Dark Web Topic Analysis (GSoC @ HumanAI)

Goal

This project, developed for Google Summer of Code at HumanAI, focuses on uncovering how language tied to criminal behavior evolves over time within Dark Web forums, providing insights for security and linguistic research.

Technical Approach

• Analyzed over 500,000 discussion points from underground marketplaces and illicit forums.
• Used BERT and BERTopic for semantic embedding and clustering to identify emerging criminal themes.
• Integrated CLIP and ViT to examine visual content alongside textual data, creating a multi-modal analysis pipeline.
• Conducted temporal analysis to track linguistic shifts and the evolution of topics across months and years.

Impact

• Identified Key Themes: Extracted 5 core topic categories, creating a temporal view of how narratives and terminology changed.
• Predictive Modeling: Deployed predictive models on Hugging Face to forecast thematic evolution.
• Open Science: The results, models, and code were communicated through an in-depth Medium article and fully open-sourced on GitHub.