对比三种RAG架构在多跳溯源中的鲁棒性，发现图结构检索普遍存在过度引用问题。
Universal Pathologies, Conditional Consequences: A Triple-Robustness Analysis of RAG for Multi-Hop Traceability
- 固定检索架构，变换嵌入模型、语料库和判断标准，进行三重鲁棒性测试。
- 图检索在所有场景下均存在过度引用，每答案引用11-15个编号，准确率仅0.12-0.23。
- 其结果可靠性依赖语料库类型，对维基链路支持度高，对航空标准文档则严重失真。
GraphRAG在多项报告中表现不如向量检索的引用精确度，但具体原因和发生场景仍局限于特定语料库。本文通过三重鲁棒性分析，在固定检索架构的前提下，分别改变嵌入模型（e5-small → Azure text-embedding-3-small）、语料库（DO-178C typed-edge requirements → MuSiQue生成的维基段落链）与判断方式（配对GPT-5.4 vs GPT-4.1），完成4,440次主矩阵运行、600次跨语料运行及1,200次忠实性判断。结果显示：(C2a) 过度引用是架构固有现象——无论何种设置，GraphRAG每答案输出11–15个引用编号，引用精确度为0.12–0.23，召回率为0.68–0.87；(C2b) 其忠实性后果依赖语料库：在DO-178C中，两跳任务忠实性从74%降至40%；而在维基链路中反而从42%升至58%，因被过度引用段落仍具主题相关性；(C1) 胜者具有分层条件性：在DO-178C上原始方法胜出，于MuSiQue上图检索胜出，且两种嵌入模型下结果一致；(C3) 单一大模型判断忠实性对检索状态极其敏感，同一模型自一致性kappa仅为0.137（GPT-5.4），41%的判断项发生变化；(C4) 仅基于密集嵌入的可学习路由器在跳数分类上达到宏平均F1 0.86。我们主张三重鲁棒性是可信RAG架构声明的最低标准。
原文摘要 · Abstract (English)
GraphRAG underperforms vector RAG on citation precision in many reports, but where and why have remained corpus-bound. We present a triple-robustness analysis that holds the retrieval architecture fixed and varies three orthogonal axes embedder (local e5-small -> Azure text-embedding-3-small), corpus (DO-178C typed-edge requirements -> Wikipedia paragraph chains via MuSiQue), and judge (paired GPT-5.4 x GPT-4.1) across 4,440 main-matrix runs, 600 cross-corpus runs, and 1,200 paired faithfulness judgments. (C2a) Over-citation is architecturally universal: GraphRAG emits 11-15 IDs per answer at citation precision 0.12-0.23 and retrieval recall 0.68-0.87 across all three settings. (C2b) Its faithfulness consequence is corpus-conditional: in typed-edge DO-178C, GraphRAG faithfulness collapses 74%->40% across hops; on Wikipedia chains the same pipeline rises 42%->58% because over-cited paragraphs remain topically supporting. (C1) Stratum-conditional winners are corpus-conditional but embedder-robust: vanilla wins 2-hop on DO-178C, GraphRAG wins 2-hop on MuSiQue, identical under either embedder. (C3) Single-judge LLM faithfulness is fragile to retrieval state: same-judge self-kappa across embedders is 0.137 for GPT-5.4 (verdict change on 41% of items). A learned router on dense embeddings alone reaches macro-F1 0.86 on hop classification (C4). We argue triple-robustness is the minimum bar for trustworthy RAG architecture claims.
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。