A Temporal-Aware Dynamic Graph Neural Network for Learning Evolving Graph Representations

Many computer science studies, including communication networks, software dependency graphs, transaction systems and knowledge graphs produce dynamic graphs in which nodes and edges are continuously updated. Standard Graph Neural Networks (GNNs) are mainly designed for static network architectures and do not accurately capture temporal dynamics or structural changes. When designing real-time dynamic systems, this issue includes data loss and low performance. To solve this issue, this study introduces a Temporal-Aware Dynamic Graph Neural Network (TDGNN) that generates both structural and temporal representations from building graphs. The method combines temporal encoding with graph convolution to maintain previous connections to change the new nodes and edges. A memory-based temporal aggregation method is provided to maintain long-term data knowledge without retraining the model from the initial stage. The experimental evaluation of standard dynamic graph datasets proposed TDGNN performs traditional static and snapshot-driven GNN models in terms of accuracy and stability. The results show that implementing temporal awareness significantly improves representation learning for dynamic graph data, make the proposed method suitable for next-generation graph-based software systems.