Oriel — AI & technology worth noticing, every day.

nvidia/Nemotron-PII-Japan

Dataset · https://huggingface.co/datasets/nvidia/Nemotron-PII-Japan

Nemotron-PII-Japan Dataset Card 日本語PII検出・編集のための合成データ / Synthetic Japanese Data for PII Detection and Redaction データセット概要 (Dataset Overview) Nemotron-PII-Japanは、日本語テキスト内の個人識別情報(PII)の検出、抽出、編集、評価を目的とした、ペルソナに基づく合成データセットです。日本の氏名、住所、識別子、金融慣習、組織、業務文書などの文脈を扱い、実在する機密性の高い文書に依存しないモデル開発を支援します。 Nemotron-PII-Japan is a synthetic, persona-grounded dataset for training and evaluating the detection, extraction, and redaction of personally identifiable information (PII) in Japanese… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-PII-Japan.

First seen by Oriel on .

AI summary

Nemotron-PII-Japan is a synthetic Japanese dataset for training and evaluating PII detection, extraction, redaction, and safety systems across realistic document contexts.

  • The dataset contains 100,000 synthetic Japanese documents spanning 43 document types and eight domains, including healthcare, finance, government, employment, education, legal, telecommunications and IT, and e-commerce.
  • It provides 56 PII labels and 810,703 character-level spans, with records containing document metadata, text, serialized spans, and inline [value]label annotations.
  • Generation combines Japanese persona sampling, deterministic format samplers, synthetic catalogs, rule-based processing, and LLM-generated text followed by validation and normalization.
  • The dataset is released as a 75.52 MB data.parquet file containing approximately 34.7 million tokens, under CC BY 4.0 for commercial and non-commercial use.
  • Annotations are not exhaustive and may contain missing labels, incorrect boundaries, or misclassifications; synthetic-data performance does not guarantee reliable real-world detection, compliance, fairness, or leakage prevention.

#pii detection #synthetic data #japanese nlp #data redaction

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Likes

Total likes reported by the source.

  • Hugging Face Datasets: 11, latest observation ; +2 over observed period since

Downloads (30d)

Rolling 30-day downloads. A change in this value is not new downloads during the observed period.

  • Hugging Face Datasets: 378, latest observation ; +166 over observed period since

Aliases

SourceExternal IDURL
Hugging Face Datasetsnvidia/Nemotron-PII-Japanhttps://huggingface.co/datasets/nvidia/Nemotron-PII-Japan Read discussion