Alternative Set Generation Methods Of Composite Web Service Based On Collaborative Filtering

Web service, which is based on interoperability, cross-platform, loosely coupled, has been widely applied on the Internet. However, as the requirements for Web service application become more complex, single Web service cannot meet users’ requirements. As a result, composite Web service, which is value-added service of Web service, composites single Web services according to some business logic so that it can meet users’requirements at a higher level. Each task of composite Web service corresponds to a Web service. Once one of these services fails, the composite Web service whose adaptive adjustment is costly forced to re-selection process. At the same time, there have been many Web services providing the same or similar functionality on the Internet. Hence, how to generate a candidate services set which could recommended the good service meeting user needs to the composite service and replace it with an alternative service when the recommended services has problems is becoming a hot issue in research field.To this problem, this thesis discusses an alternative set generation method based on collaborative filtering to generate an alternative for each task of composite Web service. The services in alternative set firstly use a personal QoS prediction method to predict the QoS Values. After that, the services should be sort by the QoS value prediction results. The service, which is in the front of the set, has priority when recommended to the composite Web service while the other services are waiting for replace the fail service. To speed up the generation efficiency, this thesis also introduces a pattern matching approach, which cooperates with the personal QoS prediction method to generate the alternative set for the Composite Web service.This thesis first proposes a generation method for usage information to generate service usage information from monitoring log and executing log. The service usage information contains all the necessary information needed by the whole QoS prediction process. Then, in order to use the pattern matching approach, this thesis puts forwards a pattern extraction method based on the DBSCAN algorithm to extract the pattern from the service usage information. After that, this thesis focuses on using the personal QoS prediction method which cooperate with the pattern matching to predict the service QoS values according the pattern results, considering user’s environmental characters, host characters, input characters and time characters, and using the collaborative filtering algorithm. In addition, this thesis presents a gray relational analysis approach to extract the personal business characters and a pattern QoS features complement method based on the traditional collaborative filtering algorithm. In the end, the thesis introduces the application of the generation mechanism in detail and represents the using experiments of this mechanism including the experiment analysis of this mechanism.