A Hybrid Recommender System Combing Singular Value Decomposition and Linear Mixed Model

Tianyu Zuo, Shengxin Zhu, Jian Lu

Abstract: We explain the basic idea of linear mixed model (LMM), including parameter estimation and model selection criteria. Moreover, the algorithm of singular value decomposition (SVD) is also mentioned. After introducing the related R packages to implement these two models, we compare the mean absolute errors of LMM and SVD when different numbers of historical ratings are given. Then a hybrid recommender system is developed to combine these two models together. Such system is proven to have higher accuracy than single LMM and SVD model. And it might have practical value in different fields in the future.