Matrix Factorization Meets Social Network Embedding for Rating Prediction

Social recommendation becomes a current research focus, which leverages social relations among users to alleviate data sparsity and cold-start problems in recommender systems. The social recommendation methods usually employ simple similarity information as social regularization on users. Unfortunately, the widely used social regularization cannot make a good analysis of the users' social relation characteristics. In order to overcome the shortcomings of social recommendations, we propose a new framework for which combines network embedding and probabilistic matrix factorization. We make use of social relation features extracted from social networks, on top of which we learn an additional layer that uncovers the social dimensions that explain the variation in people's feedback. Furthermore, the influence of different social network embedding strategies on our framework are compared. Experiments on three real datasets validate the effectiveness of the proposed solution.