[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83614-en":3,"doc-seo-83614-105":29,"detail-sidebar-cat-0-en-105":90},{"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":20,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},83614,16904993612988,"Olivia Brown","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","Evaluating Chunking Strategies for Retrieval Augmented Generation on Academic Texts","Retrieval-Augmented Generation (RAG) systems leverage Large Language Models to answer questions using information retrieved from outside their parameter space. This work evaluates whether cluster-based semantic chunking improves retrieval and answer quality compared with fixed-size and recursive chunking on long, structured academic theses. Experiments use the RAGAs framework and show that RAGAs faithfulness metrics have limited reliability. Performance differs substantially between fixed and document-specific questions, influenced by formatting and preprocessing. Under tested settings, cluster-based chunking does not outperform simpler strategies.","Evaluating Chunking Strategies for Retrieval-Augmented Generation on Academic Texts  \nValentin J. J. Kreileder  \nComputer Science Deggendorf Institute of Technology Deggendorf, Germany [kreileder@gmx.net](kreileder@gmx.net)  \nJohannes Reisinger  \nComputer Science Deggendorf Institute of Technology Deggendorf, Germany johannes.reisinger@th-deg.de  \nAndreas Fischer  \nComputer Science Deggendorf Institute of Technology Deggendorf, Germany andreas.fischer@th-deg.de  \narXiv :2607 .0 1852v 1 [ cs .IR] 2 Jul 2026  \nAbstract—Retrieval-Augmented Generation (RAG) systems use the question-answering capabilities of Large Language Models (LLMs) to access information outside their parameters. We evaluate if cluster-based semantic chunking improves retrieval and answer quality compared to fixed-size and recursive chunking evaluating on long, structured academic theses using the Retrieval Augmented Generation Assessment (RAGAs) framework. RAGAs based faithfulness shows limited reliability in this setup. Performance on fixed versus document specific questions varied substantially, likely related to the formatting of documents and preprocessing. Under the tested configuration, cluster-based chunking did not outperform simpler strategies.  \nIndex Terms—Chunking, Large language model, RetrievalAugmented Generation, Information retrieval  \nI. INTRODUCTION  \nLarge Language Models (LLMs) demonstrated impressive generative abilities, yet their responses are limited by the information encoded in their parameters, can suffer from hallucinations, and do not show their inner workings. RetrievalAugmented Generation (RAG) systems address these issues [1] . Long documents cannot be processed as a whole because of the embedding models and LLMs context window, therefore, those documents need to be split into smaller chunks. These chunks are extracted from external documents, which can be done with different strategies. Generated chunks are stored in a vector database before being retrieved with a user query. The LLM uses the query and chunks to generate an answer. The quality of the RAG-generated answer is coupled with the retrieval quality, the source data and the chunking strategy. Conventional strategies are fixed-sized chunking or format based recursive chunking. Semantic chunking gained prominence due to its potential to improve retrieval outcomes based on sentence similarity. However, semantic strategies like cluster based chunking are computationally more demanding compared to traditional methods [2] . Core contributions of this paper are as follows:  \n• We establish a measure combining faithfulness and answer relevancy called Answer Quality Score (AQS) .  \n• We show issues regarding the evaluation using RAGAson mid-range hardware.  \n• We show how different chunking strategies perform within this system using the RAGAs.  \nII. RELATED WORK  \nQu et al. demonstrate that plain fixed-size chunking is the most cost-effective approach [2] . Their experiments were conducted on a mixture of datasets including both original and artificially combined documents, differing from the documents used in this work. Their findings may not generalize to settings where documents are significantly larger and follow an internal structure. We also use smaller sentence encoder under hardware constraints, further limiting comparison. Late Chunking makes use of a long-context model to embed a full document, then applies chunking and mean pooling to produce chunk vectors with better surrounding context than pipelines embedding after chunking [3] . Within the group’s prior work, Reisinger et al. propose document-level knowledge graphs from the hierarchical structure of chunk embeddings to detect document versions and plagiarism, mitigating input-and context-conflicting hallucinations without a generative judge as done in this work [4] .  \nIII. METHODOLOGY  \nA. Chunking Strategies  \nChunking splits data into segments an LLM can process within its context window, and can be categori","cbCaisj2jwWf13j5","https://ap.wps.com/l/cbCaisj2jwWf13j5","pdf",692936,4,1,"English","en",105,"# Introduction\n# Related Work\n# Methodology\n## Chunking Strategies\n## RAGAs","[{\"question\":\"What problem does the paper address in Retrieval-Augmented Generation (RAG) systems?\",\"answer\":\"RAG answers depend on how documents are split into chunks that are embedded, stored, and retrieved. Long academic theses require chunking because models cannot process entire documents within context limits.\"},{\"question\":\"Which chunking strategies are compared in the study?\",\"answer\":\"The paper compares fixed-size chunking, recursive (format-based) chunking, and cluster-based semantic chunking, using a controlled experimental setup and RAGAs evaluation.\"},{\"question\":\"What do the results show about cluster-based chunking?\",\"answer\":\"With the tested configuration, cluster-based chunking does not outperform simpler strategies. The evaluation also indicates limited reliability for RAGAs faithfulness, and performance varies notably between fixed versus document-specific questions.\"}]",1784189272,10,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":27},"evaluating-chunking-strategies-for-retrieval-augmented-generation-on-academic-texts","",{"@graph":35,"@context":84},[36,52,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":20},"https://docshare.wps.com/document/evaluating-chunking-strategies-for-retrieval-augmented-generation-on-academic-texts/83614/",{"url":51,"name":13,"@type":53,"author":54,"headline":13,"publisher":56,"fileFormat":59,"inLanguage":23,"description":14,"dateModified":60,"datePublished":61,"encodingFormat":59,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":55},"Person",{"url":40,"name":57,"@type":58},"DocShare","Organization","application/pdf","2026-07-25","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":20},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"What problem does the paper address in Retrieval-Augmented Generation (RAG) systems?","Question",{"text":74,"@type":75},"RAG answers depend on how documents are split into chunks that are embedded, stored, and retrieved. Long academic theses require chunking because models cannot process entire documents within context limits.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"Which chunking strategies are compared in the study?",{"text":79,"@type":75},"The paper compares fixed-size chunking, recursive (format-based) chunking, and cluster-based semantic chunking, using a controlled experimental setup and RAGAs evaluation.",{"name":81,"@type":72,"acceptedAnswer":82},"What do the results show about cluster-based chunking?",{"text":83,"@type":75},"With the tested configuration, cluster-based chunking does not outperform simpler strategies. The evaluation also indicates limited reliability for RAGAs faithfulness, and performance varies notably between fixed versus document-specific questions.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":57,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,119,122,127,130,133],{"id":21,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":20,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":28,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":28,"slug":132},"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]