On a mechanism of detection of coalitions for reputation systems in P2P networks

One of the most common types of attacks on reputation systems is made by reporting unfair ratings. They can be performed by the individual malicious members or by a group of agents forming a coalition and cooperating together in order to achieve particular purpose (e.g, to gain higher position in the community or to discredit the competition). Due to the fact, that such attacks are performed often by intelligent coalitions, they are much more harmful and harder to detect. This paper describes a novel algorithm for coalition detecting in reputation systems. By observing the agents which are controversially rated by the community it is possible to construct agents cooperation matrices and identify harmful coalitions. Detailed description of the algorithm is provided and presented simulation results confirm the effectiveness of the detection.