DEVELOPMENT OF DYNAMIC GNN ARCHITECTURES FOR MODELING THE EVOLUTION OF SOCIAL NETWORKS: A COMPARATIVE ANALYSIS OF METHODS FOR PROCESSING TEMPORAL GRAPHS OF STRUCTURAL CHANGES IN SOCIAL NETWORKS IN REAL TIME

Popov Valeriy Vladislavovich

Social networks represent complex dynamic systems where connections between users and entities (posts, comments, communities) change in real time. Modeling such temporal structures for further analysis using classical mathematical modeling methods is challenging due to high dimensionality and the need to account for data additivity. Traditional graph neural networks (GNN – Graph Neural Network) demonstrate high efficiency in analyzing static graphs; however, their application is limited when working with temporal data that requires adaptation to changes in graph structure. This article examines modern approaches to the development of dynamic GNNs capable of modeling changes in social networks while considering temporal patterns. An architecture combining recurrent mechanisms and neighbor aggregation is proposed for processing streaming data. A comparative analysis of classical GNN and dynamic GNN (DGNN – Dynamic Graph Neural Network) modeling methods is conducted, with emphasis on their applicability in trend forecasting, community detection, and anomaly identification. Experiments on publicly available Twitter datasets show that dynamic GNNs achieve 8–10% higher accuracy compared to their static counterparts. Particular attention is paid to issues of scalability, privacy, and model interpretability. The research findings can be applied to develop secure and adaptive algorithms for real-time analysis of social networks.

Keywords:dynamic graph neural networks, temporal graphs, GNN, evolution modeling, real-time data processing, scalability, interpretability, DGNN, trend forecasting, community detection