Scholarlink

In disaster management, the process of measuring resilience is essential for gaining insights into community perspectives and experiences, thus enabling effective planning and response strategies. A recommended approach involves conducting resilience measurements at regular intervals, such as every six months. This interval allows practitioners to obtain a comprehensive snapshot of the community’s current state and identify areas for enhancing community resilience. However, the current practice of conducting manual open-ended interviews is arduous and time-consuming. Addressing this issue can be facilitated by integrating Generative AI and NLP technologies, which offer promise in creating a Decision Support System. This study investigates the key components of open-ended interviews necessary for reliable data collection and explores the potential for automation. The identified components and methods of automations included the answer transcriber component matched with speech-to-text model of SpeechStew, logical unit component harmonised with Information Retrieval technique of T5 doc2query, and follow-up question generator component coupled with the Large Language Model of Anthropic’s Claude 3 Sonnet. Automation up to 75% of components was achieved which despite the advantages, the study also acknowledges the inherent limitation of human intervention in conducting interviews. Therefore, a decision support system is proposed to empower practitioners with the capability to intermittently collect disaster resilience data, thereby facilitating more adaptive disaster planning. Quality metrics, derived from two case studies, were meticulously applied to the designed decision support system to ensure that the automated system delivers reliable outcomes.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automating Data Collection for Disaster Resilience Open-Ended Interviews Using AI and NLP

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Milad Katebi,
  • Mani Poshdar,
  • Mostafa Babaeian Jelodar,
  • Flavio Soares Correa da Silva

摘要

In disaster management, the process of measuring resilience is essential for gaining insights into community perspectives and experiences, thus enabling effective planning and response strategies. A recommended approach involves conducting resilience measurements at regular intervals, such as every six months. This interval allows practitioners to obtain a comprehensive snapshot of the community’s current state and identify areas for enhancing community resilience. However, the current practice of conducting manual open-ended interviews is arduous and time-consuming. Addressing this issue can be facilitated by integrating Generative AI and NLP technologies, which offer promise in creating a Decision Support System. This study investigates the key components of open-ended interviews necessary for reliable data collection and explores the potential for automation. The identified components and methods of automations included the answer transcriber component matched with speech-to-text model of SpeechStew, logical unit component harmonised with Information Retrieval technique of T5 doc2query, and follow-up question generator component coupled with the Large Language Model of Anthropic’s Claude 3 Sonnet. Automation up to 75% of components was achieved which despite the advantages, the study also acknowledges the inherent limitation of human intervention in conducting interviews. Therefore, a decision support system is proposed to empower practitioners with the capability to intermittently collect disaster resilience data, thereby facilitating more adaptive disaster planning. Quality metrics, derived from two case studies, were meticulously applied to the designed decision support system to ensure that the automated system delivers reliable outcomes.