Alleviating Sparsity Problem of Recommender System with No Extra Input Data

Abstract

Recommender systems is one of the major technical research tracks in the internet and big data era. However, in many small or medium-sized corporations that lack big data, sparsity is a frequently encountered problem in real world context settings. The sparsity problem also poses problems to recommender system designers during the initial stage after the system is put online. In spite of the criticality of the sparsity problem, the research and investigation on sparsity problem has been slim in the research community. In this paper, we propose to use ZeroMat (invented in 2021) as the preprocessing step before recommender system execution to solve the sparsity problem. We prove by experiments our hybrid method outperforms single models such as ZeroMat and the classic version of Matrix Factorization.