[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85886-en":3,"doc-seo-85886-105":30,"detail-sidebar-cat-0-en-105":92},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},85886,687197207639,"Asher","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","SVD-RAG Efficient Tree-Organized Retrieval-Augmented Generation via Singular Value Decomposition","Retrieval-Augmented Generation (RAG) improves large language models by retrieving relevant documents from external knowledge bases, yet large corpora make efficient retrieval costly. SVD-RAG applies Singular Value Decomposition (SVD) to dense sentence embedding matrices for extractive summarization inside hierarchical RAG trees, avoiding LLM-based abstractive summarization at internal nodes. The method is deterministic, cuts tree-construction token usage by ~85%, and adapts compression via an energy-ratio threshold τ. Head-to-head comparisons show near-RAPTOR retrieval quality (1–5% gap) with 317× faster tree building and improved performance on scaled multi-topic benchmarks, plus cost and sensitivity analysis.","arXiv :2607 . 103 16v 1 [ cs .IR] 11 Jul 2026  \nSVD-RAG: EFFICIENT TREE-ORGANIZED RETRIEVAL-AUGMENTED GENERATION  \nVIA SINGULAR VALUE DECOMPOSITION  \nA PREPRINT  \nZhihui Sun  \nJuly 14, 2026  \nABSTRACT  \nRetrieval-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.  \nWe 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 τ 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 317× faster (0.1s vs. 31.7s) . On a scaled multi-topic benchmark with 205 chunks and 100 queries across  \n20 topic variations, SVD-RAG achieves a 4.2× improvement in Recall@1 and 3.1 × 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.  \nKeywords RAG · Retrieval-Augmented Generation · SVD · Text Summarization · Hierarchical Retrieval · Tree-Organized Index  \n1 Introduction  \nRetrieval-Augmented Generation (RAG) has become a standard paradigm for grounding large language model outputsin external knowledge (Lewis et al., 2020) . By retrieving relevant documents before generation, RAG systems reduce hallucination and improve factual accuracy. However, as knowledge bases grow, efficient retrieval becomes increasingly challenging.  \nThe dominant approach—flat vector search with embedding similarity—suffers from two key limitations. First, pure cosine similarity between query and chunk embeddings cannot capture complex semantic relationships that require crossattention modeling (Vaswani et al., 2017) . Second, performing exhaustive search over all chunks is computationally expensive for large corpora, while reranking all candidates is cost-prohibitive.  \nRAPTOR (Sarthi et al., 2024) introduced a hierarchical solution: recursively cluster similar chunks, use an LLM to generate abstractive summaries for each cluster, and build a document tree. At query time, a beam search over the tree identifies relevant branches, and a reranker scores the final candidates. While effective, RAPTOR’s reliance on  \nLLM-based summarization for every internal node incurs substantial API costs. For a corpus of 10,000 chunks, building a tree of depth 3 with approximately 10 chunks per cluster would require on the order of 1,000 LLM summarization calls for internal nodes.  \nWe observe that the purpose of summarization in tree construction is not to generate fluent natural language, but to create information-dense representations that guide the search process. This insight motivates our use of extractive rather than abstractive summarization.  \nOur key innovation is applying Singular Value Decomposition (SVD) directly to sentence embedding matrices to identify the most semantically informative sentences in a do","cbCaisfoNodKB6vq","https://ap.wps.com/l/cbCaisfoNodKB6vq","pdf",336246,7,1,10,"English","en",105,"# Abstract\n# 1 Introduction\n## Limits of Flat Retrieval\n## RAPTOR and LLM-Cost Bottlenecks\n## Extractive Summarization Motivation\n## Core Idea: SVD on Embedding Matrices\n## Key Advantages and Experiments","[{\"question\":\"What problem does SVD-RAG address in hierarchical RAG systems?\",\"answer\":\"Hierarchical RAG methods like RAPTOR require LLM-based abstractive summarization at many internal tree nodes, which becomes prohibitively expensive at scale. SVD-RAG removes that overhead by using SVD-based extractive summarization instead.\"},{\"question\":\"How does SVD-RAG perform summarization for building the retrieval tree?\",\"answer\":\"SVD-RAG applies Singular Value Decomposition directly to dense sentence embedding matrices and selects sentences whose energy contribution in principal components is highest. The number of selected sentences is determined adaptively via cumulative energy retention.\"},{\"question\":\"What are the reported benefits compared with RAPTOR and flat embedding retrieval?\",\"answer\":\"SVD-RAG is deterministic and cost-efficient, reducing token consumption by about 85% during tree construction, and builds trees about 317× faster. In controlled comparisons it achieves retrieval quality within 1–5% of RAPTOR while outperforming flat embedding retrieval on scaled multi-topic benchmarks.\"}]",1784206957,25,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":28},"svd-rag-efficient-tree-organized-retrieval-augmented-generation-via-singular-value-decomposition","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/svd-rag-efficient-tree-organized-retrieval-augmented-generation-via-singular-value-decomposition/85886/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":24,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-07-25","2026-07-16",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"What problem does SVD-RAG address in hierarchical RAG systems?","Question",{"text":76,"@type":77},"Hierarchical RAG methods like RAPTOR require LLM-based abstractive summarization at many internal tree nodes, which becomes prohibitively expensive at scale. SVD-RAG removes that overhead by using SVD-based extractive summarization instead.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does SVD-RAG perform summarization for building the retrieval tree?",{"text":81,"@type":77},"SVD-RAG applies Singular Value Decomposition directly to dense sentence embedding matrices and selects sentences whose energy contribution in principal components is highest. The number of selected sentences is determined adaptively via cumulative energy retention.",{"name":83,"@type":74,"acceptedAnswer":84},"What are the reported benefits compared with RAPTOR and flat embedding retrieval?",{"text":85,"@type":77},"SVD-RAG is deterministic and cost-efficient, reducing token consumption by about 85% during tree construction, and builds trees about 317× faster. In controlled comparisons it achieves retrieval quality within 1–5% of RAPTOR while outperforming flat embedding retrieval on scaled multi-topic benchmarks.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":93},[94,98,102,106,111,116,120,123,128,131,134],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":107,"doc_module":4,"doc_module_name":46,"category_name":108,"show_sort_weight":109,"slug":110},5,"Comic",60,"comic",{"id":112,"doc_module":4,"doc_module_name":46,"category_name":113,"show_sort_weight":114,"slug":115},6,"Technology",50,"technology",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":22,"doc_module":4,"doc_module_name":46,"category_name":132,"show_sort_weight":22,"slug":133},"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":107,"slug":137},19,"General","general"]