Citation

Nakata T. J. Japan Soc. Safety Eng. 2020; 59(6): 373-378.

Vernacular Title

事故報告書から安全知識を抽出する自然言語処理

Copyright

(Copyright © 2020, Japan Society for Safety Engineering)

DOI

10.18943/safety.59.6_373

PMID

unavailable

Abstract

Detailed information and lessons learned from the accident are recorded in the report in natural language, and a huge amount is accumulated in society. If these can be utilized, it will be very effective in improving safety. However, statistical processing of a large amount of natural language data was difficult, and its utilization was not advanced. In this paper, we propose a natural language processing technique that derives useful knowledge for accident prevention. To prevent accidents, it is necessary to understand the pattern of accidents. There, the computer automatically extracts the scenes in the process of the accident and grasps how the scenes transition in the process of the accident. There is also a way to search for the most similar historical serious accidents to the hiyari hats that have not been damaged, and to know the possible damages. By combining multiple reports, it is possible to point out the risk of accidents that have not occurred but are likely to occur. In this paper, we introduce examples based on report data collections in aviation and oil refining.

事故の詳しい情報や教訓は報告書に自然言語の形で記録され，膨大な量が社会に蓄積されている．これらを活用できれば安全向上に大いに効果があるだろう．だが大量の自然言語データは統計処理が難しく，活用が進んでいなかった．本論文では，事故予防に役立つ知識を導き出す自然言語処理の技法を提案する．事故予防には，事故のパターンを把握することが必要となる．そこでは，事故の過程におけるシーンを計算機が自動的に抽出し，事故の進展過程でシーンがどのように遷移するかを把握する．また，被害無しで済んだヒヤリハットについて，それに最も類似する歴史的重大事故を検索し，あり得たかもしれない被害を知るという活用法もある．複数の報告を組み合わせることで未発生だが起こりそうな事故のリスクも指摘できる．本論文では，航空や石油精製での報告書データ集を題材にした実施例を紹介する．

Language: ja

Keywords

ヒヤリハット; リスク分析; 事故報告書; 人工知能; 自然言語処理