New AI framework CID-TKG improves temporal knowledge graph reasoning by simultaneously learning historical patterns and evolutionary dynamics

CID-TKG addresses limitations in existing temporal knowledge graph reasoning approaches that rely too heavily on time-invariant structures

The framework uses dual-graph approach: historical invariance graph captures long-term structural regularities while evolutionary dynamics graph models short-term temporal transitions

Dedicated encoders learn representations from each graph structure separately to better handle temporal evolution of entities and relations

Aims to reduce semantic discrepancies between historical patterns and evolutionary changes for more accurate future fact inference at unseen timestamps