Search

Explainable recommendations aim to generate personalized explanations for suggested items, which are provided based on the historical interactions (e.g., ratings) between users and items. Review contents are often taken as a proxy of explanations. However, most review-based models presume a sentiment consistency between user ratings and review contents, ignoring their inconsistency in real applications. By analyzing three real datasets, we observe that a user may share a positive (negative) opinion to an item in terms of rating value but a negative (positive) sentiment in terms of review content, and such contradicting scenario takes over 40% of all cases in general. To resolve this issue, in this paper we propose a novel explainable recommendation model called PESI , which can generate accurate P ersonalized E xplanations recommendation with the involvement of S entiment I nconsistency between ratings and reviews. Specifically, PESI consists of three modules: rating prediction, explanation generation, and a novel rating-review inconsistency extraction. The inconsistency extraction module disentangles ratings and reviews, effectively distinguishing both shared and private features, and ensuring accurate disentanglement through contrastive learning objectives. Then, the extracted inconsistent features are injected into the explanation generation module to provide more personalized and higher-quality explanations. The experimental results on the three datasets show that PESI consistently outperforms other competing methods in terms of explanation quality. [ABSTRACT FROM AUTHOR]

Data sparsity is a well-recognized issue for Top-N item recommendation, which depends on user preference gathered from their historical behaviors (i.e., implicit feedback). However, only few works have considered multiple types of auxiliary implicit feedback (e.g, click, wanted) when building recommendation models. This paper aims to resolve the data sparsity problem by (a) generating target data (e.g., purchase) from a linear regression of auxiliary feedback, and from the nearest neighbors with a set of purchased items in multiple dimensions; (b) proposing a novel ranking model to accommodate both the original and generated data. We provide an intuitive comprehension regarding the relationship between one kind of auxiliary feedback and target feedback. A series of experiments are conducted on two real-world datasets and demonstrate the superiority of our approach to other counterparts. [ABSTRACT FROM AUTHOR]

Trust-aware recommender systems have attracted much attention recently due to the prevalence of social networks. However, most existing trust-based approaches are designed for the recommendation task of rating prediction . Only few trust-aware methods have attempted to recommend users an ordered list of interesting items, i.e., item recommendation . In this article, we propose three factored similarity models with the incorporation of social trust for item recommendation based on implicit user feedback. Specifically, we introduce a matrix factorization technique to recover user preferences between rated items and unrated ones in the light of both user-user and item-item similarities. In addition, we claim that social trust relationships also have an important impact on a user’s preference for a specific item. Experimental results on three real-world data sets demonstrate that our approach achieves superior ranking performance to other counterparts. [ABSTRACT FROM AUTHOR]

Recommender systems aim to suggest items of interest from historical interactions between users and items. Advanced methods such as path-based recommendations attempt to capture high-order user–item interactions for better recommendation performance. However, these methods focus more on the local view of user–item interactions, ignoring the global view, which limits performance. Two more drawbacks further restrict the performance of existing methods. First, most do not consider the relations (or relation directions) between successive nodes when constructing a multi-hop path (i.e., high-order local interactions) from a user to a target item. Second, the influential factors of path length and number of relation types are ignored when computing path importance to aggregate paths for better local representation. To resolve these issues, we propose a recommendation model that can well capture high-order local and global (HOLG) user–item interactions. A path-embedding module learns a local representation of user–item interactions via a long short-term memory network, taking (directed) relations of successive nodes as input. Multiple local representations are aggregated with an attention network, using both path length and number of relation types as important factors. A graph-embedding module learns a global representation of user–item interactions by constructing a subgraph from sampled user–item paths. Experiments on the LinkedIn, MovieLens-1M, and Yelp datasets validate our approach, which performs best in comparison with eight baselines. [ABSTRACT FROM AUTHOR]

Although demonstrated to be efficient and scalable to large-scale data sets, clustering-based recommender systems suffer from relatively low accuracy and coverage. To address these issues, we develop a multiview clustering method through which users are iteratively clustered from the views of both rating patterns and social trust relationships. To accommodate users who appear in two different clusters simultaneously, we employ a support vector regression model to determine a prediction for a given item, based on user-, item- and prediction-related features. To accommodate (cold) users who cannot be clustered due to insufficient data, we propose a probabilistic method to derive a prediction from the views of both ratings and trust relationships. Experimental results on three real-world data sets demonstrate that our approach can effectively improve both the accuracy and coverage of recommendations as well as in the cold start situation, moving clustering-based recommender systems closer towards practical use. [ABSTRACT FROM AUTHOR]

Abstract: Providing high quality recommendations is important for e-commerce systems to assist users in making effective selection decisions from a plethora of choices. Collaborative filtering is a widely accepted technique to generate recommendations based on the ratings of like-minded users. However, it suffers from several inherent issues such as data sparsity and cold start. To address these problems, we propose a novel method called “Merge” to incorporate social trust information (i.e., trusted neighbors explicitly specified by users) in providing recommendations. Specifically, ratings of a user’s trusted neighbors are merged to complement and represent the preferences of the user and to find other users with similar preferences (i.e., similar users). In addition, the quality of merged ratings is measured by the confidence considering the number of ratings and the ratio of conflicts between positive and negative opinions. Further, the rating confidence is incorporated into the computation of user similarity. The prediction for a given item is generated by aggregating the ratings of similar users. Experimental results based on three real-world data sets demonstrate that our method outperforms other counterparts both in terms of accuracy and coverage. [Copyright &y& Elsevier]

Personalized item recommendation has become an essential target of Web applications, but it suffers from the efficiency problem due to a large volume of data. In particular, feature-based factorization machine models are generally limited by the vast number of feature dimensions, leading to catastrophic computation time. In this paper, we propose a Fast Discrete Factorization Machine (FDFM) method to resolve these issues by applying the hash coding technologies to factorization machine models. Specifically, it discretizes the real-valued feature vectors in the parameter model during the process of learning personalized item rankings, whereby the overall computational time can be greatly reduced. Besides, we propose convergence update rules to optimize the quantization loss of the binarization problem, which can be used in personalized ranking scenarios efficiently. Based on the evaluation in two real-world datasets, our proposed approach consistently shows better performance than other baselines, especially when using shorter binary codes. [ABSTRACT FROM AUTHOR]

Recently, Generative Adversarial Networks (GANs) have received much attention in recommender systems because they can capture complex data distributions. They rely on two sub-networks, generator and discriminator models, to generate 'fake' yet reliable data, so better recommendation accuracy can be obtained. However, most variants of existing GANs achieve accuracy improvements by greatly increasing the model complexity, especially the amount of parameters, leading to difficulty in the deployment of those methods. Therefore, we aim to resolve this issue by learning compact yet accurate GANs that can make up valuable data with less generator parameters. To this end, we integrate knowledge distillation (KD, in the form of a teacher–student architecture) into GANs to reduce the model complexity while improving the accuracy of GANs. To the best of our knowledge, this work is the first to learn compact GANs for recommender systems by applying a KD framework named StuGAN. Specifically, we use a student discriminator to refine the teacher's knowledge, and then both the generator and discriminator are enhanced by leveraging the refined knowledge via adversarial learning, which constrains the generator to produce fake data approximating both the ground truth and teacher's predicted preferences, and enables the discriminator to distinguish between the preferences of the generator and more confusing preferences of the teacher from the ground truth. Finally, we conduct extensive experiments on two real-world datasets (i.e., Ciao and LastFM), and the results show that our approach can reduce the volume of model parameters by as much as half while maintaining comparable recommendation accuracy. [ABSTRACT FROM AUTHOR]