用奇异值分解替代大模型摘要，实现超快且稳定的树状检索增强生成。
SVD-RAG: Efficient Tree-Organized Retrieval-Augmented Generation via Singular Value Decomposition
- 通过奇异值分解从密集句向量中提取关键句子，无需调用大模型。
- 构建树结构速度提升317倍，耗能降低85%，召回率仅比原版低1-5%。
- 自动调节压缩比例，适合大规模部署和对效率敏感的场景。
检索增强生成（RAG）系统通过从外部知识库检索相关文档来增强大语言模型。Sarthi 等人（2024）提出的 RAPTOR 将文档组织为层次树结构以实现高效检索，但需在每个内部节点进行昂贵的基于大模型的抽象摘要，导致大规模部署成本过高。本文提出 SVD-RAG，是首个将奇异值分解（SVD）应用于密集句向量矩阵以实现层次 RAG 中可提取摘要的方法。与传统基于 TF-IDF 的 LSA 不同，SVD-RAG 利用现代嵌入模型丰富的语义表示，通过主成分中的能量贡献识别最信息量的句子。该方法具有三重优势：（1）确定性——相同输入结果一致，不依赖随机生成；（2）成本高效——树构建无需额外 API 调用，仅需初始嵌入，减少约 85% 的令牌消耗；（3）内容自适应——能量比阈值 τ 自动根据内容复杂度调整压缩程度。在使用相同语料库、聚类方式和束搜索的对照实验中，SVD-RAG 在检索质量上仅比采用大模型摘要的 RAPTOR 低 1-5%（MRR 0.867 vs. 0.875，Recall@1 0.483 vs. 0.458），而树构建速度提升 317 倍（0.1 秒 vs. 31.7 秒）。在包含 205 个块、100 个查询及 20 个主题变体的扩展多主题基准测试中，相较于扁平嵌入检索，SVD-RAG 在 Recall@1 上提升 4.2 倍，MRR 提升 3.1 倍。我们提供了详细的成本分析与参数敏感性研究，代码已开源。
原文摘要 · Abstract (English)
Retrieval-Augmented Generation (RAG) systems enhance large language models by retrieving relevant documents from external knowledge bases. Recent work by Sarthi et al. (2024) introduced RAPTOR, which organizes documents into hierarchical tree structures for efficient retrieval, but requires expensive LLM-based abstractive summarization at each internal node -- making large-scale deployment prohibitively costly. We present SVD-RAG, the first method to apply Singular Value Decomposition (SVD) on dense sentence embedding matrices for extractive summarization in hierarchical RAG. Unlike classical LSA which operates on sparse TF-IDF matrices, SVD-RAG exploits the rich semantic representations of modern embedding models, identifying the most informative sentences through their energy contribution in the principal components. Our approach is (1) deterministic -- unlike LLM-based summarization, SVD produces identical results for the same input; (2) cost-efficient -- tree construction requires no additional API calls beyond the initial embedding, reducing token consumption by ~85%; and (3) content-adaptive -- the energy-ratio threshold tau automatically adjusts compression based on content complexity. In a controlled head-to-head comparison using identical corpora, clustering, and beam search, SVD-RAG achieves retrieval quality within 1-5% of RAPTOR with LLM summarization (MRR 0.867 vs. 0.875, Recall@1 0.483 vs. 0.458) while building the tree 317x faster (0.1s vs. 31.7s). On a scaled multi-topic benchmark with 205 chunks and 100 queries across 20 topic variations, SVD-RAG achieves a 4.2x improvement in Recall@1 and 3.1x improvement in MRR over flat embedding retrieval. We provide a detailed cost analysis and parameter sensitivity study. Our implementation is released as an open-source Python package.
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。