A probabilistic graph-based method to solve precision-diversity dilemma in recommender systems

Abstract Over the last few years, with significant growth of information on the web, users are swamped with a huge amount of information. Recommender system (RS) is an information retrieval technology that aims to provide relevant items to users by considering their preferences. Although several studies in the literature have proposed new RSs to improve the precision of recommendations, some studies have shown that increasing the diversity of recommendations is one way to increase the quality of the user’s experience and to solve the over-fitting problem. While some previous works have tried to make a trade-off between precision and diversity, they could not provide a linear and deterministic trade-off between precision and diversity. As such, these methods are either impractical or infeasible in a real RS. In this paper, we propose a new method to recommend items which are fit to users’ preferences. The method consists of following steps: i) finding users’ preferences using a limited user study, ii) proposing a probabilistic graph-based recommender system (PGBRS) to recommend items with a desirable level of precision and diversity by considering items that are likely to be preferred by the user in the future. PGBRS is a dataset-independent method which is significantly flexible to update with a low implementation complexity.