AuRo-Rec: An unsupervised and Robust Sybil attack defense in online recommender systems

With the explosive growth of online social networks (OSNs), the social commerce and online stores facilitating recommender systems (RSs) are a popular way of providing users customized information such as friends, books, goods, and so on. The major function of RSs is recommending items to their system users (i.e., potential consumers), however, malicious users attempt to continuously attack the RSs with fake identities (i.e., Sybils) by injecting false information. In this paper, we propose an Unsu-pervised and Robust Sybil attack defense in online Recommender systems (AuRo-Rec) which exploits dynamic auto-configuration of system parameters on top of the admission control concept. AuRo-Rec firstly provides highly trusted recommendations regardless of whether ratings are given by Sybils or not. To build the automatic parameter configuration required by Auto-Rec, we propose an unsupervised approach: Dynamic Threshold Auto-configuration (DTA). To evaluate our approaches, we conducted experiments against four possible Sybil attacks. The experimental results confirm that AuRo-Rec works robustly in terms of prediction shift (PS).