HiTS-CL: A Continual Learning Framework for Long-Horizon Temporal Knowledge Graph Extrapolation

2026-09-28

AI Summary

This study addresses the performance degradation in temporal knowledge graph reasoning (TKGR) extrapolation over long-term non-stationary data, which arises from static prefix training. To tackle this issue, we propose HiTS-CL, a novel framework that pioneers formulating the extrapolation task as a continual learning paradigm. Specifically, our method synergistically integrates current dynamic evolution, stable knowledge, and historical evidence through continual fine-tuning, multi-teacher adaptive distillation, and a selective fact memory mechanism. Experimental results demonstrate that HiTS-CL significantly improves extrapolation accuracy and effectively mitigates long-term degradation, outperforming mainstream baseline methods.

Abstract

Extrapolative temporal knowledge graph reasoning (TKGR) predicts future facts from historical snapshots. Most existing methods train once on an early prefix of the timeline and then use a frozen model for all future timestamps. We argue that this fixed-prefix protocol is misaligned with extrapolation. It learns from a static prefix, whereas the target stream is non-stationary: new entities and facts emerge, temporal dependencies shift across regimes, and recurring historical signals must be refreshed online. As a result, models trained only on early snapshots become outdated and degrade over long horizons. We address this mismatch by formulating extrapolative TKGR as continual learning over streaming snapshots. Under this view, effective extrapolation must jointly handle current dynamics, stable knowledge, and recurring historical evidence. Based on these requirements, we propose History-enhanced Two-Step Continual Learning (HiTS-CL), a backbone-agnostic continual learning framework for extrapolative TKGR. HiTS-CL tracks current dynamics via continual fine-tuning, preserves stable knowledge via multi-teacher adaptive distillation, and retains recurring historical evidence via a selective memory of recent and frequent facts. We integrate HiTS-CL into five representative TKGR backbones and evaluate it on four benchmark datasets. HiTS-CL consistently improves extrapolation accuracy, reduces long-horizon degradation, and outperforms strong continual-learning baselines, including a recent method for temporal knowledge graphs. Source code and data are available at https://github.com/liuyansong98/HiTS-CL.

Problem

• Temporal Knowledge Graph Reasoning
• Extrapolation
• Continual Learning
• Non-stationary Data Streams
• Long-Horizon Degradation

Innovation

• Continual Learning
• Temporal Knowledge Graph Extrapolation
• Multi-teacher Adaptive Distillation
• Selective Memory
• Backbone-agnostic Framework