Enhancing a Tool for Extracting Knowledge Graphs from Text through the Utilization of Generative Language Models

Abstract

In this paper, we extend a tool for extracting knowledge graph candidates by leveraging dependency structure analysis. Specifically, instead of using a dependency structure analysis tool, we replace the syntactic relationship determination based on the derivation rules of one of the wellknown parsing algorithms, the CKY algorithm, with a prompt-based determination using a Large Language Model (LLM). This approach allows a better integration of knowledge graph extraction with syntactic structure analysis. The prompts focus on bunsetsus, the basic units of meaning in Japanese, allowing us to effectively address domain-specific writing styles and handle named entities by restricting sentence sets.