Reputation in Multi Agent Systems and the Incentives to Provide Feedback

Problems with Rationality Rational agents will not submit feedback at all because it is not rational for them to do so feedback is a public good, giving feedback is associated with costs but no direct gain. Reputation Mechanism needs to set incentives to submit feedback to further trustworthy feedback Problems with Rationality Rational agents will not submit feedback at all because it is not rational for them to do so feedback is a public good, giving feedback is associated with costs but no direct gain. Reputation Mechanism needs to set incentives to submit feedback to further trustworthy feedback Main Objective: distinguishing between trustworthy and untrustworthy, and between honest and dishonest agents Untrustworthy feedback is ignored, old feedback values are discounted over time The outcome of the transaction is measured with the help of three dierent kinds of reputation and a measure for the discrepancy for the advertised and the delivered service. Incentives Liu et al. set incentives with the meta rating. Example: 1. Agent a asks agent r for recommendations 2. Agent r evaluates the state of the agent a and if it has a signicant number of direct experiences. 3. Agent r sends back the recommendation due to the state of agent a: active truthteller sends back the recommendation immediately. inactive recommenders sends back the reputation with the probability of di = δ a − (r p + r n − 2). Liars and truthtellers are distinguished by a small value of (decreasing for liars and increasing for truthtellers) active liar does not send anything back. Therefore the less active an entity is, the less possible that it receives helpful recommendations from others. Incentives Liu et al. set incentives with the meta rating. Example: 1. Agent a asks agent r for recommendations 2. Agent r evaluates the state of the agent a and if it has a signicant number of direct experiences. 3. Agent r sends back the recommendation due to the state of agent a: active truthteller sends back the recommendation immediately. inactive recommenders sends back the reputation with the probability of di = δ a − (r p + r n − 2). Liars and truthtellers are distinguished by a small value of (decreasing for liars and increasing for truthtellers) active liar does not send anything back. Therefore the less active an entity is, the less possible that it receives helpful recommendations from others. The advised X …