NR-GAT: A network risk assessment ethod based on graph attention networks

Security threats in the network are difficult to eradicate, and the use of vulnerability analysis methods to identify high-influence nodes in the network in advance in order to establish an effective security mechanism is crucial to maintaining the stability of the Internet. The centrality-based traditional methods and machine learning-based methods can only use the network structure characteristics or the characteristics of network nodes to mine the vulnerability nodes in the network by measuring the influence of nodes in the network, which has certain limitations. The vulnerability of network nodes not only depends on their own characteristics but is also closely related to the linking relationship and vulnerability of neighbouring nodes. To solve the problem, we propose a deep learning model NR-GAT based on a graph neural network to achieve the task of identifying the vulnerable nodes in the network. NR-GAT is based on graph structure learning, which takes into account the node features and the network structure information. Four types of risk features are used as inputs to the graph neural network, and the SIR (susceptible-infected-recovered) model is used to obtain the node influence labels, which are used to learn the deeply hidden representation of nodes. Through extensive experiments in three experimental regions, it is shown that our proposed model exhibits better performance than the baseline, providing a reference for identifying high-impact vulnerable nodes in networks.