Structured Scientific Text Extraction With Olmocr and Lora: Results, Metrics and Challenges

Structured Scientific Text Extraction With Olmocr and Lora: Results, Metrics and Challenges | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Structured Scientific Text Extraction With Olmocr and Lora: Results, Metrics and Challenges Bruno Leonardo Santos Menezes, Fabio Andre Machado Porto This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7055155/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract This study evaluates the performance of the olmOCR-7B-0225-preview multimodal model in the semantic extraction of scientific PDFs in the field of meteorology, analyzing the base model and a specialized textual version derived from it, adjusted via Low-Rank Adaptation (LoRA), using only textual input. A pipeline was developed for the automated collection of articles, extraction and transcription of structured content, including formulas, tables and diagrams, followed by supervised textual training, built from image-text pairs. Output analysis was based on metrics such as perplexity, Type-Token Ratio (TTR) and Semantic Diversity Index (SDI). The results indicated that the fine-tuned model offers more predictable and lexically rich transcriptions, while the base model better preserves the semantic and visual structure of the documents. The textual model adjusted via LoRA does not represent a simple incremental improvement of the base multimodal model, but rather a functional specialization. While the base model operates in a multimodal regime (images and text), the LoRA model was derived to act exclusively on textual inputs and is not capable of performing multimodal inference. The comparisons made in this study refer to distinct linguistic and behavioral dimensions, not to direct performance benchmarks in the same task. Artificial Intelligence and Machine Learning scientific text extraction multimodal OCR with LoRA olmOCR Full Text Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7055155","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":481162676,"identity":"33a7747f-5963-45e6-af17-321d1e3a12b5","order_by":0,"name":"Bruno Leonardo Santos Menezes","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAqElEQVRIiWNgGAWjYBACAzBZwMDAzw5mHSBWC5CUbCZZi8FhYrWYS58xe/DBoC7a+DDvwQ8/GO7kE9Ri2ZdjbjjD4HDutsN8yZI9DM8sGwg67AyPmTSPwQGgFh4zZgaGwwYEbYFqqcvd3EyiFubcDczEarHsYSuTBPllxmEeY8keg2eEtZjzMG+T+FBRl9vf3mP44UfFHcJa0N1JqoZRMApGwSgYBVgBAOL6NClcNz2/AAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0001-6249-2395","institution":"National Laboratory for Scientific Computing (LNCC)","correspondingAuthor":true,"prefix":"","firstName":"Bruno","middleName":"Leonardo Santos","lastName":"Menezes","suffix":""},{"id":481162677,"identity":"764b7c2e-a592-4a5a-a469-3fe7e8687f86","order_by":1,"name":"Fabio Andre Machado Porto","email":"","orcid":"https://orcid.org/0000-0002-4597-4832","institution":"National Laboratory for Scientific Computing (LNCC)","correspondingAuthor":false,"prefix":"","firstName":"Fabio","middleName":"Andre Machado","lastName":"Porto","suffix":""}],"badges":[],"createdAt":"2025-07-05 23:35:27","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7055155/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7055155/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":86224495,"identity":"42c2b237-fcbc-4296-9d29-ac7c9b71da46","added_by":"auto","created_at":"2025-07-08 07:52:08","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":475273,"visible":true,"origin":"","legend":"","description":"","filename":"StructuredScientificTextExtractionwitholmOCRandLoRAResultsMetricsandChallenges2025.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7055155/v1_covered_164027b1-b1f5-44c1-9eb6-1c2d5f192634.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eStructured Scientific Text Extraction With Olmocr and Lora: Results, Metrics and Challenges\u003c/p\u003e","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"National Laboratory for Scientific Computing (LNCC)","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":" scientific text extraction, multimodal OCR with LoRA, olmOCR","lastPublishedDoi":"10.21203/rs.3.rs-7055155/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7055155/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study evaluates the performance of the olmOCR-7B-0225-preview multimodal model in the semantic extraction of scientific PDFs in the field of meteorology, analyzing the base model and a specialized textual version derived from it, adjusted via Low-Rank Adaptation (LoRA), using only textual input. A pipeline was developed for the automated collection of articles, extraction and transcription of structured content, including formulas, tables and diagrams, followed by supervised textual training, built from image-text pairs. Output analysis was based on metrics such as perplexity, Type-Token Ratio (TTR) and Semantic Diversity Index (SDI). The results indicated that the fine-tuned model offers more predictable and lexically rich transcriptions, while the base model better preserves the semantic and visual structure of the documents. The textual model adjusted via LoRA does not represent a simple incremental improvement of the base multimodal model, but rather a functional specialization. While the base model operates in a multimodal regime (images and text), the LoRA model was derived to act exclusively on textual inputs and is not capable of performing multimodal inference. The comparisons made in this study refer to distinct linguistic and behavioral dimensions, not to direct performance benchmarks in the same task.\u003c/p\u003e","manuscriptTitle":"Structured Scientific Text Extraction With Olmocr and Lora: Results, Metrics and Challenges","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-08 07:44:00","doi":"10.21203/rs.3.rs-7055155/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"216dfcb9-3f8b-43ca-9381-5ea110260b63","owner":[],"postedDate":"July 8th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":51199645,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2025-07-08T07:44:00+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-08 07:44:00","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7055155","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7055155","identity":"rs-7055155","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}