Multi-Task Web QoS Prediction Based On Graph Neural Network

Currently,both individuals and enterprises are shifting their data storage methods from physical storage to cloud storage,which originates from the joint efforts of industry and academia in the field of cloud computing technology.The rapid development of cloud computing makes the number of Web services grow exponentially,and the screening of Web services becomes more and more complex.Quality of Service(QoS)is an important means to screen Web services,but QoS data is very sparse in practical scenarios,and the number of QoS observed by users themselves is limited,so designing a QoS prediction method is beneficial to both cloud The design of a QoS prediction method is therefore beneficial for cloud service providers to save cost and improve user experience.The existing prediction algorithms,such as collaborative filtering and matrix decomposition,do not facilitate the hidden graph structure information between users and services,resulting in limited prediction accuracy.This paper mines the non-Euclidean structure between users and services,and proposes two QoS prediction algorithms for different application scenarios based on graph neural networks and multi-task learning.For the static QoS prediction problem,this paper proposes GCN-M,a multi-task QoS prediction algorithm based on graph convolution,which conveys the information of nodes in the user and service graph structures through the graph convolution network,learns the information of interaction features of users and services,and explores the association information of two subtasks by sharing shallow parameters,so that the trained model can predict two QoS in one calculation and improve the prediction work efficiency.After a large number of comparison and ablation experiments as well as special parameter analysis,GCN-M can effectively improve the prediction accuracy with an average MAE improvement of about 13.08% on the two subtasks,while the RMSE improvement effect is about 5.74% on average.For the dynamic QoS prediction problem,this paper proposes the time-aware graph attention multi-task QoS prediction algorithm T-GAT-M,which gives different node weights in the bipartite graph of users and services through GAT graph neural network,so that users can discover more suitable services for themselves,and combines GRU to learn relevant features on the time domain and control which feature information can be input to the next layer of the network,through A large number of comparative experiments and parameter analysis,T-GAT-M is more accurate than the traditional time-series quality of service prediction algorithm,on two subtasks,on two subtasks MAE can be improved by 17.08% on average than other models,and due to the RMSE improvement effect is13.58% on average.