CID-TKG: Collaborative Historical Invariance and Evolutionary Dynamics Learning for Temporal Knowledge Graph Reasoning
Summary: arXiv:2604.09600v1 Announce Type: new
Abstract
Temporal knowledge graph (TKG) reasoning aims to infer future facts at unseen timestamps from temporally evolving entities and relations. Despite recent progress, existing approaches still suffer from inherent limitations due to their inductive biases, as they predominantly rely on time-invariant or weakly time-dependent structures and overlook the evolutionary dynamics. To overcome this limitation, we propose a novel collaborative learning framework for TKGR (dubbed CID-TKG) that integrates evolutionary dynamics and historical invariance semantics as an effective inductive bias for reasoning.
Key Features of CID-TKG
- Historical Invariance Graph: This component captures long-term structural regularities within the temporal knowledge graph.
- Evolutionary Dynamics Graph: It models short-term temporal transitions to better understand how entities and relations evolve over time.
- Dedicated Encoders: CID-TKG employs specialized encoders to learn representations from both the historical invariance graph and the evolutionary dynamics graph.
- Contrastive Objective: To enhance the alignment between different views, the framework decomposes relations into view-specific representations, promoting cross-view consistency while minimizing view-specific noise.
Motivation Behind CID-TKG
Traditional approaches to TKG reasoning often rely on static or weakly dynamic structures, which can limit the model’s ability to accurately predict future facts. The need for a methodology that can account for both historical invariance and rapid evolutionary changes in data is paramount. CID-TKG addresses this gap by providing a framework that is robust against the limitations present in existing models.
Methodology
The CID-TKG framework operates by first constructing two distinct graphs: one for historical invariance and another for evolutionary dynamics. Each graph is analyzed separately through dedicated encoders, which ensure that the unique characteristics of each structure are captured effectively. The learning process is further enhanced by the contrastive objective that aligns representations from both graphs, thus ensuring that the model maintains semantic coherence across different views.
Experimental Results
Extensive experiments have been conducted to evaluate the performance of CID-TKG. The results indicate that CID-TKG achieves state-of-the-art performance under extrapolation settings, outperforming existing models that do not integrate historical invariance and evolutionary dynamics. This demonstrates the effectiveness of the proposed framework in improving TKG reasoning capabilities.
Conclusion
In conclusion, CID-TKG represents a significant advancement in the field of temporal knowledge graph reasoning. By addressing the limitations of previous models and providing a robust framework that integrates both historical and evolutionary aspects, CID-TKG paves the way for more accurate and reliable predictions in dynamic environments.