Predicting nodal influence via local iterative metrics.

Area of Science:

• Network Science
• Complex Systems
• Data Analysis

Background:

• Nodal spreading influence is key to network activation.
• Combining local and global network information improves influence prediction.
• Previous methods often rely on single centrality metrics.

Purpose of the Study:

• Investigate the contribution of local vs. global topological information to nodal influence prediction.
• Determine if relatively local network information is sufficient for accurate influence prediction.
• Propose and evaluate an iterative metric set for predicting nodal influence.

Main Methods:

• Leveraged iterative processes similar to eigenvector centrality calculation.
• Defined an iterative metric set (order 1 to K) encoding progressively global information.
• Evaluated three iterative metrics converging to classical centrality metrics.
• Tested on real-world, synthetic, and spatially embedded networks.

Main Results:

• Prediction quality rapidly converges with low-order iterative metrics (K < 5).
• Further increases in K yield only marginal improvements in prediction quality.
• The best iterative metric set (K < 5) rivals benchmark methods using seven centrality metrics.
• Spatially embedded networks with large diameters require higher-order iterative metrics for comparable performance.

Conclusions:

• Iterative metric sets offer efficient and effective prediction of nodal spreading influence.
• Relatively local network information, captured by low-order iterative metrics, is often sufficient.
• The proposed iterative approach provides a competitive alternative to complex, multi-metric methods.