Justified group recommender systems

Abstract

Justification improves the reliability of a recommender system because it helps user/s understand the reasoning behind the recommendation. Nearest neighbor style and influence style are the common justification styles in a recommender system. Since both styles are constructed exclusively in light of user preferences on the item rather than content of an item, the recommendation cannot be adequately justified. Moreover, these justification styles are applicable for personal recommender systems rather than group recommender systems. In this paper, we introduce a novel justification style for group recommender systems having the structure “item x is recommended because those who watch y, z,..that contain features {gi,gj,…} also watch x that contains {gk,gl,…}”. Our justification style is based on precedence mining model, wherein the precedence probability of using an item by an active user is determined based on pairwise precedence relations between the items. We broaden this idea of precedence probability to accommodate the social influence factor. No past investigation deals with justified group recommender systems.