Uniform Evaluation of Recommender Algorithms for the Cold-Start Problem

Abstract

The cold-start problem is a problem that a recommender system becomes inaccurate for users who have little preference information for items or for items that have rarely been recommended to users. It is said that currently widely used recommender algorithms have difficulty addressing the cold-start problem. In recent years, researches have been conducted to address the cold-start problem. On the other hand, the scope of the cold-start problem is not clearly defined, and previous studies often differ to each other in their evaluation methods. In this paper, we compare several recommender algorithms addressing the cold-start problem with Matrix Factorization, a widely used recommender algorithm, under uniform conditions and settings. Through experiments, the characteristics of each method addressing the cold-start problem are identified from a uniform perspective.