Dataset · RedactionBench
Benchmark for finding and redacting sensitive data
RedactionBench/RedactionBench
200 complex English documents with manual PII annotations for evaluating redaction systems.
What's inside
The dataset includes 200 diverse English texts with manually and thoroughly labeled sensitive fragments (token-classification task). Along with the data, the authors propose the R-Score metric to evaluate information concealment quality independently of surface formatting.
What you can do with it
- Evaluating and comparing models for automated text anonymization (PII redaction)
- Testing NER systems and LLMs for finding sensitive data in complex documents
- Tasks
- Token classification
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 545
- Likes
- 0
How to load it
from datasets import load_dataset
ds = load_dataset("RedactionBench/RedactionBench")