ReCoPII Framework for Contextual PII Masking via Entity Relations
October 28, 2026
ReCoPII builds contextual PII masking datasets by planning ontology-constrained relations and using LLM-based verification. The resulting ReCoPII-Ko corpus achieved 88.42% annotator agreement on KEEP labels, significantly outperforming existing SPY-style augmentation.
HOW THIS AFFECTS YOU
●
builderYou can use this framework to generate more accurate training data for PII redaction systems.
●
policyThis provides a more reliable method for generating privacy-preserving datasets without exposing sensitive real-world data.