Research On Recommendation Algorithm Integrated With Social Trust

In recent years, the rapid development of Internet and E-commerce has brought convenience to people in almost every respect. But the explosive growth of information on Internet results in the Information Overload, making it hard for users to access their required information. Personalized recommendation technology provide users with services or items they may be interested in according to users’ information such as personal characteristics and historical behaviors. The researching of personalized recommendation has been a hot point of data mining and social networks, et al.Collaborative filtering is the most mature and widely used personalized recommendation algorithm, but it also has some problems such as data sparsity, cold start, scalability and so on, which needs to be improved. So in this paper, we focus on resolving the problem of data sparsity based on users’ rating data and social network information, introduce a set of new measures for user influence and social trust and propose two recommendation algorithms: a social recommendation algorithm combining trust and user influence and a novel recommendation algorithm based on matrix factorization combining credible correlation.The main research works of this paper are as follows:Firstly, traditional collaborative filtering only use the user-item rating data. It may has a low performance in terms of accuracy when the rating data is sparse. In this study, we introduce a set of new measures for social trust and user influence, then propose an integrated approach for recommendation combing users’ social information. The advantage of the proposed algorithm is taking into account of social information and rating information. It can improve recommendation accuracy while enhancing the stability of collaborative filtering.Secondly, model-based collaborative filtering has some problems such as the sparsity of social relationships and recommendation accuracy. This paper proposes a new model named concentric circle model considering indirect trust relationship, fusing more trust information. Also, we propose the concept credible correlation which considers trust and interest and improves matrix factorization’s learning ability by using credible correlation.Finally, our experiments were performed on the Dianping datasets. The results show that the proposed recommendation algorithm combining trust and user influence outperforms traditional collaborative filtering in terms of recommendation accuracy. Meanwhile, the algorithm based on matrix factorization combining credible correlation outperforms the basic matrix factorization and Social Matrix Factorization(Social MF) in terms of recommendation accuracy and stability.