Spanish GDPR PII NER — v9
Token-classification (NER) model that detects GDPR / LOPDGDD-regulated personal data in
Spanish text, including standard identifiers and GDPR Art. 9 special-category data
(health, genetic, biometric, ethnic, political, religious, union membership, sexual
orientation, criminal records).
- Base model:
kierandesmond/spanish-gdpr-pii-ner-v8 (continued fine-tuning, NOT trained from scratch)
- Architecture:
XLMRobertaForTokenClassification (xlm-roberta-base, ~277M params)
- Labels: 57 BIO tags / 28 entity types (schema identical to v6/v8 — preserved verbatim)
- Max length: 256
What's new in v9
v9 was driven by a large, statistically-meaningful held-out evaluation set (7,304 examples,
200–336 per entity, with hard negatives) — replacing the tiny n=1–6 battery used for v6/v8 that
masked real weaknesses. Re-scoring v8 on this honest eval exposed gaps (e.g. SIP_CARD 0.5%,
ETHNIC_ORIGIN 40%, GENETIC_DATA 70%) that the old battery rated as 80–100%.
v9 adds targeted real + synthetic training data for the weak entities and hard negatives for the
false-positive traps, while preserving locked entities.
Evaluation (large held-out eval set, span-level pass-rate)
| Metric | v8 | v9 |
|---|
| Macro positive pass-rate (avg over 28 entities) | 87.3% | 99.55% |
| Macro negative pass-rate (hard negatives + false-positive guards) | 76.6% | 100% |
| Lowest single entity | 0.5% (SIP_CARD) | 94.6% (GENETIC_DATA) |
Every one of the 28 entity types scores ≥ 90%. Biggest gains vs v8:
SIP_CARD 0.5%→100%, ETHNIC_ORIGIN 40%→100%, GENETIC_DATA 70%→94.6%, CREDIT_CARD 71%→100%,
BANK_ACCOUNT 80%→100%, POLITICAL_OPINION 81%→97.6%, UNION_MEMBERSHIP 82%→100%,
BIOMETRIC_DATA 82%→100%, USERNAME 85%→100%, SEXUAL_ORIENTATION 88%→100%, PERSON_NAME 90%→100%.
(One minor trade-off: HEALTH_DATA 99.4%→97.0%, still well above bar.)
Confusion guards are clean: SEXUAL_ORIENTATION↔RELIGIOUS_BELIEF = 0 confusion;
POLITICAL_OPINION↔UNION_MEMBERSHIP = 0 confusion.
Full per-entity numbers, comparison CSV, and confusion matrix:
kierandesmond/spanish-gdpr-pii-ner-v9-validation.
Training data sources (v9)
Real corpora (parquet-native, no script execution):
- HEALTH_DATA —
BSC-NLP4BIA/bsc-bio-distemist-ner (ENFERMEDAD, train split)
- GENETIC_DATA —
PlanTL-GOB-ES/pharmaconer (PROTEINAS, train split)
- Anchor / locked entities (PERSON_NAME, ADDRESS_ES, DATE_OF_BIRTH, EMAIL, PHONE_ES, NUSS) —
bigbio/meddocan (char-offset → BIO, train split) for catastrophic-forgetting protection
Synthetic (special categories have no labeled Spanish corpus) — register-varied templates
(legal / informal / clinical), contrastive minimal pairs for confusion guards, hard negatives,
and regex-generated structured identifiers (DNI/NIE/NIF/CIF/IBAN/SIP/credit-card/etc.).
Anti-leakage: synthetic eval uses template families disjoint from training; real-corpus eval
uses validation/test splits never seen in training.
Training configuration
- Continued from v8 weights; LR 1e-5, 3 epochs, effective batch 32 (16×2), warmup 0.1,
weight_decay 0.01, bf16, max_length 256.
- Final token-level seqeval on held-out eval: precision 0.965, recall 0.976, F1 0.971.
Entity types (28)
PERSON_NAME, EMAIL, PHONE_ES, ADDRESS_ES, DATE_OF_BIRTH, IP_ADDRESS, URL, USERNAME, DNI, NIE,
NIF, CIF, NUSS, IBAN_ES, SIP_CARD, PASSPORT_ES, VEHICLE_PLATE_ES, CREDIT_CARD, BANK_ACCOUNT,
HEALTH_DATA, GENETIC_DATA, BIOMETRIC_DATA, ETHNIC_ORIGIN, POLITICAL_OPINION, RELIGIOUS_BELIEF,
UNION_MEMBERSHIP, SEXUAL_ORIENTATION, CRIMINAL_RECORD.
Usage
1from transformers import pipeline
2ner = pipeline("token-classification", model="kierandesmond/spanish-gdpr-pii-ner-v9",
3 aggregation_strategy="simple")
4ner("El trabajador está afiliado a CCOO y fue diagnosticado de diabetes mellitus tipo 2 .")
Limitations & honest caveats
- Special-category recall partly depends on phrasing patterns. Although v9 uses
register-varied, template-family-disjoint synthetic data (so eval measures generalization,
not memorization), production free-text will be more varied than any synthetic set. Treat
special-category detection as assistive (flag-for-review), not fully autonomous redaction,
until validated on real production documents.
- No real labeled corpus exists for Art. 9 special categories in Spanish — those entities are
synthetic-trained, grounded in researched real-world phrasings.
- The eval set, while large (200–336/label), is built from public corpora + synthetic generation;
a production-grade acceptance test would add human-labeled documents from the target domain.
- This model is a privacy/compliance aid and does not guarantee complete PII removal.
