SCAR通过语义连续性感知，高效修复RAG中因分块导致的证据断裂问题。
SCAR: Semantic Continuity-Aware Retrieval for Efficient Context Expansion in RAG
- 根据查询与邻近段落的相关性，动态决定是否扩展上下文。
- 在边界碎片化查询上实现92.8%召回率，仅需7.84个片段（减少22.9%）。
- 跨模型通用，降低上下文令牌数27.1%，适合追求效率的RAG应用。
RAG中的固定长度分块常导致关键证据被切割至不同片段，影响检索召回率。尽管静态窗口和父级检索可提升召回，但带来显著的令牌开销。本文提出SCAR（语义连续性感知检索），一种自适应检索策略，通过权衡查询-邻近相关性与结构连续性惩罚，选择性扩展邻近片段。该策略采用相对扩展阈值，与每个检索片段自身的查询相关性挂钩，形成近似尺度不变的决策规则，无需重校准即可跨嵌入模型迁移。在四个多样化语料库（RFC、GDPR、10-K报告、并购协议；共320个查询，160个边界碎片化查询）上，SCAR在边界碎片化查询中达到92.8%召回率，仅需7.84个片段，相比静态窗口（10.16片段）减少22.9%。配对置换检验（B=10,000）确认该减少高度显著（p<0.0001，Cohen's d=-1.49，大效应），召回差异小（Cohen's d=-0.33）。该策略在三种嵌入模型（text-embedding-3-large、BGE-large-en-v1.5、zembed-1）上使用同一超参数设置实现迁移，并在10-K语料库的下游RAGAS评估中证实，生成忠实度不受影响，同时上下文令牌减少27.1%。
原文摘要 · Abstract (English)
Fixed-length chunking in Retrieval-Augmented Generation (RAG) often leads to boundary fragmentation, where critical evidence is split across segments, degrading retrieval recall. While static windowing and parent retrieval improve recall, they introduce significant token overhead. We propose SCAR (Semantic Continuity-Aware Retrieval), an adaptive retrieval policy that selectively expands neighboring chunks by weighing query-neighbor relevance against a structural continuity penalty. SCAR uses a relative expansion threshold tied to each retrieved chunk's own query-relevance, yielding an approximately scale-invariant decision rule that transfers across embedding models without recalibration. Across four diverse corpora (RFC, GDPR, a 10-K report, and a Merger agreement; N=320 queries; 160 boundary-fragmented), SCAR achieves 92.8% recall on boundary-fragmented queries with only 7.84 chunks, a 22.9% reduction compared to static windowing (10.16 chunks). Paired bootstrap tests (B=10,000) confirm the chunk reduction is highly significant (p<0.0001, Cohen's d=-1.49, large effect), with a small recall difference (Cohen's d=-0.33). The policy transfers across three embedding models (text-embedding-3-large, BGE-large-en-v1.5, zembed-1) using the same single hyperparameter setting, and downstream RAGAS evaluation on the 10-K corpus confirms SCAR preserves generation faithfulness while reducing context tokens by 27.1%.
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。