Temporal-Aware QoS Prediction via Dynamic Graph Neural Collaborative Learning

Abstract

How to effectively predict missing QoS has become a fundamental research issue for service-oriented downstream tasks. However, most QoS prediction approaches omit high-order implicit invocation correlations and collaborative relationships among users and services. Thus, they are incapable of effectively learning the temporally evolutionary characteristics of user-service invocations from historical QoS records, which significantly affects the performance of QoS prediction. To address the issue, we propose a novel framework for temporal-aware QoS prediction by dynamic graph neural collaborative learning. Dynamic user-service invocation graph and graph convolutional network are combined to model user-service historical temporal interactions and extract latent features of users and services at each time slice, while a multi-layer GRU is applied for mining temporal feature evolution pattern across multiple time slices, leading to temporal-aware QoS prediction. The experimental results indicate that our proposed approach for temporal-aware QoS prediction significantly outperforms state-of-the-art competing methods.