Matrix Factorization for Collaborative Prediction

Netflix, an online video rental company, recently announced a contest to spur interest in building better recommendation systems. Users of Netflix are able to rank movies on an integer scale from 1 to 5. A rating of 1 indicates that the user “hated it”, while 5 indicates they “loved it”. The objective of a recommendation system, or collaborative filter, is to provide users with new recommendations based on past ratings they have made. There are several methods for approaching this problem, many of which have been extensivley documented such as k-nearest neighbors and mixture of multinomials. We are particularly interested in exploring different matrix factorization techniques, namely