@techreport{oai:ipsj.ixsq.nii.ac.jp:00241593, author = {村上, 一彦 and 瀬川, 修 and 木村, 佳央 and Kazuhiko, Murakami and Osamu, Segawa and Yoshio, Kimura}, issue = {18}, month = {Dec}, note = {我々は電力会社のヒューマンエラー事象を記録した「ヒヤリハット」の事例を知識源として,ヒューマンエラーに関する対策を立案する手法の検討を行っている.本稿では,大規模言語モデル(Large Language Model: LLM)と RAG(Retrieval Augmented Generation)の枠組みを用い,不完全な外部知識を LLM の内部知識により補完する拡張方式を提案する.ヒヤリハットの数十万件の大規模事例を用いた対策立案(マルチドキュメント要約)の評価実験では,提案する知識補完の有効性が示唆される結果が得られた., We have been studying a method for planning countermeasures against human error using “near-miss” cases that record human error events in electric power companies as a knowledge source. In this paper, we propose an enhanced approach that supplemets incomplete external knowledge with the internal knowledge of Large Language Model (LLM), using the framework of LLM and Retrieval Augmented Generation (RAG). Evaluation experiments on countermeasure planning (multi-document summarization) using hundreds of thousands of large-scale near-miss cases suggests the effectiveness of the proposed knowledge supplementation method.}, title = {RAGを用いたヒューマンエラー事象の対策立案}, year = {2024} }