Web Services QoS Prediction Based On Non-Negative Multiple Matrix Factorization

As an application method of Service Oriented Architecture(SOA),Web Services uses various standards and protocols based on Extensible Markup Language(XML)to standardize the various links of services,making cross-platform resource sharing and application integration technologies develop rapidly.With the rapid increase in the number of services in the network,how to predict QoS(Quality of Service)accurately,then effectively recommend services to users is more and more important.In the classic QoS prediction algorithm,the accuracy of collaborative filtering algorithm is often restricted by the data sparseness,and it is not suitable for the cold start problem.Matrix decomposition can better deal with data sparseness,but it is still underutilized for spatio-temporal information and exists a problem that the results are not interpretable enough.In addition,most of the current research focuses on a certain QoS attribute.The prediction of comprehensive attributes is minimal.For this reason,this article did the following studies:Utilizing the deep hidden relationship between users and users,users and services,and services and services,using the Non-negative Multiple Matrix Factorization(NMMF)algorithm,make the three matrices of user similarity relation matrix,service similarity relation matrix and QoS matrix are decomposed into four matrices based on sharing the base matrix and the coefficient matrix.Then the low-latitude matrices from decomposition are multiplied to realize the prediction of unknown values in the original matrix.By studying the influence of user(service)location on the similarity,based on Pearson's similarity,different biases are set by region,and a method of using the user and service location information to promote the accuracy of similarity calculation is proposed.Based on this,two auxiliary matrices in the NMMF model are modified,and Location based Non-negative Multiple Matrix Factorization(LNMMF)prediction algorithm is proposed.The comprehensive description method of the QoS attributes of service efficiency is presented and expressed as the ratio of service throughput and response time to achieve the combined effect of service throughput and response time prediction.On the basis of studying the characteristics of the ratio data,the LNMMF algorithm is applied to the prediction of this comprehensive attribute,and the application of the algorithm is further extended.The proposed method was verified using real data sets.The results show that the NMMF algorithm can make full use of the implicit information of the auxiliary matrix and the prediction accuracy is higher than the conventional classical algorithm.By improving the accuracy of the auxiliary information through the positional relationship,the LNMMF algorithm can further play the role of an auxiliary matrix in the case of sparse data and significantly improve the prediction accuracy.