Disaster Warning Map using Premonitory Symptom Keywords Gathered from Twitter in Real Time

摘要

In this research project, the premonitory symptom keywords and location information relating to landslides, tsunamis, river flooding and sudden downpours of rain were gathered from Twitter (SNS Data) in real time. The data was analyzed, a hazard coefficient was calculated, and this coefficient was then visualized on a heat map so that a disaster warning map could be developed. In 2016, at the time of the occurrence of Typhoon No. 16 in Japan, about 4000 Tweets were acquired and an experiment was carried out for the evaluation of river flooding. Specifically, based on the results verified by the summarized damage reports from Typhoon 16 announced by the Japanese Cabinet Office (report) and on the relevant news reports, it was clearly possible to predict these events more quickly by using the disaster warning map than by relying on the news reporting services for about 57% of the actual occurrences (21 instances) where the water surfaces of rivers went over the danger level.