Zero-Shot Privacy-Aware Text Rewriting via Iterative Tree Search
This work addresses privacy concerns for users of cloud-based LLM services, offering an incremental improvement over existing text anonymization techniques.
The paper tackles the problem of privacy preservation in text when using large language models, proposing a zero-shot tree-search-based iterative rewriting algorithm that systematically obfuscates or deletes private information while maintaining text quality. The method significantly outperforms existing baselines in balancing privacy protection and utility preservation.
The increasing adoption of large language models (LLMs) in cloud-based services has raised significant privacy concerns, as user inputs may inadvertently expose sensitive information. Existing text anonymization and de-identification techniques, such as rule-based redaction and scrubbing, often struggle to balance privacy preservation with text naturalness and utility. In this work, we propose a zero-shot, tree-search-based iterative sentence rewriting algorithm that systematically obfuscates or deletes private information while preserving coherence, relevance, and naturalness. Our method incrementally rewrites privacy-sensitive segments through a structured search guided by a reward model, enabling dynamic exploration of the rewriting space. Experiments on privacy-sensitive datasets show that our approach significantly outperforms existing baselines, achieving a superior balance between privacy protection and utility preservation.