Relational Collaborative Topic Regression for Recommender Systems

摘要: Due to its successful application in recommender systems, collaborative filtering (CF) has become a hot research topic data mining and information retrieval. In traditional CF methods, only the feedback matrix, which contains either explicit (also called ratings) or implicit on items given by users, is used for training prediction. Typically, matrix sparse, means that most users interact with few items. this sparsity problem, will suffer from unsatisfactory performance. Recently, many researchers have proposed utilize auxiliary information, such as item content (attributes), alleviate problem CF. Collaborative regression (CTR) one of these methods achieved promising performance successfully integrating both information. real applications, besides there may exist relations known networks) among can be helpful recommendation. paper, we develop novel hierarchical Bayesian model Relational Topic Regression (RCTR), extends CTR seamlessly user-item network structure into same model. Experiments real-world datasets show our achieve better prediction accuracy than state-of-the-art lower empirical time. Moreover, RCTR learn good interpretable latent structures are useful