Investigation Of Service Selection For Grid Services Based On Collaborative Filtering

With the expanding of grid application areas, services in the grid wasexplosively develop A large number of services which have similar function or even thesame function come into being. As these services bring diverse choices for user, butalso it make it difficult for users to find the best service for themselves. Therefore,Quality of Service (QoS) as the core of service nonfunctional attributes, becomes thefocus of researchers. However, most existing service selection approaches ignore thediversity of the service environment and assume that different users have receivedidentical QoS from the same service provider. According to this assumption, it is hardfor service providers to satisfy the individual demands of user. The collaborativefiltering method as the key part of collaborative recommendation system has beenwidely used in online shopping, music recommendation and other fields, and it meetsthe personalized needs of users appropriately, and achieves a good effects. Therefore, itis the inevitable choice to use the collaborative filtering method to solve grid servicesselection problem.Aiming at solving the grid service selection problem, this paper investigates gridservice selection method based on collaborative filtering method and makescontribution to the following part:1. The paper summarizes the problem of grid service selection and proposes a gridservice selection method based on K nearest neighbor collaborative filtering accordingto specific scenes. At first, a similarity metric according to characteristic of service QoSdata is proposed. Secondly, try to find similar user set of currently active user thoughmining the historical QoS data by using this new similarity metric. Then, the methodpredicts the candidate service QoS value of active user by using the historical QoS dataof similar user set. Finally, the method find the optimal service by using integralstrategy on the prediction results.2. In order to solve the sparse data problem of service QoS historical data, selectionmethod based on clustering smoothing collaborative filtering. At first, cluster users to Kclass based on the historical QoS data. Secondly, fill the blank by smoothing the Qosdata based on cluster result. Thirdly, predict the candidate service QoS value of activeuser by using the historical QoS data and fill data. Finally, find the optimal service byusing integral strategy on the prediction results. Experimental results show that, the two proposed methods can effectively improvethe quality of user selection, and meet the individual demand of users in servicenon-functional attributes fields.