Research On Collaborative Filtering Recommendation Algorithms Based On Social Trust Network

Social network acts as a crucial medium for people to share and communicate their knowledge. Based on the social network, recommender systems can analyze users’ preferences in history and further recommend some interesting items to users, which can address the problems of "information overload" and neglecting users’preferences in traditional information retrieval, and meet personalized demands of different users. The core of the recommender system is recommendation algorithm. Collaborative filtering is currently one of the most widely used recommendation algorithms. However, traditional collaborative filtering recommendation algorithm meets the challenges of data sparsity, cold start and easy attack. In reality, users often tend to admit the recommendations made by their trustworthy friends. So, social trust network is applied to improve traditional collaborative filtering recommendation algorithm, which is named trust-based collaborative filtering recommendation algorithm. According to the difference sources of trust information, trust-based collaborative filtering is classified to explicit and implicit trust-based collaborative filtering respectively. Explicit trust information is the information of trust relationships that users manually build with their neighbors. Implicit trust information is the information of inferred trust relationships based on users’behaviors. Nevertheless, when explicit trust information is obtained, expressing trust statements is usually time-consuming for users and exposing users’privacy easily happens, which result in insufficient explicit trust information available and limited recommendations for explicit trust-based collaborative filtering. The existing related studies on implicit trust consider the little factors about users’behaviors and trust propagation, which lead to inferred trust relationships far away from actual trust relationships and inaccurate recommendations for implicit trust-based collaborative filtering.From the aspect of explicit trust, to alleviate the problems of data sparsity, cold start and easy attack in collaborative filtering and insufficient trust information available in explicit trust-based collaborative filtering, a latent social trust network model for collaborative filtering recommendation algorithm is proposed. The latent social trust comes from the coupling trust and the cocitation trust as well as the trust relying on the similar interests between users. Based on the latent trust information, a new social trust network can be built to search for the trustworthy neighbors and then be used to predict the target user’s preference. The experimental results demonstrate that our approach can rationally infer the trust relationships between users and highly improve the recommendation accuracy.From the aspect of implicit trust, to address the issues of data sparsity in collaborative filtering and inaccurate inferred implicit trust relationships in implicit trust-based collaborative filtering, and to avoid the drawbacks of explicit trust information, an implicit trust-based collaborative filtering recommendation algorithm by users’behavior analysis is proposed, which builds three kinds of implicit trust recommendation models, i.e., user-trust-neighbor, user-trust-item and hybrid models, by analyzing users’rating behaviors. According to neighbors’and users’historical preferences, some interesting items can be recommended to the target user. The experimental results demonstrate that our proposed method can provide users with more accurate recommendations.