A Comparative Analysis of Trust-Enhanced Recommenders for Controversial Items

Abstract

A particularly challenging task for recommender systems (RSs) is deciding whether to recommend an item that received a variety of high and low scores from its users. RSs that incorporate a trust network among their users have the potential to make more personalized recommendations for such controversial items (CIs) compared to collaborative filtering (CF) based systems, provided they succeed in utilizing the trust information to their advantage. In this paper, we formalize the concept of CIs in RSs. We then compare the performance of several well-known trust-enhanced techniques for effectively personalizing the recommendations for CIs versus random items in the RS. Furthermore, we introduce a new algorithm that maximizes the synergy between CF and its trust-based variants, and show that the new algorithm outperforms other trust-based techniques in generating rating predictions for CIs.