A Modified Regularized Non-Negative Matrix Factorization for MovieLens

Abstract

This paper studies the matrix factorization technique for recommendation systems. The problem is to modify and apply non-negative matrix factorization to predict a rating that a user is likely to rate for an item in MovieLens dataset. First, based on the original randomize non-negative matrix factorization, we propose a new algorithm that discovers the features underlying the interactions between users and items. Then, in the experimentation section, we provide the numerical results of our proposed algorithms performed on the well-known MovieLens dataset. Besides, we suggest the optimization parameters which should be applied for Matrix Factorization to get good results on MovieLens. Comparison with other recent techniques in the literature shows that our algorithm is not only able to get high quality solutions but it also works well in the sparse rating domains.