Community Detection from Aggregated Relational Data and Graph Neural Networks

S. Díaz-Aranda, J. Aguilar, R. E. Lillo

Modeling complex systems through network representations is essential for understanding interaction patterns, but collecting complete network data is often infeasible due to privacy or logistical constraints. This work aims to detect network communities using aggregated relational data (ARD), the number of links between a node and its neighbors with a specific feature. This paper proposes a methodology for using synthetic graphs estimated from ARD as input for machine learning models. The steps of the methodology comprise the generation of estimated graphs using ARD, validation and correction of the synthetic graphs with Network Scale-up Methods and link prediction techniques, node embedding learning via Graph Neural Networks, and community detection using clustering algorithms. The methodology is applied in simulated environments and real data, showing the effectiveness of the proposed framework in reconstructing network structures and accurately identifying community patterns.

Keywords: Network Community Detection Graph Neural Networks Aggregated Relational Data

J. A. Cardenas Cardenas, P. Delicado, J. Arjona

P. González Barquero, J. Goldsmith, Á. Méndez Civieta, R. E. Lillo

P. Morala Miguélez, E. Gambín Monserrat, I. Úcar Marqués