Motif-GNN: A Novel Method for Key Node Identification in Complex Networks

Critical node identification is vital for network function, structural optimization, and risk management. Existing methods are inaccurate, as they overlook local structural features and inadequately model higher-order linkage patterns. This paper proposes Motif-GNN—a framework integrating motif structure analysis and convolutional neural networks: it enumerates network higher-order motifs to build a node-level weighted motif adjacency matrix, uses a balanced mechanism to fuse multi-scale topological information, designs a multi-channel convolutional layer for local structural feature extraction, and finally applies supervised learning to predict node importance scores. Tested on communication, power, and aviation network datasets against benchmark algorithms, Motif-GNN outperforms baselines significantly in ranking consistency and propagation influence prediction, offering a new feature-driven solution for critical node mining in complex networks.