to identify and handle bias in data generation as well as training stages.

Recommender systems are typically biased toward a small group of users, leading to severe unfairness in recommendation performance, i.e., User-Oriented Fairness (UOF) issue. Existing research on UOF exhibits notable limitations in two phases of recommendation models. In the training phase, current methods fail to tackle the root cause of the UOF issue, which lies in the unfair training process between advantaged and disadvantaged users. In the evaluation phase, the current UOF metric lacks the ability to comprehensively evaluate varying cases of unfairness. In this paper, we aim to address the aforementioned limitations and ensure recommendation models treat user groups of varying activity levels equally. In the training phase, we propose a novel Intra- and Inter-GrOup Optimal Transport framework (II-GOOT) to alleviate the data sparsity problem for disadvantaged users and narrow the training gap between advantaged and disadvantaged users. In the evaluation phase, we introduce a novel metric called ?-UOF, which enables the identification and assessment of various cases of UOF. This helps prevent recommendation models from leading to unfavorable fairness outcomes, where both advantaged and disadvantaged users experience subpar recommendation performance. We conduct extensive experiments on three real-world datasets based on four backbone recommendation models to prove the effectiveness of ?-UOF and the efficiency of our proposed II-GOOT.

To improve the experience of consumers, all social media, commerce and entertainment sites deploy Recommendation Systems (RSs) that aim to help users locate interesting content. These RSs are black-boxes – the way a chunk of information is filtered out and served to a user from a large information base is mostly opaque. No one except the parent company generally has access to the entire information required for auditing these systems – neither the details of the algorithm nor the user-item interactions are ever made publicly available for third-party auditors. Hence auditing RSs remains an important challenge, especially with the recent concerns about how RSs are affecting the views of the society at large with new technical jargons like “echo chambers”, “confirmation biases”, “filter bubbles” etc. in place. Many prior works have evaluated different properties of RSs such as diversity, novelty, etc. However, most of these have focused on evaluating static snapshots of RSs. Today, auditors are not only interested in these static evaluations on a snapshot of the system, but also interested in how these systems are affecting the society in course of time. In this work, we propose a novel network-centric framework which is not only able to quantify various static properties of RSs, but also is able to quantify dynamic properties such as how likely RSs are to lead to polarization or segregation of information among their users. We apply the framework to several popular movie RSs to demonstrate its utility.

Recommender systems play an essential role in guiding users through overwhelming content by predicting items of interest based on prior interactions. This thesis advances shallow collaborative filtering methods—particularly those characterized by simplicity, interpretability, and efficiency—by extending their capabilities in three key directions: interactivity, probabilistic modeling, and temporal awareness. First, we propose TEASER, a transparent and explainable recommendation model that constructs user profiles in the space of item metadata. By embedding users directly in an interpretable feature space, TEASER enables real-time interaction with recommendation explanations and adjustments, facilitating enhanced user control and understanding. Offline experiments and an interactive demo application demonstrate that allowing user feedback significantly boosts performance over static models. Second, we examine the limitations of classical implicit matrix factorization models such as Weighted Matrix Factorization (WMF), particularly their treatment of unknown interactions. We introduce a probabilistic reinterpretation of WMF that preserves its scalability and effectiveness while modeling unknowns as truly missing data rather than implicit negatives. This perspective leads to improved robustness and interpretability without sacrificing performance. Finally, we extend the modeling capacity of shallow recommendation models to incorporate context and time. We explore tensor decomposition techniques, introducing novel variants such as Weighted Tensor Factorization (WTF) and time-aware adaptations that support continuous-time modeling. These approaches avoid coarse time binning and instead embed time through smooth functions, allowing recommender systems to capture subtle temporal dynamics in user behavior. Through a combination of theoretical insights, methodological innovations, and empirical evaluations across multiple datasets, this work demonstrates that shallow collaborative filtering models can be substantially enriched while maintaining their inherent benefits. The proposed models bridge the gap between performance and transparency, offering practical tools for building trustworthy and interpretable recommender systems.

Multi-attention recommender system for non-fungible tokens