Research On QoS Prediction Based On Deep Factorization And Neighborhood Regularization

With the increasing popularity of service-oriented computing such as big data and cloud computing,more and more Web services are deployed by service providers on the Internet.Many of these Web services have the same or similar functionality,but their non-functional attributes(Such as response time and throughput)may vary widely and these non-functional attributes are often described as Quality of Service(QoS)in service computing.How to allow users to find the most suitable Web service to satisfy their functional and non-functional requirements from a wide range of Web service providers is critical.Therefore,the QoS prediction research of Web services in the recommendation system is getting more and more attention.Predicting the QoS value of a Web service based on historical QoS records is an effective method for obtaining Web service QoS.At present,common QoS prediction methods include collaborative filtering and matrix decomposition.However,these methods have limited understanding of the deep semantics of QoS data.The nonlinear characteristics of data implied cannot be captured by these methods,which limits QoS prediction.Accuracy.In addition,because of the need for privacy protection,the QoS assistance information that can be mined and used is limited.It is necessary to mine the inherent characteristics of the data as much as possible to improve the prediction ability of the model or algorithm.For this reason,this article carries out the following work:Firstly,this paper analyzes the research status of the current Web service recommendation system field,and reviews the advantages and disadvantages of several major methods and methods of the recommendation system.Secondly,by integrating deep learning and neighbor information to improve the recommendation system,a QoS prediction model(NR-DFM)based on depth factorization and neighbor regularization is proposed.Multi-Layer Perceptron(MLP)is used to model deep interactions that learn the hidden features of users and services,and then minimize the objective function to learn the model.This model is similar to the factorization method and can be called the Deep Factorization Model.Further,this paper considers the influence of user neighbors(ie,similar users)on each other,and adds regularization terms to describe the effects of similar trust propagation.In order to smooth the learning of user characteristics.Finally,through the Web service QoS prediction on the data set WS-DREAM,it shows that the proposed method has better prediction accuracy on large data sets than the previous methods based on matrix decomposition and collaborative filtering.Deep learning can effectively capture the relationship between users and services,combine low-level features to form a more dense high-level semantic abstraction,and take into account the impact of neighbors on Web services QoS,greatly improving the accuracy of QoS prediction.