PIIBench: A Unified Benchmark for PII Detection Across Domains

PIIBench is a benchmark dataset for detecting Personally Identifiable Information (PII) in text. It contains approximately 1.39 million records annotated with BIO tags for 48 entity types, aggregated from ten source datasets. The dataset was created by Pritesh-2711 and was last updated on Hugging Face in April 2026.

Use Cases

• Training PII detection models based on the BIO-tagged text tokens.
• Benchmarking model performance across 48 different PII entity types.
• Studying PII distribution and detection challenges across multiple data sources mentioned in the description.

Strengths

• Large scale with approximately 1.39 million total records.
• Covers a wide variety of 48 PII entity types.
• Unified structure from ten different source datasets.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The specific ten source datasets are not listed in the provided metadata.
• License information is unknown, which may restrict usage.

Provenance

Source

Aggregated from ten datasets, as referenced in the description.

Collection Method

Likely compiled from existing public datasets for benchmarking purposes.

Freshness

Last updated 2026-04-20 17:42:28; freshness should be verified.