Top influencers can be identified universally by combining classical centralities

Area of Science:

• Network Science
• Computational Social Science
• Epidemiology

Background:

• Information, opinions, and epidemics spread through structured networks.
• Identifying key nodes (superspreaders) is crucial for understanding and controlling these processes.
• No single node centrality measure consistently ranks top influencers across different network types.

Purpose of the Study:

• To develop a consistently predictive method for identifying superspreaders in diverse networks.
• To investigate the synergistic effect of combining multiple centrality measures.
• To improve the accuracy of predicting influential nodes.

Main Methods:

• Utilized statistical classifiers trained on multiple node centrality indicators.
• Evaluated the predictive performance across various static, real-world network topologies.
• Compared the effectiveness of single centralities versus combined centrality approaches.

Main Results:

• Statistical classifiers using two or more centralities proved consistently predictive across diverse networks.
• Pairs of local and global centralities (e.g., neighborhood size with eigenvector centrality) showed strong cooperative predictive power.
• Superspreaders typically maximize both local and global centrality values.

Conclusions:

• Combining multiple centrality measures significantly enhances the prediction of network superspreaders.
• The interplay between local and global network properties is key to identifying influential nodes.
• Classifiers trained on seven classical indicators achieved near-maximum prediction accuracy (0.995).