Service Selection Method Based On Similar Service Clustering And User Preferences

Most of existing service selection algorithms selects the “best” service with highest evaluated value of service quality from a great number of services. Such a strategy is hard to provide highly efficient selection process and might make too many service requests rush for one service provider and bring the overload. Making one service provider hold too many requests would increase waiting time of users and decrease the quality of invoked service. In fact, many users prefer to get a better service in a short time other than a best service after a long time. Moreover, sometimes it is difficult for users to provide preference on some attributes of Qo S and corresponding preference determination mechanism needs to be proposed.To solve the above problems, this paper clusters the candidate services firstly. Candidate services could be grouped into multiple classes based on the similarity of their service quality attributes and a virtual service could be determined to represent a service class. By taking virtual service as candidates, selection efficiency could be improved a lot. The high similarity in a service class enables the member service to substitute each other, which make it easier to balance the load and find backup service if necessary. Besides, for a Qo S attribute which a user knows little about an extension of collaborative filtering method is used to get weight for this user through preferences from similar users. Based a complete preference description, a better service could be finally determined after selecting a candidate service class. A case study is used to illustrate our methods and the simulation shows it can improve the selection efficiency and load balance.