EM-trust: A Robust Reputation Algorithm for Peer-to-peer Marketplaces

Abstract We have developed EM-trust, a robust algorithm for evaluating rep- utations in peer-to-peer marketplaces. EM-trust is robust in the sense that it is far less susceptible than are previous algorithms to errors due to inaccurate feedback. Moreover, a Bayesian version of EM-trust seems ideally suited to real peer-to-peer marketplaces, in which participants are unlikely to have had previous interactions. The properties of both EM- trust variants as well as eBay’s “percent positive feedback” algorithm have been evaluated in a marketplace simulator designed to model a real peer-to-peer marketplace.