A Model For Service Reputation Evaluation Based On Objective And Automatic Feedbacks

QoS has been a research subject receiving a lot of attentions for quite a long time in SOA area. Nowadays most of the researchers use service reputation to judge the overall quality of a service while service reputation is defined as the mean of the rating from service consumers. Many experts have made a lot of researches in this area, but none of them could solve two core issues of reputation evaluation: the Subjectivity and scarcity of service consumer rating. Subjectivity is embodied in that service consumers rate services according to their own judgment. These ratings may be inaccurate, prejudiced or even malignant. Scarcity is embodied in that many service consumers may not want to rate services because rating is a manual activity which needs additional human resource cost.In order to solve these two problems, this thesis proposed a service reputation evaluation model based on objective and automatic feedbacks. Rather than using the subjective service consumer rating, this model leverages some sociological and statistical theories on top of some objective information that can be collected in real time to evaluation service reputation. Meanwhile, because of the needlessness of the subjective activities from service consumer, the collecting of feedbacks can be made automatically, thus solving the above two issues. Moreover, this model also introduced support for measurable QoS attributes. Service consumer could raise their own QoS constraints; the model will find the services with the highest reputations as well as satisfying the constraints. On the other side, although this model avoided the subjectivity and scarcity of traditional user rating, it also brought in another issue: invocation fraud. This thesis introduced the background of invocation fraud and proposed a probability distribution based solution for recognizing and filtering invocation fraud. At last, a prototype system of the model is implemented and its effectiveness was presented by experiments.According to the experiment data, the service reputation evaluation model proposed could evaluate service quality effectively while the support of measurable QoS attributes constraints could help service consumers find services that really satisfied their requirements. SOA is becoming more and more mature. The categories and number of services are increasing rapidly. The service recommendation algorithm composed of the above two parts plays a very important role in the current circumstance because it can help service consumers find proper services with high quality which is the basis of all SOA applications. Academically speaking, the thesis innovatively applied sociological and statistical theories to perform service reputation evaluation based on objective and automatic feedbacks thus expanded the scope of research ideas in QoS area. Industrially speaking, the extensible prototype system implemented in this subject allowed third parties use their own QoS attribute calculating methods and even supported substitution of service reputation evaluation algorithm, thus provided a well-defined, flexible QoS application platform.