Bazaar: Strengthening User Reputations in Online Marketplaces

Online marketplaces are now a popular way for users to buy and sell goods over the Internet. On these sites, user reputations—based on feedback from other users concerning prior transactions—are used to assess the likely trustworthiness of users. However, because accounts are often free to obtain, user reputations are subject to manipulation through white-washing, Sybil attacks, and user collusion. This manipulation leads to wasted time and significant monetary losses for defrauded users, and ultimately undermines the usefulness of the online marketplace. In this paper, we propose Bazaar, a system that addresses the limitations of existing online marketplace reputation systems. Bazaar calculates user reputations using a max-flow-based technique over the network formed from prior successful transactions, thereby limiting reputation manipulation. Unlike existing approaches, Bazaar provides strict bounds on the amount of fraud that malicious users can conduct, regardless of the number of identities they create. An evaluation based on a trace taken from a real-world online marketplace demonstrates that Bazaar is able to bound the amount of fraud in practice, while only rarely impacting non-malicious users.