Social Media Text Analysis for Disaster Management Using DistilBERT Model

There is no doubt on the fact that disasters like floods, earthquakes and other man-made ones have a destructive impact on human lives. Timely response is crucial to minimize the casualties and aid recovery. Social media in such times, are vital as people share experiences, news and request assistance. This study investigates the efficiency of Transformer-based models for fine-grained sentiment analysis of disaster tweets on social media. Five models, including Distil-BERT, are compared and analyzed using Twitter data. The analysis categorises tweets into ten classes reflecting human sentiments during a disaster. Model performance is evaluated using accuracy, precision, recall, F1-score and execution time. While all the models demonstrate strong performance, Distil-BERT achieved comparable accuracy with a significant reduction in training and execution time (45%-55%). These findings suggest the potential of Distil-BERT (Distilled BERT) as the core component in disaster management applications; enabling organizations to gain a deeper understanding of the multifaceted public sentiment during disasters and make informed decisions for improved public aid and resource allocation.