An Extension for Bounded-SVD — A Matrix Factorization Method with Bound Constraints for Recommender Systems

Abstract

In this paper, we introduce a new extension for bounded-SVD, i.e., a matrix factorization (MF) method with bound constraints for recommender system. In bounded-SVD, the bound constraints are included in the objective function so that not only the estimation errors are minimized but the constraints are also taken into account during the optimization process. Our previous results on major real-world recommender system datasets showed that bounded-SVD outperformed an existing MF method with bound constraints, BMF, and it is also faster and simpler to implement than BMF. However, an issue of bounded-SVD is that it does not take into account the bias effects in given data. In order to overcome this issue, we propose an extension of bounded-SVD: bounded-SVD bias. Bounded-SVD bias takes into account the rating biases of users and items - known to reside in recommender system data. The experiment results show that the bias extension can improve the performance of bounded-SVD in most cases.