Qos Incremental Prediction Model Under Dynamic Environment

With the rapid development of the Internet,there exist many services with the same function but non-functional attributes are totally different.It is impossible for users to try all services,so how to predict the QoS of the services accurately and recommend and select optimal Web services for users has become one of the most challenging issues in the field of service computing.Although collaborative filtering algorithms have been widely used in service recommendation area,most service recommendation algorithms are based on the fact that the data set is static,considering the QoS data does not change during the service recommendation process.However,the QoS data is dynamically changing.This is because QoS performance will be affected by the user's own network,server network,server load,etc.during the process of calling the service.For example,when time varies,the QoS will be affected accordingly.Therefore,how to predict the user's interests for unused Web services in a dynamic environment and use the QoS value to select QoS services for users has become a problem that needs to be solved in Web service recommendation.In order to overcome this shortcoming,it's necessary to adopt an incremental learning method in a dynamic data environment.When a real-time dynamic QoS data in a service invocation system is invoked by a user,the data should be immediately responded and the model is updated online.In order to provide efficient and timely service recommendations,the main works of this thesis are as follows:(1)In-depth study of the deviation,this thesis proposes eight combinations of deviations and obtains Slope One algorithm based on deviations.Finally,experiments are performed on two datasets of WS-Dream.The experimental results show that the Slope One algorithm with bias can improve the recommendation accuracy of the algorithm in some extent.(2)Due to the dynamic change characteristics of QoS,this thesis research on the incremental update mechanism of Slope One algorithm on dynamic environment.Therefore,the incremental Slope One algorithm,incremental weighted Slope One algorithm and incremental bipolar Slope One algorithm are obtained.The principle of incremental algorithm is to decompose the fixed factor and the incremental factor from the original formula,and regard the user's average deviation as the quotient of the two increment factors.When the data changes,only the value of the corresponding increment factor needs to be updated,and the final prediction result are adjusted accordingly.Finally,the experiments are designed on two datasets of WS-Dream to analyze and compare the performance of the incremental algorithm and the static algorithm.At last,the experimental results show that the incremental algorithm can shorten the training time of the model while maintaining the accuracy.