Node Influence Evaluation Through K-Shell Convoluted Proximal Aggregation
Yongji Liu, Boyuan Zhu, Fansong Chen, Weicheng Lin, Haiqiang Fei, Hongsong Zhu · 2024
Optimizing methodologies for identifying vital nodes within complex networks is an active field of study. Our findings indicate that although the traditional techniques for static undirected complex networks are effective in vital node identification, opportunities persist for enhancing both node ranking accuracy and resolution. In response, we introduce a novel approach, K-Shell Convoluted Proximal Aggregation (KSCPA). This novel method incorporates the convolution characteristic from Graph Convolutional Networks (GCNs), enabling localized node evaluation through the aggregation of adjacent node information. To evaluate the proposed method, we employed Kendall's Tau correlation index to compare with the Susceptible Infected Recovered model (SIR) for ranking accuracy and used a monotonicity function to assess ranking resolution. At the end, empirical analysis across 12 real-world network datasets reveals that KSCPA outperforms established techniques in ranking accuracy and resolution, thereby confirming its efficacy in identifying key nodes in complex networks.