Earticle

초록

영어

Local governments need to continuously enhance their staff’s disaster response capabilities by incorporating the latest knowledge and experience. Recently, after-action reports have been frequently published by local governments to review their disaster responses. Although learning from these reports is effective for improving disaster response capacity, it is not easy for local government officers to spend sufficient time learning from them. In this study, a method to efficiently extract lessons learned from after-action reports on local government disaster response was developed. First, cases of critical situations during emergency responses were collected from after-action reports published after recent disasters. Next, a deep learning model based on a language model (BERT) was developed using the collected cases as training data. The model enables the extraction of new cases similar to the collected ones from after-action reports. As the developed model was able to extract new cases with high accuracy, it may contribute to the efficient learning and analysis of lessons learned from after-action reports. On the other hand, the findings suggested the need to explore the optimal amount of training data required to achieve high accuracy and stable learning performance. After these efforts were started with after-action reports of flood disasters, the reports of earthquake and tsunami disasters were also added as the targets. Finally, a support system was developed to extract lessons from after-action reports using the proposed model and to present accumulated lessons efficiently. This system enables users to explore the latest lessons from recently published after-action reports and enhance their capabilities based on these lessons.

저자

• Miho Ohara [ The University of Tokyo; AI Hirama; Hokkaido University ]
• Yudai Hirama [ The University of Tokyo; AI Hirama; Hokkaido University ]
• Soichiro Yokoyama [ The University of Tokyo; AI Hirama; Hokkaido University ]
• Tomohisa Yamashita [ The University of Tokyo; AI Hirama; Hokkaido University ]

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  한국재난정보학회 학술발표대회

• 간기

  부정기

• 수록기간

  2005~2026

• 십진분류

  KDC 338 DDC 361