A Strategy to Detect Colluding Groups by Reputation Measures

Collusion is the malicious activity mostly frequent in agent-based recommender systems in which two or more agents agree with each other to mutually exchange high positive feedback in order to gain undue advantages by altering the correct computation of reputation measures in their agent communities. Therefore, identification of colluding agents is an important issue and several strategies have been developed to this purpose. Among them, the EigenTrust algorithm is well known, although it is limited by the necessity of knowing a priori which agents are considered as trustworthy and the impossibility of recognizing several groups of colluding agents acting simultaneously and autonomously. The focus of this paper is dealing with the above issues and, to this end, we will present a strategy to support EigenTrust both providing the necessary inputs about pre-trusted agents and recognizing groups of malicious agents. In particular, we combined EigenTrust with a clustering process in order t...