Automatic Extraction of Protected Health Information from Multilingual Hacker Communities

This research tackles the theft and resale of stolen health records discussed in hacker channels on encrypted platforms like Telegram and Discord. The authors build NERF-PHI, a named-entity-recognition system that, after machine-translating over three million multilingual hacker posts into English, automatically pulls out mentions of victims and medical data. They find that encoder-based language models are highly effective at surfacing this protected health information for investigators.