Shilling Attack Detection in Recommender Systems via Selecting Patterns Analysis

• DOI: 10.1587/transinf.2015edp7500
• Publication Date: 2016-09-30T22:22:47Z
• Authors: Wentao LI; Min GAO; Hua LI; Jun ZENG; Qingyu XIONG; Sachio HIROKAWA
• Keywords: MovieLens; Popularity; Feature (linguistics)

Collaborative filtering (CF) has been widely used in recommender systems to generate personalized recommendations. However, using CF are vulnerable shilling attacks, which attackers inject fake profiles manipulate recommendation results. Thus, attacks pose a threat the credibility of systems. Previous studies mainly derive features from characteristics item ratings user detect attackers, but methods suffer low accuracy when adopt new rating patterns. To overcome this drawback, we properties popularity profiles, determined by users' different selecting This feature extraction method is based on prior knowledge that select items rate with man-made rules while normal users do according their inner preferences. Then, machine learning classification approaches exploited make use these and remove attackers. Experiment results MovieLens dataset Amazon review show our proposed improves detection performance. In addition, justify practical value derived