Towards reliable rating filtering in grid reputation systems: A preevaluating set based bias-tuned approach

Reputation-based trust system emerges as a promising mechanism for trust establishment between unknown entities in Grid, in which the reliability of first-hand ratings plays a crucial part. In this paper, we propose a pre-evaluating set based bias-tuned approach for dishonest feedback filtering in such system, which has taken reputation's subjective feature into consideration. With the fact that different rater may have different rating criteria in mind, our filtering method will not blindly filter out apparently dishonest feedbacks, such as higher ratings from lenient raters. The basic idea for filtering is that, first align ratings according to some same rating criteria and then try to find inconsistency among tuned ratings. Specifically, rating criteria is tracked by means of the pre-evaluating set introduced especially to our trust model. Timing is performed according to the current evaluator's rating criteria with an interpolation method Inconsistency examination consists of two aspects "credibility filtering" and "on-spot filtering". The former tries to find inconsistency in ratings given to entities familiar to the current evaluator. And the latter tries to find inconsistency in the currently retrieved ratings. With a two-dimension function, results from these two filtering are combined Experiment results show that we can effectively filter out dishonest feedbacks and retain honest ones both to a large extent.