Toward the automatic detection of rescue-request tweets: analyzing the features of data verified by the press
Chenjie Song, 裕之 藤代 · 2019
Twitter has successfully been used as a critical information channel during natural disasters since the 2011 Great East Japan Earthquake. However, enormous numbers of tweets are increasingly being attached to the hashtag “#rescue,” which is used to send rescue requests. This makes it difficult to determine whether a tweet is in fact a rescue request by human effort, thus requiring automated detection. However, it is not easy to confirm whether a tweet is a valid rescue request through such means. This study therefore conducted an experiment to effectively detect rescue-request tweets by analyzing specific tweet features based on a dataset provided by the Japan Broadcasting Corporation (NHK), which verified that all people associated with the dataset were rescued during the 2018 heavy-rain disaster in western Japan. We also compared these tweet features to those from another set of tweets collected during the same disaster to verify their effects. We thus succeeded in manually identifying rescue-request tweets from those related to other information with a relatively high accuracy rate of 64.7%.