Recommendation Algorithm Fusing Implicit Similarity of Users and Trust

Abstract

The goal of the recommender system is to understand the user's preferences and make appropriate recommendations for certain behaviors. Similarity is an important part of recommendation algorithm. However, most current recommendation algorithms ignore the social relation on users, and the most similarity calculations are based on the common object, which does not fully exploit the implicit information. For these challenges, firstly, an adaptive approach which takes both trust and rating into account is proposed to measure the implicit similarity between a user and his friends when the user as a truster and a trustee, respectively. Secondly, a recommendation algorithm combining trust and ratings is proposed, which also takes the biases of users and items, and the user's implicit feedback into account to enhance the accuracy of the algorithm. Finally, the social regularization recommendation algorithm is modified. The experimental results on two different datasets show that the proposed algorithm is able to not only outperform the traditional recommendation algorithm, but also effectively alleviate the data sparsity and cold start problems.