Toward Secure Trust and Reputation Systems for Electronic Marketplaces

In electronic marketplaces, buying and selling agents may be used to represent buyers and sellers respectively. When these marketplaces are large, repeated transactions between traders may be rare. This makes it difficult for buying agents to judge the reliability of selling agents, discouraging participation in the market. A variety of trust and reputation systems have been proposed to help traders to find trustworthy partners. Unfortunately, as our investigations reveal, there are a number of common vulnerabilities present in such models---security problems that may be exploited by `attackers' to cheat without detection/repercussions. Inspired by these findings, we set out to develop a model of trust with more robust security properties than existing proposals. Our Trunits model represents a fundamental re-conception of the notion of trust. Instead of viewing trust as a measure of predictability, Trunits considers trust to be a quality that one possesses. Trust is represented using abstract trust units, or `trunits', in much the same way that money represents quantities of value. Trunits flow in the course of transactions (again, similar to money); a trader's trunit balance determines if he is trustworthy for a given transaction. Faithful execution of a transaction results in a larger trunit balance, permitting the trader to engage in more transactions in the future---a built-in economic incentive for honesty. We present two mechanisms (sets of rules that govern the operation of the marketplace) based on this model: Basic Trunits, and an extension known as Commodity Trunits, in which trunits may be bought and sold. Seeking to precisely characterize the protection provided to market participants by our models, we develop a framework for security analysis of trust and reputation systems. Inspired by work in cryptography, our framework allows security guarantees to be developed for trust/reputation models--provable claims of the degree of protection provided, and the conditions under which such protection holds. We focus in particular on characterizing buyer security: the properties that must hold for buyers to feel secure from cheating sellers. Beyond developing security guarantees, this framework is an important research tool, helping to highlight limitations and deficiencies in models so that they may be targeted for future investigation. Application of this framework to Basic Trunits and Commodity Trunits reveals that both are able to deliver provable security to buyers