Semantic Organization of User's Reviews Applied in Recommender Systems

Abstract

Recommender systems are widely used to minimize the information overload problem. A great source of information is users' reviews, since they provide both item descriptions and users' opinions. Recent works that process reviews often neglect problems such as polysemy and sinonimy. On the other hand, systems that rely on word sense disambiguation focus their efforts on items's static descriptions. In this paper, we propose a hybrid recommender system that uses word sense disambiguation and entity linking to produce concept-based item representations extracted from users' reviews. Our findings suggest that adding such semantics to items' representations have a positive impact on recommendations.