Distilling Human-Aligned Privacy Sensitivity Assessment from Large Language Models
This work addresses the problem of scalable and efficient privacy evaluation for natural language processing practitioners, though it is incremental as it builds on existing LLM-based methods.
The paper tackled the challenge of computationally expensive privacy evaluation of textual data by distilling the privacy assessment capabilities of a large language model into lightweight encoder models, achieving strong agreement with human annotations while reducing model size to as few as 150M parameters.
Accurate privacy evaluation of textual data remains a critical challenge in privacy-preserving natural language processing. Recent work has shown that large language models (LLMs) can serve as reliable privacy evaluators, achieving strong agreement with human judgments; however, their computational cost and impracticality for processing sensitive data at scale limit real-world deployment. We address this gap by distilling the privacy assessment capabilities of Mistral Large 3 (675B) into lightweight encoder models with as few as 150M parameters. Leveraging a large-scale dataset of privacy-annotated texts spanning 10 diverse domains, we train efficient classifiers that preserve strong agreement with human annotations while dramatically reducing computational requirements. We validate our approach on human-annotated test data and demonstrate its practical utility as an evaluation metric for de-identification systems.