User preference modeling by trust propagation for rating prediction

To alleviate the issue of data sparsity in collaborative filtering (CF), a number of trust-aware recommendation methods have been proposed recently. However, the existing methods that straightforwardly utilize trust relations to model user similarities in ratings or preference features can hardly provide the in-depth understanding of the trust and its relationship to user preference. They also fail to systematically model the mutual influence among users via the truster-user-trustee propagation. In this paper, we propose a novel integrated matrix factorization framework to model user preference, trust relation and the relationship between them in a systematic way. The proposed framework is able to describe how and how much users' preferences change and influence each other with trust propagation over the network. As a result, more effective user preference features can be learned from both rating and trust data. Experimental results on three real-world datasets show that our proposed methods outperform the state-of-theart CF and trust-aware methods.