On Alleviation of New User Problem in Collaborative Filtering using SNA Theory

Abstract

Collaborative filtering is the most used personalized recommendation technology. However, the traditional collaborative filtering faces the cold start problem and data sparsity, which deteriorates user experience and reduces the prediction accuracy. This paper presents a novel solution of new user problem with social network analysis (SNA) theory. First, the user relationship network is built based on the user-item rating matrix. Then all users will be divided into many different cliques according to SNA. The demographic user information of new user is used to find similar users in user attributes. Further, the candidate neighbors of new user can be obtained from cliques. Then, the preferential attachment characteristic of the free-scale network is introduced to eliminating the unsuitable neighbors. Finally, items are recommended with user-based nearest neighbor recommendation algorithm. The experiments show that the proposed approach alleviates the new user problem and improves the prediction accuracy effectively compared with other algorithms.