Abstract: In the recommend systems, modeling resource is a vital issue. Based on the study of the traditional user-based resource models, we find that it is always assumed that users are independent of each other, trust relations among the users are not fully used, leading to strong Matthew effect, cold start problem and so on. This paper designs a common user trust network based on the tag co-occurrence, upon which the PageRank algorithm is used to refine the resource model. Further, an improved diffusion process is performed to get the recommend results. Compared with the previous algorithms, experimental results show that our algorithm significantly improves accuracy， recall and F1-measure of recommendations.

Key words: resource model, trust network, tag, mass diffusion, recommender system