关键词: 协同过滤, 信任关系, 矩阵分解, 二部图, 物质扩散

Abstract: Cold-start and data sparsity issues have still been two challenges in recommender systems. In most of traditional recommender systems based on the matrix factorization model, it is often assumed that users are isolated and the relationships among users are ignored, this results in the decrease in the recommendation effects. Thus, a novel approach incorporating social trust relationship and the structure of bipartite network is proposed. Based on the matrix factorization, this proposed approach combines the social trust relationships among users with the structure of bipartite network, and employs the gradient algorithm to train model parameters. The experimental results on Epinions data set show that the proposed approach is superior to other advanced approaches in accuracy and reliability, especially while the cold-start and data sparsity issues are involved in.

Key words: collaborative filtering, trust relationship, matrix factorization, bipartite network, mass diffusion