Gur, Yuval Dan, Sheizaf Rafaeli

Enhancing user control over online recommendation processes: 'Friends' vs. "Neighbors" in the "Qsia" Recommender System

Abstract

Online Recommender Systems provide recommendations based partially on other users' rankings (or 'preferences'). A core fundamental of social-filtering-based Recommender Systems is the 'neighbors' group - the population whose tastes and preferences are close to the user's. Though taste-related recommendation is partly a social process, a preponderance of the work so far in this field has focused on the statistical formation of the "neighbors" group. Modeling ('matching' algorithms) generally neglect the social perspective. Further, Recommender Systems remained relevant only to 'low-risk' domains, acting like computerized oracles: Black boxes supplying recommendations upon demand, with inherent low user's involvement. In this research we propose motivated user involvement in the recommendation process, specifically in the formation of the 'neighbors' group. We hypothesize that such involvement will enhance user acceptance, commitment and evaluation of the output recommendations. The notion of a user controlled 'friends' group is introduced. We use the term 'friends' which is not solely rank-dependent, instead of 'neighbors' . In accordance with 'Social Comparison Theory' and derived behavioral experiments, we suggest that 'neighbors' (like-minded groups) are relevant to 'low-risk' domains but 'friends' (similar and wider personal characteristics) may be relevant to higher risk domains. The research intends to compare performance indices of 'Qsia' (a newly-constructed online collaborative system for knowledge items) in two modes: one with automatic-ranked 'neighbors' against those of enabled user-controlled 'friends' group. Future implications might be reflected in issues of online social interaction, trust, privacy, matching algorithms, user overload and information value.