European Financial Text Data With Redacted Personal Information

50 sample entries from the PII-Masking-2M European release provide a preview of financial documents with personally identifiable information redacted. The dataset is created by AI4Privacy and was last updated in April 2026. The full dataset contains 200,000 entries.

Use Cases

• Train models to identify and classify PII entities like names and account numbers using the redacted source text and PII labels.
• Benchmark the performance of different NER architectures on financial domain text using the provided label distribution statistics.
• Analyze language distribution across European financial documents for multilingual NLP model training.
• Study geographic coverage patterns within European financial data using the European coverage metadata.
• Generate synthetic financial text data by learning patterns from the redacted source text for privacy-safe model training.

Strengths

• Full dataset contains 200,000 entries, providing a substantial corpus.
• Sample of 50 entries allows for initial inspection of data structure and quality.
• Explicitly documents label and language distributions for transparency.

Limitations

• Preview size of 50 rows is too small for any meaningful model training.
• Source text and PII values are redacted in the preview, limiting its utility for analysis.
• Geographic coverage is specified as European, which may limit global applicability.

Provenance

Source

AI4Privacy, part of the PII-Masking-2M project.

Collection Method

Presumably gathered and annotated from European financial documents, with PII redacted.

Freshness

Last updated in April 2026, indicating it is a forward-dated or frequently maintained resource.

Geography

European coverage.