An Analytic Approach to Reputation Ranking of Participants in Online Transactions

Agents from a community interact in pairwise transactions across discrete time. Each agent reports its evaluation of another agent with which it has just had a transaction to a central system. This system uses these time-sequences of experience evaluations to infer how much the agents trust each another. Our paper proposes rationality assumptions (also called axioms or constraints) that such inferences must obey, and proceeds to derive theorems implied by these assumptions. A basic representation theorem is proved. The system also uses these pairwise cross-agent trustworthiness to compute a reputation rank for each agent. Moreover, it provides with each reputation rank an estimate of the reliability, which we call weight of evidence. This paper is different from much of the current work in that it examines how a central system which computes trustworthiness, reputation and weight of evidence is constrained by such rationality postulates.