Dynamic Heterogeneous Graph Representation Learning: A Survey

The study provides a systematic review of dynamic heterogeneous graph representation learning, covering modeling methods for the evolution of nodes, edges, and attributes over time. The article summarizes the paradigm shift from static graphs to dynamic graphs, focusing on analyzing information dissemination mechanisms and time-dependent modeling techniques in heterogeneous graph structures. The core contents include attention-based dynamic update strategies, multi-source heterogeneous data fusion algorithms, and lightweight representation schemes for real-time reasoning. The review also compares the performance of mainstream model architectures (such as Temporal Graph Networks and Dynamic Heterogeneous GNNs) in tasks such as node classification, link prediction, and anomaly detection, and explores the challenges and future directions in causal inference and interpretability.