Search

Heterogeneous Information Network (HIN) is a natural and general representation of data in recommender systems. Combining HIN and recommender systems can not only help model user behaviors but also make the recommendation results explainable by aligning the users/items with various types of entities in the network. Over the past few years, path-based reasoning models have shown great capacity in HIN-based recommendation. The basic idea of these models is to explore HIN with predefined path schemes. Despite their effectiveness, these models are often confronted with the following limitations: (1) Most prior path-based reasoning models only consider the influence of the predecessors on the subsequent nodes when modeling the sequences, and ignore the reciprocity between the nodes in a path; (2) The weights of nodes in the same path instance are usually assumed to be constant, whereas varied weights of nodes can bring more flexibility and lead to expressive modeling; (3) User-item interactions are noisy, but they are often indiscriminately exploited. To overcome the aforementioned issues, in this paper, we propose a novel path-based reasoning approach for recommendation over HIN. Concretely, we use a bidirectional LSTM to enable the two-way modeling of paths and capture the reciprocity between nodes. Then an attention mechanism is employed to learn the dynamical influence of nodes in different contexts. Finally, the adversarial regularization terms are imposed on the loss function of the model to mitigate the effects of noise and enhance HIN-based recommendation. Extensive experiments conducted on three public datasets show that our model outperforms the state-of-the-art baselines. The case study further demonstrates the feasibility of our model on the explainable recommendation task., 11 pages, 5 figures, Neurocomputing

Session-based recommendation targets next-item prediction by exploiting user behaviors within a short time period. Compared with other recommendation paradigms, session-based recommendation suffers more from the problem of data sparsity due to the very limited short-term interactions. Self-supervised learning, which can discover ground-truth samples from the raw data, holds vast potentials to tackle this problem. However, existing self-supervised recommendation models mainly rely on item/segment dropout to augment data, which are not fit for session-based recommendation because the dropout leads to sparser data, creating unserviceable self-supervision signals. In this paper, for informative session-based data augmentation, we combine self-supervised learning with co-training, and then develop a framework to enhance session-based recommendation. Technically, we first exploit the session-based graph to augment two views that exhibit the internal and external connectivities of sessions, and then we build two distinct graph encoders over the two views, which recursively leverage the different connectivity information to generate ground-truth samples to supervise each other by contrastive learning. In contrast to the dropout strategy, the proposed self-supervised graph co-training preserves the complete session information and fulfills genuine data augmentation. Extensive experiments on multiple benchmark datasets show that, session-based recommendation can be remarkably enhanced under the regime of self-supervised graph co-training, achieving the state-of-the-art performance., Comment: Accepted by CIKM'21

Social relations are often used to improve recommendation quality when user-item interaction data is sparse in recommender systems. Most existing social recommendation models exploit pairwise relations to mine potential user preferences. However, real-life interactions among users are very complicated and user relations can be high-order. Hypergraph provides a natural way to model complex high-order relations, while its potentials for improving social recommendation are under-explored. In this paper, we fill this gap and propose a multi-channel hypergraph convolutional network to enhance social recommendation by leveraging high-order user relations. Technically, each channel in the network encodes a hypergraph that depicts a common high-order user relation pattern via hypergraph convolution. By aggregating the embeddings learned through multiple channels, we obtain comprehensive user representations to generate recommendation results. However, the aggregation operation might also obscure the inherent characteristics of different types of high-order connectivity information. To compensate for the aggregating loss, we innovatively integrate self-supervised learning into the training of the hypergraph convolutional network to regain the connectivity information with hierarchical mutual information maximization. The experimental results on multiple real-world datasets show that the proposed model outperforms the SOTA methods, and the ablation study verifies the effectiveness of the multi-channel setting and the self-supervised task. The implementation of our model is available via https://github.com/Coder-Yu/RecQ., Comment: 11 pages, Accepted by WWW'21. Correct some typos in the previous version

With the prevalence of social media, there has recently been a proliferation of recommenders that shift their focus from individual modeling to group recommendation. Since the group preference is a mixture of various predilections from group members, the fundamental challenge of group recommendation is to model the correlations among members. Existing methods mostly adopt heuristic or attention-based preference aggregation strategies to synthesize group preferences. However, these models mainly focus on the pairwise connections of users and ignore the complex high-order interactions within and beyond groups. Besides, group recommendation suffers seriously from the problem of data sparsity due to severely sparse group-item interactions. In this paper, we propose a self-supervised hypergraph learning framework for group recommendation to achieve two goals: (1) capturing the intra- and inter-group interactions among users; (2) alleviating the data sparsity issue with the raw data itself. Technically, for (1), a hierarchical hypergraph convolutional network based on the user- and group-level hypergraphs is developed to model the complex tuplewise correlations among users within and beyond groups. For (2), we design a double-scale node dropout strategy to create self-supervision signals that can regularize user representations with different granularities against the sparsity issue. The experimental analysis on multiple benchmark datasets demonstrates the superiority of the proposed model and also elucidates the rationality of the hypergraph modeling and the double-scale self-supervision., Comment: 11 pages, 6 figures, CIKM 2021

Social recommender systems are expected to improve recommendation quality by incorporating social information when there is little user-item interaction data. However, recent reports from industry show that social recommender systems consistently fail in practice. According to the negative findings, the failure is attributed to: (1) A majority of users only have a very limited number of neighbors in social networks and can hardly benefit from social relations; (2) Social relations are noisy but they are indiscriminately used; (3) Social relations are assumed to be universally applicable to multiple scenarios while they are actually multi-faceted and show heterogeneous strengths in different scenarios. Most existing social recommendation models only consider the homophily in social networks and neglect these drawbacks. In this paper we propose a deep adversarial framework based on graph convolutional networks (GCN) to address these problems. Concretely, for (1) and (2), a GCN-based autoencoder is developed to augment the relation data by encoding high-order and complex connectivity patterns, and meanwhile is optimized subject to the constraint of reconstructing the social profile to guarantee the validity of the identified neighborhood. After obtaining enough purified social relations for each user, a GCN-based attentive social recommendation module is designed to address (3) by capturing the heterogeneous strengths of social relations. Finally, we adopt adversarial training to unify all the components by playing a Minimax game and ensure a coordinated effort to enhance recommendation performance. Extensive experiments on multiple open datasets demonstrate the superiority of our framework and the ablation study confirms the importance and effectiveness of each component., Accepted by TKDE

Traditional recommender systems often suffer from the problem of data sparsity, because most users rate only a few of the millions of possible items. With the development of social platforms, incorporating abundant social relationships into recommenders can help to overcome this issue, because users' preferences can be inferred from those of their friends. Most existing social recommenders are based on matrix factorization, a collaborative filtering model that has been proven to be effective. In this paper, we introduce a novel social recommender based on the idea that distance reflects likability. Compared with matrix factorization, the proposed model enables us to obtain a spatial understanding of the latent factor space and how users and items are positioned inside the space by combining the factorization model and distance metric learning. In our method, users and items are initially mapped into a unified low-dimensional space. The positions of users and items are jointly determined by ratings and social relations, which can help to determine appropriate locations for users who have few ratings. Finally, the learned metrics and positions are used to generate understandable and reliable recommendations. Experiments conducted on real-world data sets have shown that compared with methods based on only matrix factorization, our method significantly improves the recommendation accuracy.

Recommendation systems often fail to live up to expectations in real situations because of the lack of user feedback, known as the data sparsity problem. A large number of existing recommendation methods resort to side information to gain a performance improvement. However, these methods are either too complicated to follow or time-consuming. To alleviate the data sparsity problem, in this paper we propose UGAN, which is a general adversarial framework for recommendation tasks and consists of a generative model and a discriminative model. In UGAN, the generative model, acts as an attacker to cheat the discriminative model to capture the pattern of the original data input and generate similar user profiles, while the counterpart, the discriminative model aims to distinguish the forged samples from the real data. By competing with each other, two model are alternatively updated like playing a minimax game until the generative model has learned the original data distribution. The experimental results on two real-world datasets, Movielens and Douban, show that the user profiles forged by UGAN can be easily integrated into a wide range of recommendation methods and significantly improve their performance, which provides a promising way to mitigate the adverse impact of missing data.

The recommendation systems have been widely employed due to the effectiveness on mitigating the information overload issue. At present, the recommendation systems have made great progress, but they are under the threat of shilling attack because of their open nature. Shilling attack is the way by which the attackers can manipulate the recommendation results and cause great harm to recommendation systems. Existing shilling attack detection models are mainly based on statistical measures to extract features like the rating deviation, which are generally susceptible to attack strategies. Once the attacker changes attack strategy, the detection model which is based on the statistical method may fail. Some researchers have identified that implicit features hidden in user-user interactions and user-item interactions can be utilized to solve the problem. Their research aims to learn potential relationship between users to update features. However, the research ignores the significance of learning features by employing label information. To solve this problem, in this paper, we propose a novel detection model, named BayesDetector, which takes not only the user-user and user-item interactions but also the label information into consideration in the process of learning user implicit features. Furthermore, to take full advantage of user labels, the Bayesian model is added to the feature learning. Experiments on two datasets, Amazon and Movielens, show that BayesDetector significantly outperforms the state-of-the-art methods.

Social recommendation has attracted increasing attention over the years due to the potential value of social relations, which can be harnessed to mitigate the dilemma of data sparsity in traditional recommender systems. However, recent studies show that social recommenders fail in the practical use in industry for the reason that some problems in social relations, such as the noise, lead to a degradation in recommendation quality. To solve the problem, in this paper, a social recommender, SocialEM, which integrates the neural user embedding and collaborative filtering is proposed. Enlightened by the factorization of the word co-occurrence matrix which is equivalent to the skip-gram model in word2vec, SocialEM will jointly decomposes the user-item rating matrix and the user-user co-occurrence matrix with shared user latent factors. For each pair of users, the co-occurrence matrix encodes the number of being trusted together by other users in the social relation network. Experiments conducted on the real-world datasets have shown that the side effect of social relations can be diminished by tuning parameters for SocialEM. And compared with previous studies, our method significantly improves the quality of recommendations.

© 2018, ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering. In social network, people generally tend to share information with others, thus, those who have frequent access to the social network are more likely to be affected by the interest and opinions of other people. This characteristic is exploited by spammers, who spread spam information in network to disturb normal users for interest motives seriously. Numerous notable studies have been done to detect social spammers, and these methods can be categorized into three types: unsupervised, supervised and semi-supervised methods. While the performance of supervised and semi-supervised methods is superior in terms of detection accuracy, these methods usually suffer from the dilemma of imbalanced data since the number of unlabeled normal users is far more than spammers’ in real situations. To address the problem, we propose a novel method only relying on normal users to detect spammers exactly. We present two steps: one picks out reliable spammers from unlabeled samples which is imposed on a voting classifier; while the other trains a random forest detector from the normal users and reliable spammers. We conduct experiments on two real-world social datasets and show that our method outperforms other supervised methods.

The recommender system based on collaborative filtering is vulnerable to shilling attacks due to its open nature. With the wide employment of recommender systems, an increasing number of attackers are disordering the system in order to benefit from the manipulated recommendation results. Therefore, how to effectively detect shilling attacks now becomes more and more crucial. Most existing detection models recognize attackers in statistics-based manners. However, they failed in capturing the fine-grained interactions between users and items, leading to a degradation in detection accuracy. In this paper, inspired by the success of word embedding models, we propose a collaborative shilling detection model, CoDetector, which jointly decomposes the user-item interaction matrix and the user-user co-occurrence matrix with shared user latent factors. Then, the learned user latent factors containing network embedding information are used as features to detect attackers. Experiments conducted on simulated and real-world datasets show that CoDetector has a good performance and generalization capacity and outperforms state-of-the-art methods.

Social relations can help to relieve the dilemmas called cold start and data sparsity in traditional recommender systems. Most of existing social recommendation methods are based on matrix factorization, which has been proven effective. In this paper, we introduce a novel social recommender based on the idea that distance reflects likability. It aims to make users in recommender systems be spatially close to their friends and items they like, and be far away from items they dislike by connecting factorization model and distance metric learning. In our method, the positions of users and items are decided by the ratings and social relations jointly, which can help to find appropriate locations for users who have few ratings. Finally, the learnt metric and locations are used to generate understandable and reliable recommendations. The experiments conducted on the real-world dataset have shown that, compared with methods only based on factorization, our method has advantages on both interpretability and accuracy.

© Springer International Publishing AG 2017. Social networks act as the communication channels for people to share various information online. However, spammers who generate spam information reduce the satisfaction of common users. Numerous notable studies have been done to detect social spammers, and these methods can be categorized into three types: unsupervised, supervised and semi-supervised methods. While the performance of supervised and semi-supervised methods is superior in terms of detection accuracy, these methods usually suffer from the dilemma of imbalanced data since the labeled normal users are far more than spammers in real situations. To address the problem, we propose a novel method only relying on normal users to detect spammers. Firstly, a classifier is built from a part of normal and unlabeled samples to pick out reliable spammers from unlabeled samples. Secondly, our well-trained detector, which is based on the given normal users and predicted spammers, can distinguish between normal users and spammers. Experiments conducted on real-world datasets show that the proposed method is competitive with supervised methods.

Recommender systems can help to relieve the dilemma called information overload. Collaborative filtering is a primary approach based on collective historical ratings to recommend items to users. One of the most competitive collaborative filtering algorithm is matrix factorization. In this paper, we proposed an alternative method. It aims to make users be spatially close to items they like and be far away from items they dislike, by connecting matrix factorization and distance metric learning. The metric and latent factors are trained simultaneously and then used to generate reliable recommendations. The experiments conducted on the real-world datasets have shown that, compared with methods only based on factorization, our method has advantage in terms of accuracy.

Recommender systems (RSs) now play a very important role in the online lives of people as they serve as personalized filters for users to find relevant items from an array of options. Owing to their effectiveness, RSs have been widely employed in consumer-oriented e-commerce platforms. However, despite their empirical successes, these systems still suffer from two limitations: data noise and data sparsity. In recent years, generative adversarial networks (GANs) have garnered increased interest in many fields, owing to their strong capacity to learn complex real data distributions; their abilities to enhance RSs by tackling the challenges these systems exhibit have also been demonstrated in numerous studies. In general, two lines of research have been conducted, and their common ideas can be summarized as follows: (1) for the data noise issue, adversarial perturbations and adversarial sampling-based training often serve as a solution; (2) for the data sparsity issue, data augmentation--implemented by capturing the distribution of real data under the minimax framework--is the primary coping strategy. To gain a comprehensive understanding of these research efforts, we review the corresponding studies and models, organizing them from a problem-driven perspective. More specifically, we propose a taxonomy of these models, along with their detailed descriptions and advantages. Finally, we elaborate on several open issues and current trends in GAN-based RSs., 24 pages, 12 figures, 8 tables, the journal of Information Sciences

To explore the robustness of recommender systems, researchers have proposed various shilling attack models and analyzed their adverse effects. Primitive attacks are highly feasible but less effective due to simplistic handcrafted rules, while upgraded attacks are more powerful but costly and difficult to deploy because they require more knowledge from recommendations. In this paper, we explore a novel shilling attack called Graph cOnvolution-based generative shilling ATtack (GOAT) to balance the attacks' feasibility and effectiveness. GOAT adopts the primitive attacks' paradigm that assigns items for fake users by sampling and the upgraded attacks' paradigm that generates fake ratings by a deep learning-based model. It deploys a generative adversarial network (GAN) that learns the real rating distribution to generate fake ratings. Additionally, the generator combines a tailored graph convolution structure that leverages the correlations between co-rated items to smoothen the fake ratings and enhance their authenticity. The extensive experiments on two public datasets evaluate GOAT's performance from multiple perspectives. Our study of the GOAT demonstrates technical feasibility for building a more powerful and intelligent attack model with a much-reduced cost, enables analysis the threat of such an attack and guides for investigating necessary prevention measures., Comment: 16 pages, 21 figures, Information Sciences - Journal - Elsevier