Research On QoS Prediction Approaches Based On Neural Networks For Web Services

With the development of cloud computing,big data and service oriented architecture(SOA)technology,big service ecosystem arise at the historic moment.It has become a challenging research topic to how to efficiently recommend services that satisfy users' personalized needs from large scale services with the same functional properties and different attributes.At present,QoS has been widely used in service composition,service selection and service recommendation.However,in the real Internet environment,users cannot access the QoS values of these services by invoking all services,which can lead to the user-service matrix be very sparse.So it could make the prediction accuracy of some prediction methods decreased when predicting the missing QoS value.To this end,the recommendation system should to recommend services for users to meet their personalized need in the case of the QoS information sparsity.At the same time,two kinds of Web service quality prediction methods based on neural network algorithm are proposed to solve the problem of poor scalability of some algorithms.The main contributions of this article are as follows:(1)A quality of service prediction method based on SOM algorithm(SOMQP)is proposed.By introducing the domain function with topological structure,all neurons are places in a topological structure based on prior knowledge,which achieves a more stable clustering result.Then,according to the clustering results,a new top-k selection mechanism is applied to select similar users and similar services for target user and target service.Finally,a mixed prediction method is used to predict the missing values.The experimental results show that the prediction accuracy of the proposed method is better than the mainstream prediction algorithm in the case of very sparse data.(2)This paper proposes a quality of service prediction method based on the covering algorithm(UIQPCA).The method has two stages:the QoS prediction based on historical time slice and the QoS prediction based on current time slice.In the prediction phase based on the history time slice,first,according to the historical QoS data that users linvoke services on the front time slices,the missing values on the next time slice of the services are predicted.If the user does not invoke the service on the previous time slice,the covering algorithm is used to find the similar users and similar services of target users and services,then we predict the missing values based on historical QoS data of the similar users and similar services.After predicting the missing values of the historical time slices on the first stage,the QoS values of all the user-service on the current time slice can be predicted according to the data of the historical time slices.The experimental results show that this method can better solve the prediction problem of QoS dynamic change.(3)Finally,we conduct extensive experiments on the widely used real-word Web service dataset,WS-Dream,to evaluate the prediction accuracy and efficiency of the proposed methods.The experimental results demonstrate that compared with the classical CF algorithm,the prediction algorithm based on the improved PCC method and the prediction algorithm based on the k-means method,the prediction accuracy of SOMQP is increased by 34.9%,29.5%and 4.3%,and the prediction accuracy of UIQPCA is increased by 22.5%,11.9%and 8.7%.