Research On QoS Prediction Of Web Services Based On Location Clustering And Tensor Decomposition

More and more developers have tended to buy various Web services to build their application platforms.In the face of the large number of related products on the market and the high degree of homogenization of functions,it is necessary for service providers to help users determine the most suitable Web services through technical means.Therefore,service recommendation technology emerges as the times require and becomes one of the effective means to solve the above problems.However,as the basis for service recommendation,QoS data often contains missing parts.Too many QoS data will seriously affect the quality of service recommendation.Therefore,it is necessary for data scientists to predict and fill in missing values in QoS data before making recommendations.In this thesis,for the QoS missing value prediction problem in service recommendation,the tensor method is used to complete the following three aspects:Firstly,in view of the problem that the traditional tensor model does not pay attention to the QoS data location information,this thesis clusters the QoS data containing the location information,and aggregates the services with similar geographical locations to form several small QoS tensors.The tensor model is used to predict the missing values separately,which improves the accuracy of the prediction.Secondly,this thesis improves the solution of QoS missing value prediction.On the one hand,it considers the error of the corresponding part in the missing QoS tensor and the corresponding position in the prediction result.On the other hand,this thesis is inspired by the image noise reduction problem.The QoS tensor after interpolated is treated as data containing noise and then denoised.Based on the location clustering of QoS information,this thesis combines these two solution ideas with the improved high-order orthogonal iterative algorithm,and proposes two QoS missing value prediction models of CHOOI1 and CHOOI2.The feasibility of the method and the accuracy of the prediction results are verified on the public data set WSDREAM dataset 2.In order to further improve the accuracy of model prediction,this thesis also combines the CHOOI2 model with the best performing NTF model in the traditional tensor model,and proposes the QTF model.Experiments show that the QTF model shows better prediction accuracy and running time than the former two in the QoS missing value prediction task.Thirdly,for the high-order orthogonal iterations occupying large memory space and slow iteration,this thesis starts its steps from the iterative process and further completes the parallelization of QTF model,and implements it under Apache Spark distributed computing framework.Parallelization work is mainly focused on its CHOOI2 model part.From the experimental results,the parallel version of the algorithm has a certain improvement in runtime compared to the serial version,and has greater potential to handle QoS prediction tasks on large-scale data sets.