Real-Time Disaster Forecasting: Harnessing Social Media and NLP for Early Crisis Detection

Authors- K. Srikanth, Yarra Lalitha Dabi Varshini, Rayudu Satish, Yarra Koushik, Kondisetti Phaneendra Saketh, Pulidindi Harsha Vardhan

Abstract-Natural disasters such as wildfires, earthquakes, floods, cyclones, and heatwaves have significantly impacted social media users, who actively express their sentiments about these crises online. Analysing location-specific public emotions during such events is essential for policymakers and emergency response teams to make informed decisions. To address this need, we introduce a fully automated artificial intelligence (AI) and natural language processing (NLP) framework designed to extract and analyse sentiment trends related to disasters at specific locations. Our proposed system utilizes AI-driven sentiment analysis, named entity recognition (NER), anomaly detection, regression modelling, and the Getis-Ord Gi* algorithm to process multilingual social media content. Supporting 110 languages, our approach facilitates comprehensive sentiment monitoring and disaster intelligence extraction from social media data. The system was implemented and tested on live Twitter data collected from September 28 to October 6, 2021, processing 67,515 entities in 39 languages and identifying 9,727 location-based entities with over 70% confidence. These insights enabled real-time mapping of disaster-affected locations along with sentiment-based disaster intelligence. Performance evaluation demonstrated the system’s effectiveness, achieving an average precision of 0.93, recall of 0.88, and an F1-score of 0.90, with an overall accuracy of 97%, highlighting the reliability of our AI-driven disaster monitoring framework.

DOI: /10.61463/ijset.vol.13.issue2.226