Werea-KVKK-PII-150M v2
Research-preview Turkish PII recogniser for the Werea KVKK PrivacyOps project. It is trained for these synthetic span categories: ADDRESS, BIOMETRIC_DATA, CRIMINAL_CONVICTION_SECURITY, CREDIT_CARD, DATE_OF_BIRTH, DEVICE_ID, EMAIL, FINANCIAL_DATA, GENETIC_DATA, HEALTH_DATA, IBAN_TR, IPV4, PASSPORT_TR, PERSON, PHONE_TR, POLITICAL_OPINION, RELIGION, SEXUAL_LIFE, TCKN, UNION_MEMBERSHIP, VEHICLE_PLATE_TR. The public training set is deterministic and entirely synthetic; it contains no customer records. Use the deterministic rules layer for checksummed identifiers and evaluate every category on an institution-specific, lawyer-reviewed corpus before production deployment.
Validation F1 on the synthetic split: 1.0000. This score must not be interpreted as real-world compliance performance.
The model cannot determine legal compliance or choose a legal basis. Human review is mandatory. See the dataset card and Werea PrivacyOps documentation.