A Robust Item Recommendation Technique Based on Message Passing
Soonhyoung Kwon, Sang‐Chul Lee, Sang‐Wook Kim · 2012
Due to the rapid growth of e-commerce, various types of items are being sold online these days. For sales promotion, sellers want to find out which items need to be exposed to a target user so that she/he is willing to buy some of those items. The recommender system identifies and recommends those items to a user by analyzing the user's information such as profiles and transactions. Some venders, however, try to attack the system by employing a large number of abusers in order for the system to recommend their own items to users. This paper proposes a recommendation system that is robust against such abuser attacks. The proposed method models users, items, and their relationships as a bipartite graph, and employs message passing performed over the graph, which weakens the influence of the abusers' attacks. We verify the robustness of our method through extensive experiments by comparing it with the existing recommendation methods in terms of accuracy.