Pragmatic De-Identification of Cross-Domain Unstructured Documents: A Utility-Preserving Approach with Relation...

Area of Science:

• Computer Science
• Information Science
• Data Privacy

Background:

• Daily generation of vast amounts of personal information necessitates robust privacy measures.
• Compliance with evolving global privacy regulations is crucial for leveraging data.
• Existing de-identification methods struggle to balance privacy with data utility.

Purpose of the Study:

• To introduce READI, a novel framework for de-identifying unstructured documents.
• To enhance data de-identification quality by improving entity detection.
• To evaluate READI's effectiveness in reducing false positives and preserving data utility.

Main Methods:

• Utilizing Named Entity Recognition (NER) and Relation Extraction (RE) technologies.
• Developing a utility-preserving framework named READI.
• Evaluating the approach on two distinct datasets against state-of-the-art methods.

Main Results:

• READI significantly reduces false positives in de-identified text.
• The Relation Extraction-based Approach for De-Identification (READI) improves the utility of de-identified data.
• Demonstrated effectiveness on multiple datasets compared to existing techniques.

Conclusions:

• READI offers a superior approach to unstructured document de-identification.
• The framework effectively balances privacy preservation with data usability.
• READI aids organizations in meeting compliance requirements for personal information.