Graph neural network method for detecting coordination structures in network data

Abstract

The purpose of the article is to develop a graph neural network–based method for detecting coordination structures in network data based on graph topology analysis and machine learning techniques.
Research methodology. The study is based on representing network data as an interaction graph, where vertices correspond to objects and edges represent their relationships. Structural representations of graph vertices are obtained using a GraphSAGE [4] graph neural network architecture, which aggregates information from the local neighborhood. The subsequent identification of coordination structures is performed using the density-based clustering algorithm DBSCAN [3] without an a priori specification of the number of clusters.
Research results. The proposed method was evaluated on a synthetic network graph that reproduces key properties of real-world systems. Experimental results demonstrate high effectiveness of the approach: the average values of precision, recall, and F1-score for detecting coordination structures were 0.91, 0.98, and 0.94, respectively. The obtained results confirm the ability of the method to reliably identify dense coordination groups even in the absence of explicit individual anomalies.
Practical significance. The proposed graph neural network–based method is universal, does not require fully labeled data, and can be integrated into large-scale network data analysis systems for detecting coordination structures in applied monitoring and information analytics tasks.