Double-Sided Recommendations: A Novel Framework for Recommender Systems

Recommender systems actively provide users with suggestions of potentially relevant items. In this paper we introduce double-sided recommendations, i.e., recommendations consisting of an item and a group of people with whom such an item could be consumed. We identify four specific instances of the double-sided recommendation problem and propose a general method for solving each of them (social comparison-based, group-priority, item-priority and same-priority methods), thus defining a framework for generating double-sided recommendations. We present the experimental evaluation we carried out, focusing on the restaurant domain as a use case, with the twofold aim of 1) assessing user liking for double-sided recommendations and 2) comparing the four proposed methods, testing our hypothesis that their perceived usefulness varies according to the specific problem instance users are facing. Our results show that users appreciate double-sided recommendations and that all four methods -and, in particular, the group-priority one- can generate useful suggestions.