Exploiting Trust and Suspicion for Real-time Attack Recognition in Recommender Applications

As is widely practiced in real world societies, fraud and deception are also ubiquitous in the virtual world. Tracking and detecting such malicious activities in the cyber space is much more challenging due to veiled identities and imperfect knowledge of the environment. Recommender systems are one of the most attractive applications widely used for helping users find their interests from a wide range of interesting choices that makes them highly vulnerable to malicious attacks. In this paper we propose a three dimensional trust based filtering model that detects noise and attacks on recommender systems through calculating three major factors: Importance, Frequency, and Quality. The results obtained from our experiments show that the proposed approach is capable of correctly detecting noise and attack and is hence able to decrease the absolute error of the predicted item rating value.