Research On Graph-based QoS Prediction For Cloud Service Recommendation

With the rise of Internet and cloud computing,the demand of a large number of industry applications promotes the rapid development of cloud service market.In recent years,there are many cloud services with similar functions,but different quality of service(QoS).QoS reflects the non-functional characteristics of services.In the current situation of highly homogeneous services,cloud service recommendation technology based on QoS has been proved to be an effective means of screening and evaluating cloud services.However,due to the large number of cloud services and expensive calls,QoS is extremely sparse for ordinary users,so predicting QoS is the first problem to be solved in cloud service recommendation.In recent years,many scholars have proposed a large number of QoS prediction methods.However,the existing methods are still lack of effective information integration,and the deep mining of the relationship between information in the face of growing cloud services and dynamic network environment,so it is unable to accurately predict the QoS.Inspired by the graph structure,which can effectively integrate a variety of information and reflect the indirect and non indirect connections between nodes,this paper studies the QoS prediction problem in cloud service recommendation based on graph model.The main work and innovation can be summarized into the following two parts:(1)Research on QoS prediction based on multi-source information.In order to make efficient use of multi-source information and deeply explore the relationship between multi-source information,this paper proposes a graph based matrix factorization(GMF)QoS prediction method.Firstly,a variety of information reflecting user characteristics,service characteristics,and user service interaction characteristics are collected and quantified to build a full graph model;Secondly,in order to further enhance the strong connection between nodes and reduce the interference of noise edge,this paper uses the graph partition algorithm to cut the integrated-graph into several sub-graphs;Finally,the improved PMF algorithm is used to predict the QoS on the sub-graph.The experimental results confirm that GMF algorithm can improve the QoS prediction accuracy by integrating multi-source information through graph model.(2)Research on QoS prediction based on adaptive fusion.In order to further improve the accuracy of QoS prediction,this paper proposes a graph-based matrix factorization with Gaussian mixture model(GMF-GMM).Firstly,PMF is applied to obtain the global QoS prediction value to make full use of the global characteristics;Secondly,in order to deeply explore the potential relationship between global and local QoS,this paper introduces Gaussian mixture distribution to complete the global and local adaptive fusion.The experimental results show that compared with GMF,GMF-GMM significantly improves the QoS prediction accuracy,which proves the necessity of introducing global information into the prediction process;Furthermore,compared with the traditional fusion method by adjusting parameters,GMF-GMM is more time-saving and more accurate by adaptive fusion.