Bayesian memory-based reputation system

Abstract

Reputation System provides a way to maintain trust through social control by utilizing feedbacks about the service providers' past behaviors. Conventional Memory-based Reputation System (MRS) is one of the most successful mechanisms in terms of accuracy. Though MRS performs well on giving predicted values for service providers offering averaging quality services, our experiments show that MRS performs poor on giving predicted values for service providers offering high and low quality services. We propose a Bayesian Memory-based Reputation System (BMRS) which uses Bayesian Theory to analyze the probability distribution of the predicted valued given by MRS and makes suitable adjustment. The simulation results, which are based on EachMovie dataset, show that our proposed BMRS has higher accuracy than MRS on giving predicted values for service providers offering high and low quality services.