TD-GCN: A novel fusion method for network topological and dynamical features

Abstract

The topological and dynamical features of complex networks hold abundant information. How to fully utilize this information for more accurate network structure mining is a significant issue. In this paper, we propose a novel method that simultaneously takes into account both network topological and dynamical features via graph convolutional networks (TD-GCN). Specifically, we obtain the topological features of the network by using the second-order adjacency matrix of the complex network, which captures indirect connections between nodes, for a more detailed representation of network structure, and use the SIS model to generate node state data in the complex network as the dynamical features of the network. The network topological and dynamical features are fused through the graph convolutional neural network. To verify the effectiveness and applicability of our method, we conduct extensive experiments on both simulated networks and real-world networks with various network scales. We comprehensively compare the proposed method with other existing methods in the domains of network link prediction and network node ranking learning. The experimental results show that our method can better capture the characteristic information in complex networks and has better performance compared with other methods.