A Unified Probabilistic Matrix Factorization Recommendation Algorithm

Abstract

Rating information is usually used to calculate and predict in traditional recommendation systems. They can obtain the explicit characteristics of the users, but without implicit information and enough semantic interpretation, which affect recommendation results. To address the issue, this paper proposes a unified probabilistic matrix factorization recommendation algorithm fusing social tagging. The algorithm constructs user-resource rating matrix, user-tag tagging matrix, resources-tag correlation matrix and uses unified probabilistic matrix factorization to get the latent feature vectors of three matrices, to recommend for users by optimizing model parameter. The experimental results show that the proposed algorithm can effectively improve the quality of recommendation.