Analysis of Conversation Data With AI Chatbot at the Time of Natural Disaster


In recent years, many natural disasters such as typhoons and earthquakes have occurred in Japan. Thus far, disaster information from local governments and television has been the main information source for such disasters. In addition to such information sources, the provision of disaster and evacuation information through dialogue with AI chatbots has recently begun as a new information providing medium. This study analyzed dialogue data of victims collected from AI chatbots. The dialogue data used in the analysis is a dataset that collected 160,196 dialogues from September 23, 2019 to October 31, 2019, affected by Typhoons 15 and 19. According to the analysis results, those with the highest number of dialogues with AI chatbots were in the order of the method of obtaining the victim certification, the procedure of financial support, house repair, disaster prevention information, and lifeline information. Also, when examining the contents of the dialogue with the AI chatbot in chronological order, information acquisitions are concentrated during the week immediately after opening the AI chatbot account, and especially information about the victim certification and home repair were concentrated immediately after the disaster. Based on these results, it is important to provide information on disasters step by step at the disaster prediction stage, disaster occurrence stage, disaster damage processing stage, and subsequent continuous support stage.

Author Information
Nagayuki Saito, International Professional University of Technology, Japan
Nao Fukushima, Council on AI for Disaster Resilience, Japan
Ryusuke Yonekura, LINE Corporation, Japan
Kazuto Ikeda, LINE Corporation, Japan
Kiyotaka Eguchi, Council on AI for Disaster Resilience, Japan

Paper Information
Conference: ACSS2020
Stream: Other

The full paper is not available for this title

Virtual Presentation

Comments & Feedback

Place a comment using your LinkedIn profile


Share on activity feed

Powered by WP LinkPress

Share this Research

Posted by James Alexander Gordon