PII Masking Text Corpus with 54 Entity Classes

Pii-masking-200k is a text dataset for training models to remove personally identifiable information. It contains example texts targeting 229 discussion contexts and is labeled with 54 distinct PII classes. The dataset was created by the Ai4Privacy community and last updated in April 2024.

Use Cases

• Train a named entity recognition model to identify 54 PII classes within conversational text.
• Fine-tune a text generation model to mask or redact sensitive entities across 229 targeted discussion contexts.
• Develop a privacy-preserving data preprocessing pipeline for AI assistant training data using the labeled PII examples.
• Benchmark the performance of different anonymization techniques on the defined set of 54 sensitive data types.

Strengths

• Contains labels for 54 distinct classes of personally identifiable information.
• Targets 229 specific discussion contexts for privacy masking applications.
• Created by a dedicated community (Ai4Privacy) focused on building open privacy datasets.

Limitations

• The exact number of text examples (rows) is not specified in the provided input.
• No sample data or column structure is available to assess data format and annotation schema.
• Potential bias towards the 229 targeted discussion contexts, which may not cover all real-world scenarios.

Provenance

Source

huggingface

Collection Method

Created by the Ai4Privacy community to build open datasets for privacy masking.

Time Range

null

Freshness

Last updated on 2024-04 21.

Geography

null