Measurement of influential nodes in networks based on community structure information entropy

Identifying influential nodes in complex networks is critical for regulating information dissemination and mitigating disasters induced by misinformation, yet current methods for detecting such nodes typically rely exclusively on local topological features or global positional information-failing to leverage critical insights from neighbors with similar community memberships (i.e., nodes within the same community share the same community affiliation) in the network's inherent community structure, which frequently results in low recognition accuracy and poor generalization across diverse datasets. To address this gap, we propose a novel semi-local metric, Local Community Structure Entropy (LCE), for identifying influential nodes in complex networks: LCE fully integrates information entropy with the network's intrinsic community architecture by incorporating the connectivity magnitude of community structures, and specifically, we first partition the network into distinct communities using a community detection algorithm, then calculate each node's entropy centrality based on first-order neighbor information, before further extending the analysis to second-order neighbor data to derive a comprehensive entropy centrality index. To validate LCE's performance, we conduct extensive experiments on both social and synthetic networks, employing multiple evaluation metrics (the Susceptible-Infected-Recovered (SIR) model, Kendall's correlation coefficient, complementary cumulative distribution function (CCDF), community partitioning effectiveness, spreading efficiency, and computational complexity) and performing comparative tests against seven benchmark algorithms (Degree Centrality (DC), Betweenness Centrality (BC), K-shell decomposition (KS), Chance-Based Centrality (CBC), Community-Based Measure (CBM), Intra-Community Centrality (ICC), and Entropy-Renewal Heuristic (EnRenew)), and experimental results demonstrate that LCE outperforms all competing methods in terms of recognition capability while simultaneously improving efficiency and universality.