New RAG method Eraser4RAG removes private data, outperforms GPT-4o

Researchers have developed Eraser4RAG, a novel method to remove sensitive information from documents used in Retrieval-Augmented Generation (RAG) systems. This approach constructs a knowledge graph to identify and separate private from public information, then fine-tunes a model to rewrite documents, excluding private triples while preserving public knowledge. Experiments show Eraser4RAG outperforms GPT-4o in effectively erasing private data while maintaining the utility of public information for generative tasks. AI

IMPACT Enhances privacy in RAG systems by enabling custom erasure of sensitive data without compromising generative capabilities.

RANK_REASON The cluster contains a research paper detailing a new method for privacy in RAG systems. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• Eraser4RAG
• FLAN-T5
• GPT-4o
• PPO algorithm
• retrieval-augmented generation
• Yujing Wang

• paper
• safety
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →