InfGCN: Identifying influential nodes in complex networks with graph convolutional networks

Abstract

Identifying influential nodes in a complex network is very critical as complex networks are ubiquitous. Traditional methods, such as centrality based methods and machine learning based methods, only consider either network structures or node features to evaluate the significance of nodes. However, the influential importance of nodes should be determined by both network structures and node features. To solve this problem, this paper proposes a deep learning model, named InfGCN, to identify the most influential nodes in a complex network based on Graph Convolutional Networks. InfGCN takes neighbor graphs and four classic structural features as the input into a graph convolutional network for learning nodes’ representations, and then feeds the representations into the task-learning layers, comparing the ground truth derived from Susceptible Infected Recovered (SIR) simulation experiments with quantitative infection rate. Extensive experiments on five real-world networks of different types and sizes demonstrate that the proposed model significantly outperforms traditional methods, and can accurately identify influential nodes.