Research On QoS Prediction Based On Multi-source Spatio-Temporal Information

With the continuous advancement of cloud computing technology,the cloud service market has developed rapidly,and cloud service providers provide cloud services with rich functions through the Internet.However,there are a large number of homogeneous cloud services in the cloud service market,which have similar functions but are not matched to users in the same degree.Quality of Service(QoS)is a collection of nonfunctional attributes widely used to evaluate the matching degree between user and service,so it is necessary to accurately predict the QoS of services.In recent years,research results of cloud service recommendation based on QoS prediction have been fruitful,and related technologies have been widely used in industry and academia.However,the diversification of cloud service deployment environments and the large increase in the sparsity of QoS data caused by the continuous growth of the number of cloud services all pose greater challenges to the QoS prediction methods.Indepth analysis of various factors affecting QoS and making full use of the contextual information of service invocation become the key to improving the accuracy of QoS prediction.In view of this,this paper deeply mines the multi-source spatio-temporal context information of service invocations,analyzes their interrelationships and their impact on QoS,and finally realizes high-precision QoS prediction.The work of this paper is summarized as follows:(1)A space-time aware QoS prediction method is proposed.This method proposes a new time series similarity to measure the similarity between users or services in order to select suitable neighbors for users or services;secondly,use neural network to extract the context information of similar users and similar services;finally,a joint deep network with a core of pairwise multi-layer deep network is constructed for integrating contextual information such as invocation time,user,service,similar users and similar services,learning them fully interactively and achieving the final QoS prediction.Experimental results show that extracting and utilizing multi-source spatio-temporal contextual information of service invocations can effectively improve the accuracy of QoS prediction.(2)A QoS prediction method based on deep fusion of spatio-temporal information is proposed.This method uses a pairwise multi-layer deep network to interactively learn the multi-source spatio-temporal context information of service invocations,and mines the correlation between them;after that,a spatial context-aware time series prediction model SCA-GRU is proposed,which uses context-aware gates to perceive the environment of service invocations,so as to explore the rules of QoS value fluctuations in specific service invocation environments and realize QoS prediction.Experimental results show that SCA-GRU can effectively learn the fluctuation characteristics of QoS fluctuations.