Communication networks, such as social and collaborative networks, are characterized by a highly dynamic, constantly changing environment. T
Communication networks, such as social and collaborative networks, are characterized by a highly dynamic, constantly changing environment. This makes the analysis of such networks, such as the formation of communities, challenging. The adaptive temporal graph neural network (AT-GNN) was introduced here to overcome these challenges by incorporating temporal segmentation, feature extraction, and attention mechanisms. Based on two large-scale datasets, the Stanford Network Analysis Project (SNAP) and the Digital Bibliography and Library Project (DBLP), the AT-GNN model considers structural and temporal features for predicting community behaviors. Temporal segmentation was done through clustering while using node and edge attribute extraction. The preprocessing stage involved embedding layers, attention mechanisms, and recurrent layers. These components enabled the AT-GNN model to adjust the weight of essential relationships through dynamic networks, enhancing the explainability of community changes. A comparison was made between the proposed model and best-performing models, showing improved predictive accuracy of 98%, precision of 92%, recall of 95%, and F1-score of 93%. This work emphasizes the scalability, flexibility, and dynamism of the AT-GNN model and offers a starting point for studying dynamic systems. Future work will extend to graphs in continuous time and to enormously large networks, improving the model's effectiveness in real-time dynamic networks. These developments highlight the applicability of AT-GNN in various real-world settings, such as social, biological, and organizational networks.
