earticle

영어

The GraphRag technique has recently been studied meaningfully as a technique for securing high performance in search and inference without hallucination for domain-specific knowledge. GraphRAG builds a graph of the document's core concepts and the relationships between them, resulting in a connected knowledge network. To validate the practicality of GraphRAG, we implemented an EPUB reader using GraphRAG in our previous research and evaluated its retrieval performance, achieving 90% accuracy in both factual retrieval tasks and complex inferential queries, demonstrating superior performance. In this paper, we describe details of the implementation of the EPUB reader capable of retrieval and inference using the GraphRAG technique. Specifically, we detail the processes of building a graph database from EPUB files, conducting retrieval, and visualizing the constructed graphs. Furthermore, we explain how the EPUB reader was designed to run on resource-constrained devices such as the Raspberry Pi 5 single-board computer. We expected that this system serves as a representative implementation example for applications leveraging GraphRAG technology.