User-credibility Based Service Reputation Management for Service Selection

In composite services, the atomic service reputation is becoming important when many similar functional services could be provided for selecting under the heterogeneous and loose-coupled circumstance. However, existing reputation measurement methods pay little attention on user credibility which has a great influence on accuracy. In this paper, we define user credibility as the ability of honest users to provide rational feedbacks, and propose a novel service reputation management model based on service invocation information and user feedback. This method consists of three parts: malicious feedbacks distinguishing, user credibility evaluating and service reputation predicting. Firstly, a clustering algorithm is conceived to filter out malicious feedbacks. Then a feedback deviation based algorithm is proposed to evaluate the user credibility considering service QoS similarity. Finally, an algorithm on the basis of user credibility and honest feedbacks is applied to predict service reputation. Experiment proves that our model can effectively detect malicious feedbacks and precisely measure service reputation with low error. Keywords-service reputation; user credibility; malicious feedbacks;