Reproduce “Protest Activity Detection and Perceived Violence Estimation from Social Media Images” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Donghyeon Won, Zachary C. Steinert-Threlkeld, Jungseock Joo
This research introduces an automated visual model for characterizing protest activities from social media images, specifically estimating perceived violence and associated sentiments. Utilizing a newly created UCLA Protest Image Dataset with over 40,000 images, the deep learning model achieves a 90% Pearson's correlation for perceived violence estimation and high accuracy for visual attribute classification, enabling a richer understanding of protest dynamics beyond textual analysis.
Key insights
- Visual analysis of social media images provides unique and objective insights into protest characteristics, such as perceived violence levels and participant attributes, which complement textual data.
- Specific visual attributes, like the presence of 'Fire' and 'Law enforcement,' are strongly correlated with higher perceived violence, while 'Signs' indicate less violent or more organized demonstrations.
- Deep learning models, particularly multi-task CNNs, can effectively quantify subjective human perceptions like 'perceived violence' and 'sentiment' from complex visual data with high reliability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.