CRmod: Context-Aware Rule-Guided reasoning over temporal knowledge graph
Lin Zhu,Die Chai,Luyi Bai
DOI: https://doi.org/10.1016/j.ins.2024.120343
IF: 8.1
2024-04-01
Information Sciences
Abstract:Temporal knowledge graphs have been widely used in artificial intelligence, but they are still incomplete. Therefore, the reasoning task is still a research hotspot. The existing temporal knowledge graph reasoning methods are mainly based on embedding. These methods do not take the textual information of entities and relations into account, making the reasoning results not relevant. Besides, other reasoning methods ignore interpretable logic rules. To improve reasoning results relevance and interpretability, we propose the temporal knowledge graph reasoning model CRmod (Context-aware Rule-guided model). Through the confidence of contextual neighbors and the relevance of different neighbor 4-tuples, CRmod takes context information into account and improves the relevance of reasoning results. At the same time, considering the diversity of path rules, six temporal logic rules are proposed. Besides, time information consistency of temporal logic rules is constrained, which achieves the effect of real-time reasoning. We evaluated our model on three datasets: ICEWS18, ICEWS05-15, and GDELT. Specifically, in the experimental results concerning the Mean Reciprocal Rank (MRR) metric on the ICEWS18 and ICEWS05-15 datasets, our model showed improvements over the previous State-Of-The-Art (SOTA) models by 3.24% and 16.88%, respectively. Experiments show that CRmod does improve the relevance and interpretability of the reasoning results compared with the existing temporal knowledge graph reasoning models.
computer science, information systems