ReCoPII: A Relation-Grounded Framework for Building Contextual PII Masking Datasets
Paper · https://arxiv.org/abs/2610.35925
arXiv:2610.35925v1 Announce Type: new Abstract: Training contextual masking systems for personally identifiable information (PII) requires examples distinguishing maskable from preservable mentions of one entity type, but privacy restrictions limit access to real data. Author-based augmentation (SPY) labels mentions unrelated to the author as non-PII, presuming a single data subject and overlooking others. We propose ReCoPII (Relation-grounded Contextual PII masking), a framework that builds such data without this premise: it plans ontology-constrained entity relations, writes insertion sentences with placeholders, reconstructs annotations with code, checks the result with a separate LLM, and derives MASK/KEEP labels from entity relations. On Korean documents from six domains, it yields ReCoPII-Ko, a corpus with relation, privacy, and MASK/KEEP annotations, which we validate. Across 300 pairs, four annotators agree with 88.42% of its sampled KEEP labels, against 26.27% for SPY-style augmentation of the same sources. Classifiers trained on it over-mask less than under that reference on a matched test. Contextual masking fine-tuned jointly with a PII detector removes most over-masking of detection alone within a fixed leakage budget; on human-verified entity records, the four detectors meeting it over-mask at most 5.04% with at most 4.18% leakage. The detector is comparable to XLM-R-large on KDPII with about a quarter of its parameters.
First seen by Oriel on .
AI summary
ReCoPII generates relation-grounded contextual PII masking data without assuming a single data subject, and its Korean corpus improves KEEP-label agreement over SPY-style augmentation.
- The pipeline plans ontology-constrained entity relations, generates placeholder-based insertion sentences, reconstructs annotations in code, and checks outputs with a separate LLM.
- ReCoPII-Ko covers Korean documents from six domains and includes relation, privacy, and MASK/KEEP annotations.
- Across 300 pairs, four annotators agreed with 88.42% of sampled KEEP labels, compared with 26.27% for SPY-style augmentation on the same sources.
- Jointly fine-tuning contextual masking with a PII detector reduces over-masking within a fixed leakage budget; four qualifying detectors over-masked at most 5.04% with leakage at most 4.18%.
- The detector is comparable to XLM-R-large on KDPII with about one quarter of its parameters.
#pii detection #privacy #dataset generation #korean nlp
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| arXiv AI Papers | oai:arXiv.org:2610.35925v1 | https://arxiv.org/abs/2610.35925 Read discussion |