Historical facts learning from Long-Short Terms with Language Model for Temporal Knowledge Graph Reasoning

Wenjie Xu, Ben Liu, Miao Peng, Zihao Jiang, Xu Jia, Kai Liu, Lei Liu, Min Peng

Abstract: Highlights•A PLM-based framework HFL that can learn historical facts for TKGR is proposed.•HFL models time intervals with tokens and adopt a token-only input paradigm.•Short-term and long-term historical facts are incorporated when reasoning.•We evaluate HFL on four benchmark datasets and propose variants on LLMs.