Sparse QoS prediction via effective mining topology and collaborative features

Collaborative quality of services (QoS） prediction has gained considerable attention lately, focusing on predicting missing values based on past user experiences with cloud services. To address the problem of improving QoS prediction accuracy under sparse data scenarios, a topology and collaborative feature encoder architecture (TCFE) consisting of two specially tailored components is proposed for sparse QoS prediction. First, the topology feature encoder component considers network topology information and user-to-service communication path features to improve network topology modeling and path extraction; then the collaborative feature encoder component enhances user/service embedding on the user-service interaction bipartite graph by using graph convolutional networks, and finally the Kolmogorov-Arnold network is used to achieve the effective integration and extraction of collaboration and contextual features. Experimental results show that TCFE outperforms some state-of-the-art QoS prediction methods in terms of accuracy and efficiency.