Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·28d ago·5 min read
Meddies-PII: A Multilingual Framework for Personally Identifiable Information Extraction in Clinical De-identification
TL;DR
Meddies-PII is a multilingual framework designed for extracting personally identifiable information (PII) from clinical documents. It includes a dataset of one million synthetic documents and a model that outperforms existing systems in PII extraction tasks.
✦ Why It Matters
Start using the Meddies-PII framework to improve PII extraction in your clinical applications today.
Key Takeaways
- Utilize the Meddies-PII framework for multilingual PII extraction.
- Leverage the one million synthetic documents for training models.
- Implement the BIOES token classification approach for better accuracy.
- Access the publicly released dataset and tools for your research.
How It Works
Meddies-PII generates synthetic clinical documents using attribute-conditioned prompts, which specify desired characteristics of the text. This process is followed by validation through thirteen deterministic gates that check for structural and annotation consistency, ensuring high-quality data for training the PII extraction model.