Research On Web Service QoS Prediction Method Based On Multi-feature Extraction

At present,with the development and improvement of emerging technologies,such as cloud computing,big data,Internet-of-things,mobile computing and smart city,Web services have become an important carrier for IT resource delivery.Such a trend promotes the rapid development of services in three directions(i.e.,intelligence,personalization and integration),and spawns many new services.At the same time,due to the increase in available services,it is a difficult task for a user to select services that match his or her requirements.In order to effectively solve this problem,the service recommendation strategy has attracted wide attention from various research scholars.The service recommendation task is to recommend suitable services to meet users' functional requirements,and also requires that the recommended services should meet the non-functional requirements(quality-of-service,QoS for short),especially for those services with the same or similar functions.However,in a real environment,the number of services that a user has invoked before is usually quite limited,resulting in a serious lack of historical QoS values.Therefore,it becomes an inevitable task to predict missing QoS values in high accuracy.And researches on Web service QoS prediction have become a research hotspot in the field of new service computing.In recent years,some researchers have proposed some QoS prediction methods for Web services.Among all prediction methods,the collaborative filtering(CF)method has been deeply studied and applied,mainly because of its simplicity and effectiveness.The basic CF method does not depend on structural information of the recommend task,and can achieve good prediction accuracy.The CF-based prediction methods can be divided into two categories,i.e.,neighborhood-based methods and model-based methods.However,these two methods have their own shortcomings.The neighborhood-based methods can not effectively solve the problem of data sparseness and noise data.The model-based method is not capable of learning deep features of users and services,this may lead to a decline in prediction accuracy.To solve the exiting problems,this paper proposes to combine the advantages of neighborhood-based method,model-based method and deep learning technique to explain and express the invocation behavior of users and services.This paper wants to improve the accuracy of QoS prediction and further alleviate the data sparse problem by making full use of the deep implicit features.Therefore,this paper proposes a new feature-based QoS prediction model,which combines the deep learning method and CF method.This model uses the powerful feature extraction ability of the deep learning method to obtain deep implicit features and improve the accuracy of the prediction results.The main contributions of this paper are as follows:(1)This paper proposes a neighbor selection method based on environmental characteristics.Due to the complexity and instability of the Web environment,some QoS data might be abnormal,which causes some noise data and in turn affects the prediction accuracy.Therefore,this paper proposes a new feature-based neighbor selection method by fully considering invocation behavior preferences of users and services,location information.This method can effectively deal with data sparseness.First,this method users a new similarity calculation to obtain the similarity of users and services.Then,selecting a clustering method by judging whether data is sparse.If data is not sparse,the final neighbor set is obtained by a K-prototypes clustering method and a verification by neighbor reversibility.If else,this method directly uses a fuzzy clustering method to obtain the final neighbor set.(2)This paper proposes a Web services QoS prediction method based on multi-feature extraction.Aiming at the problem that data sparse and shallow model can not learn nonlinear features,a QoS prediction model based on multi-feature extraction is proposed.This model integrates the convolutional neural network,the variational auto-encoder and the restricted Boltzmann machine into the traditional collaborative filtering method,which is composed of a Neighbor Feature based Matrix Factorization model(NF?MF),a Neighbor Compensation based Hybrid Collaborative Filtering(NC?HCF)and a Global Feature based Matrix Factorization model(GF?MF),and finally the combined model is named FE?EM(Feature Extract based Ensemble Model).The sub-model NF?MF reconstructs the feature matrix of users and services by introducing the convolutional neural network,and then predict QoS values.The sub-model NF?HCF uses the restricted Boltzmann machine model to predict QoS values of the candidate neighbor sets and uses the the improved similarity calculation to obtain the final prediction results.The sub-model GF?MF uses the variational auto-encoder model to mine the global features and reconstructs the feature matrix to obtain the prediction results.The combined model FE?EM aggregates the results of the three sub-models to obtain the results.(3)This paper uses the WSDream dataset,which is widely used in QoS prediction accuracy evaluation.It contains 339 users and 5825 services,including two types of QoS attributes(response time and throughput).Among them,this paper mainly predicts the response time attribute of the data set.the proposed model is superior to the compared traditional and state-of-the-art methods in prediction accuracy and has high stability regarding parameter settings.