An Exclusion-Based Interface Making Social Sensing Reliable in Disasters

This research addresses the critical challenge of efficiently collecting damage information from social media immediately after a disaster happens. Social media serves as an essential ``social sensor,'' but its data stream is heavily contaminated with noise such as mis/disinformation and extraneous content, severely hindering rapid situational awareness. Given the indispensable need for human verification to ensure information reliability, this paper proposes an interface based on a novel ``subtractive strategy.'' This strategy systematically eliminates unnecessary information, preserving a concentrated, high-relevance information set for human judgment. The core feature is a ``stepwise exclusion'' mechanism that allows users to iteratively filter out irrelevant data clusters, thereby dramatically improving the visibility of critical damage posts. To verify the system's efficacy, we conducted a user study using a dataset of tweets from the July 2020 heavy rainfall event in Japan. The results demonstrated that while a simple baseline system (text and image presentation) had a slight initial lead in discovered posts, the proposed system's overall performance surpassed the baseline after approximately 20 minutes period. These findings collectively indicate that our exclusion-based interface substantially improves the efficiency of extracting critical disaster intelligence from social networking services.