A Novel Quality Of Service Prediction Method Based On Multivariate LSTM Model

With the development of Service-Oriented Architecture(SOA),the number of Web APIs is growing every year,and there are a large number of services that provide similar functionality.The proliferation of multiple Web services with identical or similar functionalities increases the difficulties of selecting appropriate services for service consumers.The non-functional properties of services thus become major considerations in decision making of service selection and recommendation.Quality of Service(QoS)is employed for describing non-functional properties of services,and it becomes the determining factor in selecting the best service from those providing similar functionality.QoS data could be either static or dynamic.The static QoS data can be declared by service providers and exposed as part of the service description.For the dynamic QoS properties,different users may have different service experiences at different time slots due to the changing network environments and geographical locations.The dynamic QoS properties are thus unsteady and time-dependent.Therefore,how to predict future QoS values according to the past observed values or QoS data shared by services invoker becomes a vital issue in service computing.In service selection and recommendation systems,it is essential to select services that can provide the best QoS performance from candidate services.Particularly,the past observed values can help us identify the service that provides the best QoS.Many approaches have been proposed to predict unknown QoS values of Web services according to their historical invocation records.Despite the good results obtained by these approaches,a limitation still exists,that is,(?)these approaches usually analyze each individual QoS data as a basic unit while ignoring the intrinsic characteristics of these time-series QoS data.In an extremely dynamic environment,how to capture the intrinsic and time-varying characteristics of QoS data from a finer-grained perspective becomes one of the essential issues to achieve accurate prediction;(?)these approaches are usually aimed at a specific single task prediction,such as,predicting the response time of the service.While these approaches usually don't consider the learning representation of sharing multiple related tasks.As a result,how to use multitask prediction method to predict the QoS of specific time and the QoS relationship between adjacent moments to improve the generalization ability of the model becomes another essential issue to improve the performance of QoS prediction.For the first problem,this paper proposes a hybrid QoS prediction approach by combining the Empirical Mode Decomposition(EMD)and the multivariate LSTM(Long Short Term Memory)model,named E-mLSTM.The approach aims to capture the potential information in the historical sequence from a finer-grained perspective and perform accurate QoS forecasting.Experiments conducted on two real-world datasets show the effectiveness of the proposed model in QoS prediction.For the second problem,based on the proposed E-mLSTM,this paper proposes a multitask QoS prediction approach by combining the attention mechanism and E-mLSTM,named AE-mLSTM.The approach can learn the shared representation among different tasks,and predict the QoS of specific time and the QoS relationship between adjacent moments.Experiments demonstrate that our model is superior to several mainstream methods in QoS prediction.