Presentation Information
[1K3-GS-3a-01]Knowledge Graph Representation Aggregation Method for Link Prediction Based on Linguistic Ontology Hierarchical Structures
〇Ayano Ide1, Koki Okajima1, Yasuaki Nakamura1 (1. NTT, Inc.)
Keywords:
Knowledge graph,Knowledge graph construction,Link prediction
Recent advances in large language models (LLMs) have made it easier to construct knowledge graphs from unstructured text.
However, knowledge graphs constructed using LLMs often exhibit variability in their representations, causing semantically similar entities or relations to be represented as distinct ones. This variability leads to degraded link prediction accuracy.
To address this issue, we propose a data-driven knowledge graph representation aggregation method that hierarchically aggregates entities and relations based on the hierarchical structure of a linguistic ontology, thereby reducing representation variability.
Experiments on public datasets demonstrate that the proposed method constructs knowledge graphs with fewer entities and relations while preserving high semantic similarity to the original graph. Furthermore, the reconstructed graph improves link prediction accuracy.
Overall, these results show that the proposed method effectively controls knowledge representation variability in LLM-based knowledge graph construction and improves link prediction accuracy.
However, knowledge graphs constructed using LLMs often exhibit variability in their representations, causing semantically similar entities or relations to be represented as distinct ones. This variability leads to degraded link prediction accuracy.
To address this issue, we propose a data-driven knowledge graph representation aggregation method that hierarchically aggregates entities and relations based on the hierarchical structure of a linguistic ontology, thereby reducing representation variability.
Experiments on public datasets demonstrate that the proposed method constructs knowledge graphs with fewer entities and relations while preserving high semantic similarity to the original graph. Furthermore, the reconstructed graph improves link prediction accuracy.
Overall, these results show that the proposed method effectively controls knowledge representation variability in LLM-based knowledge graph construction and improves link prediction accuracy.