Improving Social Recommendation By Fusing Distrust Information

With the rapid development of Internet, applications of recommender systems are being widely used. As the core component of recommendation system, recommendation algorithm determines the types of recommended goods of obtaining of the users. The performance of the recommendation algorithm directly affects the consumption experience of the users and economic benefits of the business. Traditional Collaborative Filtering has been one of the most widely used recommendation algorithms, unfortunately it suffers from cold-start and data sparsity problems.With the development of social networks, more recommendation systems are trying to generate more eligible recommendation through excavating users’ potential preferences using their social relationships. Almost all social recommender systems employ only positive inter-user relations such as friendship or trust information. However, incorporating negative relations in recommendation has not been investigated thoroughly in literature. In this paper, we propose a novel model-based method which takes advantage of both positive and negative inter-user relations. We apply matrix factorization techniques and utilize both rating and trust information to learn users’ reasonable latent preference. Then we incorporate two regularization terms to take distrust information into consideration, such that improving the precision of recommendation.Recommending friends and items are equally important in social recommendation applications. Existing works usually consider them as two independent tasks and address them separately. This work presents a new method to collaboratively perform the two tasks, in which three types of sparse data, ratings, trust and distrust, are completed with each other by a coupled low-rank approximation model. The data sparsity problems suffered from by respective tasks, which greatly decline prediction quality, are well mitigated in this way. The experimental validation on three real data sets shows that the performance of the two presented models are more outstanding, which also support theoretical models and analysis.