NIE-GAT: node importance evaluation method for inter-domain routing network based on graph attention network

Evaluating node importance in inter-domain routing networks to protect key nodes is crucial to the stability of the Internet. Existing evaluation methods struggle to balance accuracy and speed: centrality-based methods are quick but lack accuracy, while impact-based methods are precise but have high computational complexity. Approaches in the complex network field do not consider the business relationships in the inter-domain routing network, so their results may not match reality. This paper proposes a novel method for node importance evaluation in the inter-domain routing networks, called NIE-GAT. The key idea of NIE-GAT is to map a node’s static features to its dynamic cascading failure impact by a deep learning model based on the graph attention network. Additionally, several node centrality features based on the business relationships are defined. A method for calculating node impact labels with the cascading failure model is also designed. The experimental results on real inter-domain routing networks of different scales show that the evaluation accuracy of NIE-GAT for the top 3% nodes reaches more than 93%, which is over 7% higher than the existing best method. For large-scale networks with more than ten thousand nodes, NIE-GAT can complete the evaluation within 2 min.