Abstract

The problem of inaccurate ratings is one of the basic issues in the trust model of Multi-Agent System (MAS). The inaccuracy may come from the witness deliberate lying or other factors that are independent of the witness. In most existing trust models, the ratings different from history records or inconsistent with most opinions are regarded as inaccurate. It's not only inappropriate but also helpless to pick out the good interaction partner. In this paper, a new approach is proposed to support evaluators to revise the ratings from evaluators' viewpoints. In addition, a learning algorithm is put forward which can make evaluators perceive the non-witness-side factors leading to inaccurate ratings. By the aid of our work, the evaluators can find out the better interaction target and get greater utility in the future interaction. The experiment shows that with our method evaluators can select the better target in both static and dynamic environments.

Paper Details