NERaseText: Sensitive-Aware Text Sanitization under Differential Privacy

The widespread adoption of large language models (LLMs) has increased the need for privacy-preserving text processing techniques. Existing differentially private text sanitization methods apply uniform privacy parameters across all vocabulary elements, failing to recognize that certain words carry different levels of sensitive information. We propose NERaseText, a Named Entity Recognition-assisted differentially private text sanitization framework that dynamically allocates privacy budgets based on the sensitivity level of individual words. Our approach achieves comparable results to a state-of-the-art framework, while also providing tighter privacy guarantees to sensitive words and utilizing a lower privacy budget.