Social Media Scanning for Crisis Management

Abstract

Hurricane Maria exposed critical gaps in Puerto Rico's emergency response infrastructure. This research presents the Social Media Emergency Scanner, a system using NLP and deep learning to automatically detect and classify emergency-related posts from social media. Using the X (Twitter) v2 API, 4,250 bilingual (Spanish/English) posts were collected across multiple disaster years. Over 52% of posts contained actionable emergency information, with post utility improving significantly over time. Dominant emergency types included earthquakes and traffic accidents. A TF-IDF/Logistic Regression classifier was deployed, with BERT-Att-BiLSTM identified as the optimal future model. This tool aims to streamline disaster response coordination in Puerto Rico.