Research Of Web Service QoS Prediction Model Based On Bayesian Tensor Decomposition

The QoS attribute is an important index to measure the quality of Web services.To predict the vacant QoS values of web services can provide better services to users.Therefore,it is of great significance to predict QoS values of web services accurately.At present,most of the prediction methods are collaborative filtering algorithms based on neighbor or model.Most of them use a two-dimensional matrix of "user-service" which is established according to the historical information of users and services,and the QoS values prediction are static.The Web service invocation scenario on the current Internet is quite complicated.The user accesses the Web service in different time periods and different geographical locations,and the QoS value s is dynamically changed at this time.The traditional two-dimensional matrix data can not meet the demand of QoS prediction,and the matrix data does not consider the influence of time or other factors,resulting in the low precision of QoS values prediction.In order to solve this problem,this paper establishes the three-dimensional tensor of "user-service-time" by introducing the time information into the tensor model,And then we solve the tensor decomposition model by using the Bayesian algorithm.The specific work done in this article is as follows:(1)This paper proposes a method of Web service QoS prediction based on Bayesian tensor decomposition,BRTF.The algorithm first establishes a tensor model with time information,then to formulates the robust CP factorization under the probabilistic framework,which uses the Bayesian inference method,At the same time,a hierarchical probabilistic framework for CP decomposition is constructed.The framework takes into account data noise and the outliers generated by data interaction between user and service,which makes the prediction model more true and effective.(2)In this paper,the algorithm is applied to WSDream,a real Web service dataset.The MAE and RMSE values predicted by different data set density are simulated under the Matlab operating environment.Then we compare the three algorithms of UMEAN,UPCC and WSRec with BRTF algorithm,and analyze its feasibility and advantage.The experimental results show that the algorithm has a smaller MAE and RMSE values than other recommended algorithms,In other words,the prediction value is closer to the real value,and it has better QoS prediction effect.At the same time,When the data is sparse,it also shows relatively good prediction accuracy.