A Customized Text Sanitization Mechanism with Differential Privacy

TL;DR

This paper introduces CusText, a novel text sanitization method based on epsilon-differential privacy that is compatible with any similarity measure and offers improved privacy-utility trade-offs in NLP tasks.

Contribution

CusText is a new text sanitization mechanism that overcomes limitations of existing methods by supporting any similarity measure and providing token-level privacy customization.

Findings

CusText outperforms existing mechanisms in privacy-utility trade-offs.

It is compatible with any similarity measure.

Extensive experiments validate its effectiveness.

Abstract

As privacy issues are receiving increasing attention within the Natural Language Processing (NLP) community, numerous methods have been proposed to sanitize texts subject to differential privacy. However, the state-of-the-art text sanitization mechanisms based on metric local differential privacy (MLDP) do not apply to non-metric semantic similarity measures and cannot achieve good trade-offs between privacy and utility. To address the above limitations, we propose a novel Customized Text (CusText) sanitization mechanism based on the original $\epsilon$-differential privacy (DP) definition, which is compatible with any similarity measure. Furthermore, CusText assigns each input token a customized output set of tokens to provide more advanced privacy protection at the token level. Extensive experiments on several benchmark datasets show that CusText achieves a better trade-off between privacy and utility than existing mechanisms. The code is available at https://github.com/sai4july/CusText.