A neural diffusion model for identifying influential nodes in complex networks

Abstract

Identifying influential nodes in complex networks through influence diffusion models is a challenging problem that has garnered significant attention in recent years. While many heuristic algorithms have been developed to address this issue, neural models that account for weighted influence remain underexplored. In this paper, we introduce a neural diffusion model (NDM) designed to identify weighted influential nodes in complex networks. Our NDM is trained on small-scale networks and learns to map network structures to the corresponding weighted influence of nodes, leveraging the weighted independent cascade model to provide insights into network dynamics. Specifically, we extract weight-based features from nodes at various scales to capture their local structures. We then employ a neural encoder to incorporate neighborhood information and learn node embeddings by integrating features across different scales into sequential neural units. Finally, a decoding mechanism transforms these node embeddings into estimates of weighted influence. Experimental results on both real-world and synthetic networks demonstrate that our NDM outperforms state-of-the-art techniques, achieving superior prediction performance.