An Improvement of Matrix Factorization with Bound Constraints for Recommender Systems

Abstract

This paper presents an improvement of bounded-SVD bias - a matrix factorization (MF) method for recommender systems. In bounded-SVD bias, the bound constraints are included in the objective function, which ensures that they are satisfied during the optimization process. As shown in one of our previous work, this helps bounded-SVD bias outperform an existing MF method with bound constraints, called Bounded Matrix Factorization. However, there is one issue with bounded-SVD bias that it is prone to overfitting. In this work, we introduce a new term to the objective function of bounded-SVD bias that can help it avoid overfitting. We also perform various experiments using real-world datasets, the results of which show an improvement in terms of accuracy and thus the superiority of the proposed method.