Socrates-RAG：面向协同证据投毒的前提导向主动检索

AI 摘要

该论文提出 Socrates-RAG，一种面向检索增强生成（RAG）的主动检索策略，用于应对协同证据投毒。它把竞争性答案表示为假设，选择能区分这些假设的未决前提，并利用新获取的证据细化后续查询，再决定回答或弃答。在 48 个反事实评估场景中，该方法将安全准确率从 79.2% 提升至 93.8%，提升集中在需要两跳推理的案例上。

正文节选

Socrates-RAG: Premise-Directed Inquiry against Coordinated Evidence Poisoning Abstract Retrieval-augmented generation (RAG) defenses typically decide how to filter or aggregate a fixed retrieved set. In open-corpus question answering, however, decisive evidence may be absent from the initial context but retrievable, making the next query part of the reliability problem. We introduce Socrates-RAG, a premise-directed active retrieval policy that represents competing answers, selects an unresolved

发布时间：2026-09-30 12:00

抓取时间：2026-09-30 12:16

来源机构：arXiv