Effective and efficient identifying influential nodes in large scale networks by structural entropy

Abstract

Identifying influential nodes in large-scale networks is a pivotal challenge in network analysis. Traditional node identification methods, such as those based on node degree, primarily emphasize local neighbors without considering a node’s global importance within the network. Conversely, global feature-based methods like betweenness centrality (BC) are computationally prohibitive for large-scale networks. To address these limitations, we propose a novel community-level node influence calculation method grounded in structural entropy. This approach integrates both local significance within a community and global influence across communities. The method begins by employing community detection algorithms to cluster closely related nodes into communities. Subsequently, node influence is quantified by analyzing changes in structural entropy resulting from a node’s departure from its community and its integration into an adjacent community. Experimental evaluations on eleven real-world networks demonstrate that our method reduces the computational time for influential node identification by a factor of 76 compared to BC and other conventional approaches. Furthermore, in a network comprising seventy thousand nodes, our method enhances network efficiency by 20% relative to the LE method, underscoring its efficiency and effectiveness.