用可解释的符号规则实现时间知识图谱预测，效果媲美顶尖模型。
CountTRuCoLa: Rule Learning for Interpretable Temporal Knowledge Graph Forecasting
- 基于四种简单规则类型，融合时效性与频率的置信度函数。
- 在九个数据集上表现接近顶尖模型，多数情况下更优。
- 预测结果可追溯至具体规则和观测，适合需要可解释性的场景。
我们提出一种基于符号规则的可解释方法，用于时间知识图谱预测。受近期基于循环事实的强基线启发，该方法学习四类简单规则，包括结合时效性与频率的置信度函数的时间规则。在九个数据集上的评估显示，该方法性能与最先进模型相当，并超越大多数同类模型，且每次预测均可直接追溯到生成它的规则与观测。此外，该方法在超大数据集上仍能运行，而其他方法常因运行时或内存问题失效。
原文摘要 · Abstract (English)
We address the task of temporal knowledge graph forecasting with an inherently interpretable method based on symbolic rules. Motivated by recent work proposing a strong baseline based on recurrent facts, our approach learns four simple rule types, including temporal rules with confidence functions that combine both recency and frequency. Evaluated on nine datasets, our method achieves performance that is competitive with state-of-the-art models and outperforms the majority of them, while each prediction remains directly traceable to the rules and observations that produced it. Moreover, our approach remains functional on very large datasets, where other methods encounter runtime or memory failures.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。