Temporal knowledge graph (TKG) forecasting is a rapidly evolving field, crucial for applications requiring understanding and prediction of events over time. A new study introduces an innovative approach to address the limitations of existing methods, often plagued by "episodic amnesia" and decay of long-term dependencies.

Entity State Tuning (EST)

The proposed framework, called Entity State Tuning (EST), is based on the idea of endowing TKG forecasting systems with persistent and continuously evolving entity states. EST maintains a global state buffer and progressively aligns structural evidence with sequential signals through a closed-loop design.

Architecture and Operation

The EST architecture includes a topology-aware state perceiver module that injects entity-state priors into structural encoding. Subsequently, a unified temporal context module aggregates the state-enhanced events with a pluggable sequence backbone. A dual-track evolution mechanism writes the updated context back to the global entity state memory, balancing plasticity and stability.

Results and Performance

Experimental results on multiple benchmarks show that EST consistently improves diverse backbones and achieves state-of-the-art performance, highlighting the importance of state persistence for long-horizon TKG forecasting. The project code is available on GitHub.