Defending shilling attacks in recommender systems using soft co-clustering

Shilling attacks have been a significant vulnerability to collaborative filtering based recommender systems recently. There are various studies focusing on detecting shilling attack users and developing robust recommendation algorithms against shilling attacks. Although many studies have been devoted in this area, few of them use soft co-clustering and consider both labelled and unlabelled user profiles. In this work, the authors explore the benefits of combining soft co-clustering algorithm with user propensity similarity method and present a soft co-clustering with propensity similarity model or CCPS for short, to detect shilling attacks. Then they perform experiments using MovieLens dataset and Jester dataset to analyse it with respect to shilling attack detection to demonstrate the effectiveness of CCPS model in detecting traditional and hybrid shilling attacks and enhance the robustness of recommender systems.