Extracts insights from 26K+ protest events using BERTopic, Top2Vec, and LLMs for real-world applications like crisis monitoring, policy research, and social unrest analysis.

This project tackles the challenge of protest detection from images, identifying whether an image represents a protest or not. Inspired by the paper ["Protest activity detection and perceived violence estimation from social media images"], I've extended it using: EfficientNet architecture, Data Augmentation and Comparative Model Analysis.