Application of belief propagation to trust and reputation management

This paper introduces the first application of Belief Propagation (BP) in reputation systems. We view the reputation management as an inference problem, and hence, describe the reputation management problem as computing marginal likelihood distributions from complicated global functions of many variables. However, we observe that computing the marginal probability functions of the reputation variables is computationally prohibitive for large scale reputation systems. Therefore, we propose to utilize the BP algorithm to efficiently (i.e., in linear complexity) compute these marginal probability distributions; leading to a fully iterative probabilistic and BP-based approach (referred to as BP-ITRM). BP-ITRM describes the reputation system on a factor graph, using which we can obtain a qualitative representation of how the service providers (sellers) and consumers (buyers) are related. Further, by using such a graph representation, we compute the marginal probability distribution functions of the variables representing the global reputation values via an iterative message passing algorithm. We show that BP-ITRM significantly outperforms the well-known and commonly used reputation management schemes such as the Averaging Scheme, Bayesian Approach and Cluster Filtering in the presence of attackers. Further, its complexity is linear in the number of service providers and consumers, far exceeding the efficiency of other schemes.