Collaborative Recommender Systems Based on User-Generated Reviews: A Concise Survey

Abstract

Recommender systems are powerful tools that help users to deal with information overload problem. Collaborative Filtering (CF) approach has been widely used to build recommender systems over the past decades. However, the performance of CF is limited by sparsity and cold start problems, which are very common in real world situations. In recent years, many review-based approaches have been developed to integrate textual reviews into recommendation process, since they provide much more information about item/user profiles than ratings. The use of text analysis and opinion mining methods helps extracting such information. In this paper, we first introduce standard CF techniques and their main challenges. Then, we present different kind of information that can be extracted from user reviews. After that, we describe recent works that exploit review elements to improve the CF-based recommendations. Finally, we discuss their practical implications.