Natural Language Processing for Recommender Systems

Oren Sar Shalom, Haggai Roitman, Pigi Kouki

Abstract: In the race for improved modeling, modern recommender systems enhance collaborative filtering (CF) by using diverse signals that provide additional information on users’ preferences or items’ traits. Arguably, the most meaningful signal for recommenders is textual data, which includes examples like user-generated reviews, textual-item descriptions and even conversational interaction in natural language. Additionally, the output of a typical recommender may include free-form text as well, when auto generated explanations are associated with the suggested items. In this chapter, we describe cases where Natural Language Processing (NLP) can aid recommender systems. We first identify the possible tangent points between NLP and recommenders. Next, we present systems that successfully exploit the interaction between these two fields. Finally, for each such case we indicate its relative advantages and limitations.