A New View on Normativeness in Distributed Reputation Systems: Beyond Behavioral Beliefs

摘要

Autonomous entities in artificial societies are only willing to cooperate with entities they trust. Reputation systems keep track of the entities' behavior and, thus, are a widely used means to support trust formation. In a P2P network, the reputation system needs to be distributed to the individual entities. In previous work, we have shown that some of the limitations of distributed reputation systems can be overcome by making use of hard evidence. In this paper, we take this idea one step further by deriving beliefs of others' trustworthiness from one's own experiences and the available hard evidence. For this purpose, we justify why a self-interested autonomous entity may choose to behave according to the norms of the system designer. As a consequence, the proposed belief model does not only incorporate behavioral beliefs but also beliefs regarding the normative-ness of an entity. We prescribe how beliefs are revised if new evidence becomes available. The introduced models for recommendations and belief formation enable us to prove that self-interested entities always issue truthful recommendations regarding transactional behavior. The simulative evaluation shows that a self-interested entity can be expected to be normative and, thus, to comply with our system design.