Mitigating privacy risks in Retrieval-Augmented Generation via locally private entity perturbation

Longzhu He, Peng Tang, Yuanhe Zhang, Pengpeng Zhou, Sen Su

Abstract: Highlights•A novel privacy-preserving Retrieval-Augmented Generation (RAG) pipeline: LPRAG.•Entity-specific perturbation ensures targeted privacy preservation for text.•Differentially private perturbation of words, numbers, and phrases.•Extensive experiments are conducted to verify the superiority of our method.