Optimal Ranking for Video Recommendation

Item recommendation from implicit feedback is the task of predicting a personalized ranking on a set of items (e.g. movies, prod- ucts, video clips) from user feedback like clicks or product purchases. We evaluate the performance of a matrix factorization model optimized for the new ranking criterion BPR-Opt on data from a BBC video web application. The experimental results indicate that our approach is su- perior to state-of-the-art models not directly optimized for personalized ranking.