Web Service QoS Prediction Based On Classification

With the rapid development of Internet technology,Service Oriented Architecture(SOA)is prevalent in the area of distributed systems and software integration.Under this circumstance,the number of Web services has grown rapidly in recent years,which makes it increasingly difficult for users to select services that meet their needs from great amount of Web services.Hence,in order to meet the needs of each user,how to choose from a large-scale Web services group with high-quality services and make personalized recommendations is a very challenging task.As we know,the methods for Web service selection and recommendation based on QoS(Quality of Service)are very popular now.It is critical to accurately predict missing QoS values before making personalized recommendations for each user.Collaborative Filtering(CF)is widely employed for making Web service recommendation.This approach uses the history of a user’s call records to analyze each user’s preferences and identifies similar groups,making recommendations very intelligently.However,traditional Neighborhood-based Collaborative Filtering(CF)models fail to capture the potential factors among user and service like network location and geographic location,which have distinct influence to Web service recommendation.In addition,CF is likely to suffer from low prediction when the data used for prediction is very sparse.To solve the problems above,we proposed two novel approaches:(1)we make full use of the potential user-service features,and propose a collaborative filtering algorithm based on Classification: first of all,by leveraging the historical user-service QoS,we can get the personalized characteristics like users’ longitude,latitude,services’ provider number and services’ region number,then classify the users-service to get the user-service tag by the naive Bayes algorithm.Finally,the user-based collaborative filtering algorithm is used to predict the QoS value of the target service in the similar user set of the target user,thus improving the accuracy of the prediction.(2)Aiming at the problem that the prediction accuracy of collaborative filtering algorithm is limited by similar user selection,we proposed a similar user selection method based on DBSCAN co-occurrence matrix,which improves the accuracy ofsimilar user selection.Furthermore,the classification accuracy of the classifier suffers from the validity of the target characteristic vector,we presented the frequency vector feature of user and service,which can significantly identify the user-service personality characteristics and improve the classification accuracy of AdaBoost classifier.According to the probability neighbor model of the output of the classifier,a new aggregation model is proposed,which improves the prediction accuracy by combining the results of the two probabilistic neighbor models.In this paper,experiments on real datasets using the two methods,and compared with some well-known methods.The experimental results show that the accuracy of the two methods in this paper have improved.