Graph convolutional network for structural equivalent key nodes identification in complex networks

Abstract

Identifying key influential nodes in complex networks is crucial for applications such as social network analysis, epidemiology, and recommendation systems. This paper proposes SE_GCN (Structural Equivalence with Graph Convolutional Network), a method that combines structural equivalence with Graph Convolutional Networks (GCNs) to identify key nodes in complex networks. SE_GCN leverages structural similarities among nodes at various hop distances to construct a comprehensive feature matrix, which is directly used for node embedding. GCNs are employed to process this feature matrix, learning effective representations of nodes within the network. The fully connected layer of SE_GCN computes the embedded score of each node, and a sigmoid function predicts the influential probabilities of nodes. The performance of SE_GCN is evaluated by comparing it with the Susceptible-Infected-Recovered (SIR) epidemiological model, Kendall's tau correlation, and Jaccard similarity. The proposed method is assessed using baseline methods in terms of infection rate, seed set size, correlation coefficient, and similarity index across several synthetic and real-world networks. The results demonstrate that SE_GCN outperforms existing methods, highlighting its effectiveness and robustness in identifying influential nodes.