MIRAGE 用跨文档论证图抵御长文本 RAG 中的误导信息污染。
MIRAGE: Defending Long-Form RAG Against Misinformation Pollution
- 构建 NLI 驱动的跨文档主张图，识别矛盾与一致信息
- 在混合与全污染数据下，将事实性提升至接近纯净数据水平
- 无需训练、适配多模型，适合部署于真实场景的 RAG 系统
检索增强生成（RAG）通过引入外部证据提升大模型的事实性，但现实检索常被污染：语义相关段落可能包含细微错误、误导性表述或虚构内容。我们提出 MIRAGE，一种无需训练、模型无关的长文本 RAG 防护机制。MIRAGE 构建基于自然语言推理（NLI）的跨文档主张图，并引入受保护主张门控机制，可选择性地依据多方支持的一致子集生成答案，或直接跳过检索、参数化回答。我们还发布了一种最小编辑污染协议，涵盖四种扰动类型（明确错误、冲突、误导、虚构），用于构建匹配的干净、混合及全污染评估环境。在四个长文本问答基准上，针对多个商用与开源大模型的实验表明，污染会严重降低原始 RAG 的表现，而 MIRAGE 在混合与全污染条件下均能稳定恢复事实性，优于现有鲁棒 RAG 方法。代码与数据集已公开于 https://github.com/SaadElDine/MIRAGE。
原文摘要 · Abstract (English)
Retrieval-Augmented Generation (RAG) improves factuality by grounding LLMs in external evidence, but real-world retrieval is often polluted: semantically relevant passages may contain subtle misinformation, misleading framings, or fabrications. We introduce MIRAGE, a training-free, model-agnostic defense for long-form RAG. MIRAGE builds an NLI-based cross-document claim graph and applies a Defended-Claims Gate to either condition generation on a consistent, multi-source supported subset or to block retrieval and answer parametrically. We also release a minimal-edit pollution protocol spanning four perturbation families (Unambiguous, Conflicting, Misleading, Fabricated) to construct matched clean, mixed, and fully polluted evaluation regimes. Across four long-form QA benchmarks and multiple commercial and open-weight LLMs, pollution severely degrades vanilla RAG, while MIRAGE consistently restores factuality under mixed and fully polluted evidence and outperforms prior robust-RAG methods. Our implementation and datasets are available at https://github.com/SaadElDine/MIRAGE.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。