A Node Importance Evaluation Method Based on Graph-Transformer

Real-world systems can be abstracted as complex networks. Node importance evaluation is a commonly used network analysis technique. Evaluating node importance positively contributes to the development of subsequent research in networks. Considering the challenges of long-range dependencies between nodes being compressed in graph neural networks, this paper proposes a Node Importance Evaluation Method based on Graph-Transformer (NIE-GT). The Graph-Transformer model is used for node representation learning, effectively capturing dependencies between nodes. The method is applied to the Cora, P2p-Gnutella08, and Plant datasets through the Susceptible Infected Recovered (SIR) model and network robustness experiments for assessment. Results show that the important nodes evaluated by the proposed method have stronger information dissemination capabilities and gain more accurate importance scores than other methods. Ablation experiments demonstrate that utilizing a global attention mechanism improves the effectiveness of node importance evaluation.