Preference Networks: Probabilistic Models for Recommendation Systems

Recommender systems are important to help users se-lect relevant and personalised information over massive amounts of data available. We propose an unified frame-work called Preference Network (PN) that jointly models various types of domain knowledge for the task of recom-mendation. The PN is a probabilistic model that system-atically combines both content-based filtering and collab-orative filtering into a single conditional Markov random field. Once estimated, it serves as a probabilistic database that supports various useful queries such as rating pre-diction and top-N recommendation. To handle the chal-lenging problem of learning large networks of users and items, we employ a simple but effective pseudo-likelihood with regularisation. Experiments on the movie rating data demonstrate the merits of the PN