User-Entity Differential Privacy in Learning Natural Language Models

Phung Lai, Tong Sun, Rajiv Jain, Franck Dernoncourt, Jiuxiang Gu, Nikolaos Barmpalios, Han Hu, Hai Phan

Keywords: differential privacy, natural language models

Abstract: In this paper, we introduce a novel concept of user-entity differential privacy (UeDP) to provide formal privacy protection simultaneously to both sensitive entities in textual data and data owners in learning natural language models. To preserve UeDP, we developed a novel algorithm, called UeDP-Alg, optimizing the trade-off between privacy loss and model utility with a tight sensitivity bound derived from seamlessly combining sensitive and non-sensitive textual data together. An extensive theoretical analysis and evaluation show that our UeDP-Alg outperforms baseline approaches in terms of model utility under the same privacy budget consumption on several NLM tasks, using benchmark datasets.

Community Implementations: [ 3 code implementations](https://www.catalyzex.com/paper/arxiv:2211.01141/code)