A Collaborative Filtering Approach Based on User´s Reviews

This paper proposes a collaborative filtering approach that uses users´ reviews to produce item descriptions that represent a consensus of users regarding items´ features. While earlier works focused on using structured metadata to represent items, recent approaches study how to use user-provided text, such as reviews, to produce better insights about the semantics in the content. Some involved problems, such as noise, personal opinions and false information are reduced by an algorithm based on sentiment analysis and natural language processing. We provide an evaluation using the MovieLens dataset, and the results are promising when compared to recommenders based only on structured metadata.