Relational Topic Factorization for Link Prediction in Document Networks

Link prediction is one of the fundamental problems in complex networks. In this paper, we focus on link prediction in document networks, in which nodes are text documents. We propose the relational topic factorization model (RTF), a model that combines topic models and matrix factorization. We also develop an efficient Monte Carlo EM algorithm for learning the parameters. Empirical results show that our model outperforms other state-of-the-art ones, and can give better understanding of the documents.