Relational semantic-enhanced logic rule learning for knowledge graph completion

doi:10.1007/s13042-024-02434-7

Relational semantic-enhanced logic rule learning for knowledge graph completion

2024-10-30

0

PRE

AI

DOI：10.1007/s13042-024-02434-7

Original

Original request for help

Save

Abstract

En 中文

Knowledge graph completion (KGC) aims to automatically infer missing information in a knowledge graph (KG). Logical rule learning-based KGC has garnered significant attention due to its ability to provide logical reasoning and strong interpretability. In these models, the input paths are crucial for training the models to learn logical rules. However, topology-based methods like random walk path sampling often ignore the semantic information of relationships in the KG, which can lead to insufficient sampling, thereby affecting the quality of rules and the performance of KGC. To address this issue, we propose a path sampling method enhanced by relational semantics. First, building on random walk sampling, we prompt a large language model (LLM) to infer additional paths by understanding semantic logical connections between relationships in the KG. Second, to address unreasonable paths that may arise from potential hallucinations of the LLM, we propose a path filtering method based on a statistically analyzed relation set. Through this process, we obtain richer and more reasonable paths for logical rule learning, ultimately generating high-quality logical rules to improve the performance of KGC. Experimental results demonstrate that our model outperforms other baseline methods on three public datasets.

Keywords:

Logical rule learning

Knowledge graph completion

Relational semantic information

Path expansion

Relational semantic-enhanced logic rule learning for knowledge graph completion

Abstract

Journal

Organization