Computer Engineering and Applications ›› 2024, Vol. 60 ›› Issue (9): 181-187.DOI: 10.3778/j.issn.1002-8331.2212-0197

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Temporal Knowledge Graph Reasoning with Graph Reconstruction

XU Zhihong, ZHANG Tianrun, WANG Liqin, DONG Yongfeng   

  1. 1.School of Artificial Intelligence & Data Science, Hebei University of Technology, Tianjin 300401, China
    2.Hebei Key Laboratory of Big Data Computing, Tianjin 300401, China
    3.Hebei Engineering Research Center of Data-Driven Industrial Intelligent, Tianjin 300401, China
  • Online:2024-05-01 Published:2024-04-29

融合图谱重构的时序知识图谱推理

许智宏,张天润,王利琴,董永峰   

  1. 1.河北工业大学 人工智能与数据科学学院,天津 300401
    2.河北省大数据计算重点实验室,天津 300401
    3.河北省数据驱动工业智能工程研究中心,天津 300401

Abstract: To address the problem that most existing temporal knowledge mapping algorithms are based on static knowledge mapping snapshot sequences and cannot adequately capture fine-grained temporal features, a graph reconstruction for temporal knowledge reasoning (GRTKR) model based on mapping reconstruction is designed. The model completes the temporal knowledge graph reconstruction by sampling the temporal neighbourhood of entities, and combines the explicit temporal features provided by the temporal encoder with the implicit temporal features provided by the neighbourhood feature aggregator to improve the modelling capability of the temporal data. Experiments on the temporal knowledge graph datasets ICEWS14, ICEWS05-15, and YAGO11K validate the effectiveness of the method and show significant improvements in MRR, Hits@1, Hits@3, and Hits@10 evaluation metrics compared to the mainstream baseline model.

Key words: temporal knowledge graph, reasoning, graph reconstruction, graph convolutional network, gated recurrent unit

摘要: 针对现有时序知识图谱模型多数基于静态知识图谱快照序列进行推理,无法充分捕获细粒度时序特征的问题,设计了基于图谱重构的时序知识图谱推理模型(graph reconstruction for temporal knowledge reasoning,GRTKR)。该模型通过对实体的时间邻域进行采样完成时序知识图谱重构,结合时间编码器提供的显式时序特征与邻域特征聚合器提供的隐式时序特征来提升对时序数据建模的能力。在时序知识图谱数据集ICEWS14、ICEWS05-15、YAGO11K上的实验验证了方法的有效性,并且相比于主流基线模型,MRR、Hits@1、Hits@3、Hits@10评价指标均有明显提升。

关键词: 时序知识图谱, 推理, 图谱重构, 图卷积神经网络, 门控循环单元