An Adaptive Framework for Temporal Knowledge Graph Reasoning with Dynamic Event-Driven Dependencies

Temporal knowledge graph reasoning plays an important role in dynamic systems, event prediction, and complex temporal analysis. However, existing reasoning methods often struggle with challenges such as long-term temporal dependencies, data sparsity, and burst events. To address these issues, we propose an Adaptive Framework for Temporal Knowledge Graph Reasoning with Dynamic Event-Driven Dependencies (ADED). ADED consists of three core modules: the temporal adaptive graph convolution module, which dynamically adjusts graph convolution weights to accommodate the temporal characteristics of different time periods; the temporal multimodal fusion module, which handles both explicit and implicit temporal relationships in heterogeneous data sources; and the event-driven variable-length dependency module, which identifies burst events and automatically adjusts the reasoning window, enhancing the model's adaptability to temporal changes and anomalies. Experimental results show that ADED significantly outperforms existing temporal reasoning models across multiple benchmark datasets, excelling not only in prediction accuracy but also in interpretability and computational efficiency.