Non-personalized Movie Recommendation by Maximum k-Coverage

Abstract

Turning first-time users into returning ones is a major task for the success of e-commerce systems. However, providing effective recommendations for these users remains as an open challenge for the area due to the absence of consumption information. In this context, non-personalized RSs emerge as the main approach adopted in real scenarios. Such approaches are based on the premise that the consumption is generally biased towards items that arouse interest in the majority of a population. Despite being valid for mass consumption, by adopting this premise RSs fail to help users interested in items different from the common taste. In this context, this work proposes a new RSs based on Maximum k-Coverage strategy to merge popular and non-popular items in order to retain different profiles of first-time users. The premise of this approach is that maximizing diversity, while maintaining the relevance of the recommended items, satisfies the preferences of different user profiles. Indeed, the results show a mean gain of 13.5% w.r.t. utility and diversity, when compared to traditional strategies based on Popularity, Best Rated and Recent Items. In addition, our results indicate that the proposed strategy is able to smooth the popularity bias in recommendations, satisfying at least 97% of different users.