| In recent years,with the rapid development of cloud computing,the number of cloud services has grown exponentially,which makes it more difficult for users to choose services.Quality of Service(QoS)is an important indicator for evaluating cloud services and a powerful reference for user service selection and recommendation.In order to help users choose suitable services,service recommendation systems often rely on QoS Prediction technology to screen candidate services,and QoS prediction has also become a popular research problem in service computing.Existing QoS prediction methods usually combine context information to improve the accuracy of QoS prediction by mining the implicit features.The most commonly used context information is location and time.However,among the location-aware QoS prediction methods,collaborative filtering methods do not have the ability to learn nonlinear features;although deep learning methods that can learn nonlinear features,these methods generally have the problem of gradient disappearance and can not effectively exert the advantages of deep learning in QoS prediction.While among the time-aware QoS prediction methods,it often only predicts the historical unknown QoS,ignoring the prediction value of time series for the future QoS values.Aiming at the above problems,this paper studies QoS prediction methods based on location and time context information.The main research contents are as follows:(1)This paper expounds the current research progress of QoS prediction,summarizes the advantages of context-aware QoS prediction technology,and analyzes the shortcomings of location-aware and time-aware QoS prediction methods.(2)In order to solve the problem of gradient disappearance problem,which affects the QoS prediction performance of existing deep learning motheds,this thesis proposed a location-aware QoS prediction method(PLRes)based on residual network.This method combines the geographical location and network location,and constructs QoS probability distribution characteristics to improve the accuracy of QoS prediction.(3)In order to solve the problem of predicting the future QoS value,this paper proposes a time-aware QoS prediction method(TF-TCN)based on temporal convolutional network.Firstly,the historical unknown QoS is filled by a tensor factorization model based on attention mechanism.Then the future multi-time QoS values is predicted by time convolution network based on the complete filled QoS series data.(4)The two QoS prediction methods are tested on the public dataset WS-Dream,which is a real-world dataset generated by invoking Web services.This paper sets different densities of sparse matrix for a large number of experiments to fully verifies the effectiveness of this two methods. |
