GraphRAG　ニ　ヨル　ニホンゴ　ブンガク　テキスト　ノ　チシキ　グラフ　ブンセキ　エドガワランポ　D ザカ　ノ　サツジン　ジケン　ヲ　ジレイ　トシテ

Knowledge Graph Analysis of Japanese Literary Texts Using GraphRAG : A Case Study of Edogawa Ranpo’s “The Murder on D-Slope”

コウ, シンブン

Division of Language and Culture, Graduate School of Humanities, The University of Osaka

ネットワーク可視化

物語構造

Network Visualization

Narrative Structure

This study applies GraphRAG, a retrieval-augmented generation method proposed by Microsoft in 2024, to Japanese literary texts and examines its effectiveness in supporting textual interpretation through response generation and the visualization of knowledge structures. While conventional RAG relies on referencing unstructured external documents for generation, it faces limitations in capturing complex contexts such as interpersonal relationships and causal structures. In contrast, GraphRAG integrates knowledge graphs to structurally visualize the relationships between entities and generate contextually coherent responses. In this research, Edogawa Ranpo’s short story The Case of the Murder on D. Hill is used as a case study. Two types of queries̶one targeting thematic understanding and the other aiming at identifying the perpetrator̶are analyzed using GraphRAG-generated responses. Furthermore, the resulting knowledge graphs are visualized using Gephi and Neo4j to explore character relationships, structural centrality, and causal connections through network-based representations. The results indicate that GraphRAG successfully generates semantically coherent responses in Japanese literary contexts and demonstrates high utility in supporting narrative interpretation through structural visualization.