Research On QoS Prediction Algorithm For Web Service Recommendation

With the rapid development and deep application of “Internet+”,Web services,as an important technology to support interoperability of heterogeneous systems,collaboration of distributed application systems and data sharing,have become an indispensable "digital link" in cross-industry and cross-platform business collaboration.On the one hand,users can invoke corresponding Web services to realize business functions based on their own business needs,without having to relearn and implement specific services with equivalent functions.On the other hand,service providers can also obtain corresponding benefits by providing Web services.In this context,more and more Web services are deployed on the Internet,and the number of Web services with equivalent functions is also increasing.As a result,the problem of information overload of Web services is becoming increasingly severe.Qo S,as important attributes for measuring non-functional quality of service such as response time and throughput of Web services,can objectively distinguish and evaluate Web services with equivalent functions.Therefore,Qo S-based Web service recommendation has become an important measure to solve information overload of Web services.However,in practical application scenarios,Qo S data is not only sparse but also has numerous influencing factors.The influencing factors of Qo S vary among different application scenarios.If unknown Qo S cannot be accurately predicted based on application scenarios,it is difficult to recommend suitable Web services for users from numerous Web services with equivalent functions.This dissertation focuses on Qo S prediction needs with different influencing factors,and combines the Qo S data features of Web services,the autonomous system and comprehensive location of users and web services,as well as their correlation,and Qo S data privacy information to design and implement Qo S prediction algorithms suitable for different scenarios,thereby enhancing and improving the effectiveness and user experience of Web service recommendations.Specifically,the innovative works in this dissertation can be summarized as the following three points:(1)For the application scenarios without spatio-temporal information,aiming at the impact of potential local data features implied by Qo S data distribution differences on the accuracy of Qo S prediction,a Web service quality features-aware Qo S prediction algorithm ALSHMF is proposed.Through experimental analysis,it was found that "Qo S data of Web services do have distribution differences and the differences are related to the accuracy of Qo S prediction".Based on this,the traditional Jaccard similarity calculation method is amplified and integrated with Locality Sensitive Hashing algorithms to improve the aware ability of Qo S potential local data features,and achieve Web service neighbors’ selection with similar Qo S potential local data features.Then,matrix factorization and weighting mechanism are introduced based on the Web service neighborrelationships and Qo S overall data features to achieve accurate prediction of Qo S.A series of experiments based on real datasets show that the potential local data features implied by Qo S data distribution differences are important factors that affect the accuracy of Qo S prediction.The fusion of Qo S potential local data features-aware and weighted prediction mechanism can effectively improve the accuracy of Qo S prediction.(2)Numerous studies have shown that the spatio-temporal information of users and Web services is an important factor affecting the accuracy of Qo S prediction.Therefore,for the application scenarios with spatio-temporal information,aiming at the problem that auto-nomous systems,location information and their correlations of user and Web services affect the accuracy of Qo S prediction,a new Qo S prediction algorithm LDNN based on location aware and deep neural network is proposed.The connection of data preprocessing,features-learning neural network and Qo S prediction neural network is implemented in a serial manner to construct a bottom-up hierarchical data processing model.The unified cost function is used for model training to realize the learning of high-dimensional non-linear features in auto-nomous systems and comprehensive location information of users and Web services,and the learned features and the comprehensive location correlations obtained through processing are used to achieve Qo S prediction,so as to improve the effectiveness of Web service recommendation.The experimental results show that auto-nomous systems,location information and their correlations can provide valuable information for Qo S prediction.The LDNN that integrates auto-nomous systems,location information and their correlations can effectively improve the accuracy of Qo S prediction,and has better prediction accuracy in the datasets with different matrix densities.(3)In addition to spatio-temporal information,application scenarios involving user privacy also impose requirements on Qo S data privacy security.Therefore,for the application scenarios with privacy protection requirements,aiming at the problem of Qo S data privacy security and the impact of comprehensive information on the accuracy of Qo S prediction,a hybrid Qo S prediction algorithm DVO+LCLMF with Qo S data privacy protection function is proposed.Introducing spatial vector rotation to construct a privacy protection algorithm DVO based on conformal transformation,which achieves Qo S data obfuscation by vector rotation.The multiple neighbors’ selection of Web services is achieved by location distribution and Qo S features clustering of Web services,and then the matrix factorization model is introduced based on the multiple neighbor relationships to construct a Web service location and Qo S features aware Qo S prediction algorithm LCLMF,and fuse LCLMF with DVO to simultaneously achieve privacy protection of historical Qo S data and accurate prediction of unknown Qo S.The experimental results show that DVO+LCLMF can not only maintain Qo S data availability while protecting their privacy,but also comprehensively utilize the location information and Qo S features of Web services to improve the accuracy of Qo S prediction.Compared with LCLMF,the average fluctuation ranges of MAE and RMSE of DVO+LCLMF with privacy protection function is less than 0.6%.