A New Analysis Method for Simulations Using Node Categorizations

TL;DR

This paper introduces a novel node categorization analysis method for network phenomena, utilizing local statistics and self-organizing maps to better understand large-scale network behaviors beyond global metrics.

Contribution

It presents a new analysis approach combining local network statistics with SOM-based categorization to visualize phenomena on networks, especially large-scale ones.

Findings

Method effectively visualizes network phenomena by node categories.

Validated using two simulation models.

Improves understanding of local structure effects on network dynamics.

Abstract

Most research concerning the influence of network structure on phenomena taking place on the network focus on relationships between global statistics of the network structure and characteristic properties of those phenomena, even though local structure has a significant effect on the dynamics of some phenomena. In the present paper, we propose a new analysis method for phenomena on networks based on a categorization of nodes. First, local statistics such as the average path length and the clustering coefficient for a node are calculated and assigned to the respective node. Then, the nodes are categorized using the self-organizing map (SOM) algorithm. Characteristic properties of the phenomena of interest are visualized for each category of nodes. The validity of our method is demonstrated using the results of two simulation models. The proposed method is useful as a research tool to understand the behavior of networks, in particular, for the large-scale networks that existing visualization techniques cannot work well.