Presentation Information
[B-6-53]Knowledge Graph–Based Preprocessing to Improve RAG Accuracy for Technical Document
〇Takayuki Fujiwara1, Hiroki Inoue1 (1. NTT, Inc.)
Keywords:
Large Language Model,Retrieval-Augmented Generation,Knowledge Graph
With the increasing sophistication and scale of network-related systems, technical documents, including specifications, have become enormous. Thoroughly understanding these documents is a significant burden, especially for newly appointed engineers, and there is a high demand for interactive guidance on specific areas of interest. Retrieval-Augmented Generation (RAG) is a technology that combines document retrieval with large-scale language models (LLMs) to enable question-and-answer sessions based on document content, and is expected to be applied to support the understanding of internal company documents. This study focuses on internal network-related technical documents and aims to realize an inquiry system infrastructure for users who want to understand the document content, by examining pre-processing methods to improve RAG accuracy.