An Evaluation Model for Node Influence Based on Heuristic Spatiotemporal Features.

Area of Science:

• Network Science
• Complex Systems Analysis
• Data Mining

Background:

• Accurate node influence assessment is crucial for system stability.
• Traditional static methods struggle with dynamic network evolution and structural redundancy.
• Empirical networks exhibit dynamic characteristics and topological deviations upon copying.

Purpose of the Study:

• To propose a novel heuristic-based spatiotemporal feature node influence assessment model (HEIST).
• To address the limitations of traditional methods in capturing dynamic node influence.
• To enhance the accuracy and robustness of node influence assessment in empirical networks.

Main Methods:

• Optimized network copying using the zero-model method to reduce redundancy.
• Divided networks into subnets for feature modeling and influence differentiation.
• Employed a spatiotemporal depth-perception module integrating Graph Convolutional Neural Networks (GCN) and Long Short-Term Memory (LSTM) networks.
• Utilized a heuristic assessment algorithm for nonlinear optimization of node influence.

Main Results:

• The HEIST model demonstrated robust performance in capturing the dynamic evolution of node influence.
• Integration of GCN and LSTM improved spatial and temporal perception of node influence.
• Experimental results showed Kendall coefficients exceeding 90% across multiple datasets.
• The model exhibited strong generalization performance on empirical networks.

Conclusions:

• The proposed HEIST model effectively assesses dynamic node influence in complex networks.
• HEIST offers a significant improvement over traditional static assessment methods.
• The model's spatiotemporal feature extraction enhances accuracy and robustness in real-world network analysis.