Key node identification method based on hybrid centrality and graph neural network

Key node identification has become an important part of complex network research. However, existing key node identification methods suffer from low computational efficiency and insufficient robustness. This paper proposes to obtain a mixed centrality by combining the local and global influences of nodes, and after combining it with the graph convolutional neural network, a key node identification model is proposed. The model constructs multi-scale centrality features using mixed node centrality and builds two structural channel sets to embed neighborhood topology information. Subsequently, an attention mechanism is introduced to assign weights to different channels automatically. Finally, feature aggregation and regression prediction of nodes are carried out by the graph neural network. Experimental results on multiple real social networks and synthetic scale-free networks show that NLGCN outperforms classical centrality methods and existing graph learning models in terms of propagation ability, node ranking consistency, robustness, and generalization performance.