A fuzzy logic based reputation model against unfair ratings

Reputation systems have become more and more important in facilitating transactions in online systems. However, the accuracy of reputation systems has always been a concern for the users due to the existence of unfair ratings. Though many approaches have been proposed to mitigate the adverse effect of unfair ratings, most of them use the credibility of the rating provider alone to decide whether the rating is unfair without considering other aspects of the rating itself. Models that do consider multiple aspects often combine them through arbitrarily set weights. Therefore, they cannot work well when the credibility is not evaluated accurately or when the weights are not set properly. To resolve this problem, in this paper, we propose a reputation model which considers and combines the temporal, similarity and quantity aspects of the user ratings based on fuzzy logic to improve the accuracy of reputation evaluation. Experimental results based on a set of real user data from a cyber competition show that the proposed model is more robust against unfair ratings than the existing approaches, especially under Sybil attack conditions.