Matrix Factorization and Regression-Based Approach for Multi-Criteria Recommender System

Recommender systems (RS) try to solve information overload problem by providing the most relevant items to users from a large set of items. Collaborative filtering (CF), a popular approach in building RS, generates recommendations to users based on explicit ratings provided by the community of users. Currently many online platforms allow users to evaluate items based on multiple criteria along with an overall rating instead of single overall rating. Previous research work has shown that considering these multiple criteria ratings for recommendations improved the predictive accuracy of recommender systems. In this paper, we propose a novel approach to increase predictive accuracy of multi-criteria recommender systems (MCRS). Firstly, we use matrix factorization to predict individual criteria ratings and then compute weights of individual criteria ratings through linear regression. Finally we predict overall rating using a weighted function of multiple criteria ratings. Through experiments on Yahoo! Movies dataset, we compare our proposed approach to baseline approaches and demonstrate its effectiveness in terms of predictive accuracy measures.