{"id":"https://openalex.org/W7155062398","doi":"https://doi.org/10.48550/arxiv.2604.16422","title":"Injecting Structured Biomedical Knowledge into Language Models: Continual Pretraining vs. GraphRAG","display_name":"Injecting Structured Biomedical Knowledge into Language Models: Continual Pretraining vs. GraphRAG","publication_year":2026,"publication_date":"2026-04-03","ids":{"openalex":"https://openalex.org/W7155062398","doi":"https://doi.org/10.48550/arxiv.2604.16422"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.16422","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.16422","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.16422","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5116850710","display_name":"Jaafer Klila","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Klila, Jaafer","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5004852014","display_name":"Sondes Bannour Souihi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Souihi, Sondes Bannour","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134193608","display_name":"Rahma Boujelben","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Boujelben, Rahma","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5110302916","display_name":"Nasredine Semmar","orcid":"https://orcid.org/0000-0002-4685-6053"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Semmar, Nasredine","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5102717363","display_name":"Lamia Hadrich Belguith","orcid":"https://orcid.org/0000-0002-5063-9246"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Belguith, Lamia Hadrich","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":5,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.36820000410079956,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.36820000410079956,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11710","display_name":"Biomedical Text Mining and Ontologies","score":0.13490000367164612,"subfield":{"id":"https://openalex.org/subfields/1312","display_name":"Molecular Biology"},"field":{"id":"https://openalex.org/fields/13","display_name":"Biochemistry, Genetics and Molecular Biology"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},{"id":"https://openalex.org/T13702","display_name":"Machine Learning in Healthcare","score":0.13050000369548798,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/unified-medical-language-system","display_name":"Unified Medical Language System","score":0.8679999709129333},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6725999712944031},{"id":"https://openalex.org/keywords/knowledge-base","display_name":"Knowledge base","score":0.6385999917984009},{"id":"https://openalex.org/keywords/graph","display_name":"Graph","score":0.5825999975204468},{"id":"https://openalex.org/keywords/construct","display_name":"Construct (python library)","score":0.5361999869346619},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.5047000050544739},{"id":"https://openalex.org/keywords/knowledge-graph","display_name":"Knowledge graph","score":0.5041000247001648},{"id":"https://openalex.org/keywords/question-answering","display_name":"Question answering","score":0.49140000343322754},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.45660001039505005}],"concepts":[{"id":"https://openalex.org/C69505689","wikidata":"https://www.wikidata.org/wiki/Q455338","display_name":"Unified Medical Language System","level":2,"score":0.8679999709129333},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7588000297546387},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6725999712944031},{"id":"https://openalex.org/C4554734","wikidata":"https://www.wikidata.org/wiki/Q593744","display_name":"Knowledge base","level":2,"score":0.6385999917984009},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.6079000234603882},{"id":"https://openalex.org/C132525143","wikidata":"https://www.wikidata.org/wiki/Q141488","display_name":"Graph","level":2,"score":0.5825999975204468},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5490999817848206},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.5361999869346619},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.5047000050544739},{"id":"https://openalex.org/C2987255567","wikidata":"https://www.wikidata.org/wiki/Q33002955","display_name":"Knowledge graph","level":2,"score":0.5041000247001648},{"id":"https://openalex.org/C44291984","wikidata":"https://www.wikidata.org/wiki/Q1074173","display_name":"Question answering","level":2,"score":0.49140000343322754},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.45660001039505005},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4390000104904175},{"id":"https://openalex.org/C23123220","wikidata":"https://www.wikidata.org/wiki/Q816826","display_name":"Information retrieval","level":1,"score":0.39590001106262207},{"id":"https://openalex.org/C176225458","wikidata":"https://www.wikidata.org/wiki/Q595971","display_name":"Graph database","level":3,"score":0.3912000060081482},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.3625999987125397},{"id":"https://openalex.org/C2985722590","wikidata":"https://www.wikidata.org/wiki/Q11190","display_name":"Medical knowledge","level":2,"score":0.32829999923706055},{"id":"https://openalex.org/C115925183","wikidata":"https://www.wikidata.org/wiki/Q1412694","display_name":"Knowledge-based systems","level":2,"score":0.3246999979019165},{"id":"https://openalex.org/C120567893","wikidata":"https://www.wikidata.org/wiki/Q1582085","display_name":"Knowledge extraction","level":2,"score":0.3109000027179718},{"id":"https://openalex.org/C2474386","wikidata":"https://www.wikidata.org/wiki/Q461183","display_name":"Text corpus","level":2,"score":0.30399999022483826},{"id":"https://openalex.org/C207685749","wikidata":"https://www.wikidata.org/wiki/Q2088941","display_name":"Domain knowledge","level":2,"score":0.30399999022483826},{"id":"https://openalex.org/C2777530160","wikidata":"https://www.wikidata.org/wiki/Q41796","display_name":"Sentence","level":2,"score":0.2924000024795532},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.2897999882698059},{"id":"https://openalex.org/C2522767166","wikidata":"https://www.wikidata.org/wiki/Q2374463","display_name":"Data science","level":1,"score":0.287200003862381},{"id":"https://openalex.org/C3018949938","wikidata":"https://www.wikidata.org/wiki/Q17166101","display_name":"Text messaging","level":2,"score":0.2831999957561493},{"id":"https://openalex.org/C96711827","wikidata":"https://www.wikidata.org/wiki/Q17012245","display_name":"Entity linking","level":3,"score":0.2685999870300293},{"id":"https://openalex.org/C14156362","wikidata":"https://www.wikidata.org/wiki/Q3235388","display_name":"Descriptive knowledge","level":2,"score":0.26660001277923584}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.16422","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.16422","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.16422","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.16422","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"sustainable_development_goals":[{"score":0.7740537524223328,"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"The":[0],"injection":[1],"of":[2],"domain-specific":[3],"knowledge":[4,36,46,59,70,202],"is":[5],"crucial":[6],"for":[7,33,84],"adapting":[8],"language":[9],"models":[10,112],"(LMs)":[11],"to":[12,97,211],"specialized":[13],"fields":[14],"such":[15],"as":[16],"biomedicine.":[17],"While":[18],"most":[19],"current":[20],"approaches":[21],"rely":[22],"on":[23,113,130,150,154,186,191],"unstructured":[24],"text":[25,170],"corpora,":[26],"this":[27,95],"study":[28],"explores":[29],"two":[30,100,132],"complementary":[31],"strategies":[32],"leveraging":[34],"structured":[35],"from":[37,72,94],"the":[38,131,147,163,206],"UMLS":[39,73,208],"Metathesaurus:":[40],"(i)":[41],"Continual":[42],"pretraining":[43],"that":[44,56],"embeds":[45],"into":[47],"model":[48,165],"parameters,":[49],"and":[50,77,105,119,127,188,199],"(ii)":[51],"Graph":[52],"Retrieval-Augmented":[53],"Generation":[54],"(GraphRAG)":[55],"consults":[57],"a":[58,67,90],"graph":[60,71,96,210],"at":[61],"inference":[62],"time.":[63],"We":[64,87,109,204],"first":[65],"construct":[66],"large-scale":[68],"biomedical":[69,169],"(3.4":[74],"million":[75,79],"concepts":[76],"34.2":[78],"relations),":[80],"stored":[81],"in":[82],"Neo4j":[83,209],"efficient":[85],"querying.":[86],"then":[88],"derive":[89],"~100-million-token":[91],"textual":[92],"corpus":[93],"continually":[98],"pretrain":[99],"models:":[101],"BERTUMLS":[102,142],"(from":[103,107],"BERT)":[104],"BioBERTUMLS":[106],"BioBERT).":[108],"evaluate":[110,128],"these":[111],"six":[114],"BLURB":[115,140],"(Biomedical":[116],"Language":[117],"Understanding":[118],"Reasoning":[120],"Benchmark)":[121],"datasets":[122,136],"spanning":[123],"five":[124],"task":[125],"types":[126],"GraphRAG":[129,178],"QA":[133],"(Question":[134],"Answering)":[135],"(PubMedQA,":[137],"BioASQ).":[138],"On":[139],"tasks,":[141],"improves":[143],"over":[144,181],"BERT,":[145],"with":[146,176],"largest":[148],"gains":[149],"knowledge-intensive":[151],"QA.":[152],"Effects":[153],"BioBERT":[155],"are":[156],"more":[157],"nuanced,":[158],"suggesting":[159],"diminishing":[160],"returns":[161],"when":[162],"base":[164],"already":[166],"encodes":[167],"substantial":[168],"knowledge.":[171],"Finally,":[172],"augmenting":[173],"LLaMA":[174],"3-8B":[175],"our":[177],"pipeline":[179],"yields":[180],"than":[182],"3":[183],"points":[184,190],"accuracy":[185],"PubMedQA":[187],"5":[189],"BioASQ":[192],"without":[193],"any":[194],"retraining,":[195],"delivering":[196],"transparent,":[197],"multi-hop,":[198],"easily":[200],"updated":[201],"access.":[203],"release":[205],"processed":[207],"support":[212],"reproducibility.":[213]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-04-22T00:00:00"}