The Oblivion Problem: Exploiting forgotten items to improve recommendation diversity
Fernando Mourão, Claudiane Fonseca, Camila Souza Araújo, Wagner Meira · Conference on Recommender Systems · 2011
Recommender Systems (RSs) have become a crucial tool to assist users in their choices on various commercial applications. Despite recent advances, there is still room for more effective tech niques that are applicable to a larger range of domains. A major challenge recurrently researched is the lack of diversity in the recom mendation lists provided by current RSs. That is, besides being effective to suggest interesting items to users, a good RS should provide useful and diversified items. In order to address this proble m, we evaluate the use of forgotten items in recommendation. By forgotten items, we mean items that have been very relevant to users in the past but are not anymore. Therefore, we formally define th e Oblivion Problem, which is the problem of recommending forgotten items, propose a methodology for verifying it in real scenarios, and perform a deep characterization of this problem in a relevant music domain, the Last.fm system. Applying our methodology to Last.fm has demonstrated the existence of the oblivion problem in practice, as well as showed the utility of this methodology. Further, the behavior exhibited by forgotten items in Last.fm suggests that defining techniques that incorporate such items into RSs con sists in a promising research direction.