A comparative study of methods for measuring node influence in complex networks

Identifying influential nodes in complex networks has attracted significant research attention due to its applications in diverse fields ranging from viral marketing to epidemics and behavioral analytics. Most of the existing algorithms primarily assess node influence by measuring their spreading power, encompassing a spectrum from structural centrality metrics to machine learning and graph neural network based approaches. This study first provides a comprehensive review of various influential node identification methods and subsequently conducts an extensive comparative analysis. Using a unified platform applied to multiple standard real-world networks, we evaluate the performance, predictive accuracy, and computational cost of these methods. Our findings indicate that no single algorithm consistently outperforms others across all network types. While learning-based methods, particularly graph neural network, generally achieve higher predictive accuracy, they exhibit lower interpretability and impose greater computational costs. Therefore, selecting an appropriate method necessitates a balance between complexity and accuracy, tailored to the specific application context.