Persian PII-Masking Token Classification Corpus, 690K Initial Clean

A cleaned corpus of Persian text data annotated for Personally Identifiable Information (PII) masking, intended for token-classification tasks. The dataset was created by Reza2kn and underwent an initial cleaning process that removed near-duplicate entries. It was last updated on May 31, 2026.

Use Cases

• Train PII detection models based on token-classification annotations.
• Benchmark Persian text anonymization algorithms using the cleaned corpus.
• Fine-tune language models for privacy-preserving NLP tasks in Persian.

Strengths

• The cleaning process removed 462 rows identified as near-duplicates via cosine similarity >= 0.95.
• An audit summary is provided, detailing the raw and kept row counts (225,418 and 224,956 respectively).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown from the provided input, which may limit suitability assessment.
• Freshness should be verified as the last update timestamp is in the future (2026-05-31).

Provenance

Source

huggingface

Collection Method

Cleaned initial-round artifact from a PII-masking corpus, with near-duplicate rows removed.

Freshness

Last updated 2026-05-31 04:48:24

Geography

Persian/Iranian language focus