Examination of noise sweeping methods for extracting rescue request tweets with high reliability

Abstract

This research aims to provide a platform for rapid and effective information sharing on X (Twitter) during disasters. It seeks to efficiently extract vital information from the overwhelming volume of data available. A significant challenge in utilizing social media for disaster response is ensuring the reliability of the information and preventing important information from being overlooked. Ensuring reliability requires human judgment, raising concerns that pinpoint information extraction with conventional text-mining based approach might miss critical information. To address this, our study proposes a method to minimize human verification costs by sweeping out irrelevant (noise) information while retaining essential data. This approach fosters cooperative information collection between humans and computers. In this paper sorted out information that should actually be retained from the perspective of both posted images and posted text, and information that is noise. Our proposal approaches succeeded to narrow down the posts that highly required for human judgement is only 0.564% of posts we have gathered.