![[2602.12389] Evolving Beyond Snapshots: Harmonizing Structure and Sequence via Entity State Tuning for Temporal Knowledge Graph Forecasting](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.12389] Evolving Beyond Snapshots: Harmonizing Structure and Sequence via Entity State Tuning for Temporal Knowledge Graph Forecasting
Summary
This paper presents Entity State Tuning (EST), a novel framework for improving temporal knowledge graph forecasting by maintaining persistent entity states, enhancing prediction accuracy over long horizons.
Why It Matters
Temporal knowledge graphs are crucial for various AI applications, including recommendation systems and event prediction. The proposed EST framework addresses the limitations of existing methods that struggle with long-term dependencies, thereby advancing the field of AI forecasting and improving model performance.
Key Takeaways
- Entity State Tuning (EST) enhances temporal knowledge graph forecasting.
- EST maintains persistent entity states to improve long-term predictions.
- The framework integrates structural and sequential data effectively.
- Experiments demonstrate EST's superiority over existing methods.
- The approach balances stability and adaptability in forecasting.
Computer Science > Artificial Intelligence arXiv:2602.12389 (cs) [Submitted on 12 Feb 2026] Title:Evolving Beyond Snapshots: Harmonizing Structure and Sequence via Entity State Tuning for Temporal Knowledge Graph Forecasting Authors:Siyuan Li, Yunjia Wu, Yiyong Xiao, Pingyang Huang, Peize Li, Ruitong Liu, Yan Wen, Te Sun, Fangyi Pei View a PDF of the paper titled Evolving Beyond Snapshots: Harmonizing Structure and Sequence via Entity State Tuning for Temporal Knowledge Graph Forecasting, by Siyuan Li and Yunjia Wu and 7 other authors View PDF HTML (experimental) Abstract:Temporal knowledge graph (TKG) forecasting requires predicting future facts by jointly modeling structural dependencies within each snapshot and temporal evolution across snapshots. However, most existing methods are stateless: they recompute entity representations at each timestamp from a limited query window, leading to episodic amnesia and rapid decay of long-term dependencies. To address this limitation, we propose Entity State Tuning (EST), an encoder-agnostic framework that endows TKG forecasters with persistent and continuously evolving entity states. EST maintains a global state buffer and progressively aligns structural evidence with sequential signals via a closed-loop design. Specifically, a topology-aware state perceiver first injects entity-state priors into structural encoding. Then, a unified temporal context module aggregates the state-enhanced events with a pluggable sequence backbone. Su...