QoSGNN: boosting QoS prediction performance with Graph Neural Networks

QoS (Quality of Service) prediction is a fundamental task for downstream service computing tasks. Yet, how to predict QoS still remains a non-trivial issue. Existing works try to address this problem by either better modeling the impact between users and services or learning better user/service feature embeddings. Despite the challenges, modeling the two aspects jointly can improve the performance of QoS prediction. However, existing works have rarely attempted this approach. In this regard, Graph Neural Networks (GNNs), with their ability to model these two aspects simultaneously, present an exciting opportunity to advance QoS prediction. Unfortunately, GNNs have not yet been systematically studied in this context, and their potential for improving QoS prediction warrants further investigation. In this article, we apply GNN to the QoS prediction task by proposing a framework called QoSGNN. In this framework, we systematically explore all steps required when applying GNN to QoS prediction and discuss each step's design principles and advantages. Based on these design principles, we provide an implementation of QoSGNN and conduct extensive experiments on a real-world QoS dataset. The experimental results show that our proposed QoSGNN not only significantly outperforms the baselines in terms of QoS prediction performance but also substantially improves other aspects such as cold start, scalability, robustness, and fairness.