Deep Learning-Based Detection of Vulnerable Victims on Social Media for Post-Disaster Impact Assessment
Shenbagalakshmi Gunasekaran, S Abhishta, E. J. G. Merin, Nanditha Karthikeyan, Neethu Das H · 2025
Social media's development has opened fresh possibilities for the gathering and sharing of information after natural disasters. This has produced necessary new perspectives on the changing circumstances on the ground. Analyzing large user-generated content like text, images, and videos helps tremendously improve disaster response by offering immediate situational awareness and efficient resource distribution. Using transformer-based models including BERT and Convolutional Neural Networks (CNNs), the accuracy of vulnerability identification is improved. The strategy is to provide priority relief to those most in need thereby assisting emergency responders. This study also addresses cyberbullying, a rising problem on social media platforms affecting mental health and internet security. Leveraging BERT's sophisticated language processing capabilities to precisely classify text and address issues such complex language, sarcasm, and covert bullying, in this research, a deep learning approach essentially identifies cyberbullying on social media, particularly in post-disaster environments. Constant performance of the model is validated using parameters like accuracy, precision, recall, and F1-score. The automatic identification of harmful information by this system shows how artificial intelligence may improve online safety and resilience, therefore enabling the development of more safe digital environments and future advancement in civic content modification.