{"id":"https://openalex.org/W7160650750","doi":"https://doi.org/10.48550/arxiv.2605.05594","title":"The Cost of Context: Mitigating Textual Bias in Multimodal Retrieval-Augmented Generation","display_name":"The Cost of Context: Mitigating Textual Bias in Multimodal Retrieval-Augmented Generation","publication_year":2026,"publication_date":"2026-05-07","ids":{"openalex":"https://openalex.org/W7160650750","doi":"https://doi.org/10.48550/arxiv.2605.05594"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.05594","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.05594","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":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.05594","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5073350093","display_name":"Hoin Jung","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jung, Hoin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135686226","display_name":"Xiaoqian Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Xiaoqian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.39590001106262207,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.39590001106262207,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.06199999898672104,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.05889999866485596,"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/context","display_name":"Context (archaeology)","score":0.554099977016449},{"id":"https://openalex.org/keywords/copying","display_name":"Copying","score":0.49459999799728394},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.47600001096725464},{"id":"https://openalex.org/keywords/closed-captioning","display_name":"Closed captioning","score":0.46459999680519104},{"id":"https://openalex.org/keywords/debiasing","display_name":"Debiasing","score":0.4438999891281128},{"id":"https://openalex.org/keywords/illusion","display_name":"Illusion","score":0.4212000072002411},{"id":"https://openalex.org/keywords/reliability","display_name":"Reliability (semiconductor)","score":0.4065999984741211},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.37400001287460327}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7046999931335449},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.554099977016449},{"id":"https://openalex.org/C2779151265","wikidata":"https://www.wikidata.org/wiki/Q1156791","display_name":"Copying","level":2,"score":0.49459999799728394},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4918999969959259},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.47600001096725464},{"id":"https://openalex.org/C157657479","wikidata":"https://www.wikidata.org/wiki/Q2367247","display_name":"Closed captioning","level":3,"score":0.46459999680519104},{"id":"https://openalex.org/C2779458634","wikidata":"https://www.wikidata.org/wiki/Q24963715","display_name":"Debiasing","level":2,"score":0.4438999891281128},{"id":"https://openalex.org/C184047640","wikidata":"https://www.wikidata.org/wiki/Q182593","display_name":"Illusion","level":2,"score":0.4212000072002411},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.41029998660087585},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.4065999984741211},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.37439998984336853},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.37400001287460327},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.36880001425743103},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.3328999876976013},{"id":"https://openalex.org/C9770341","wikidata":"https://www.wikidata.org/wiki/Q1938983","display_name":"Geospatial analysis","level":2,"score":0.32010000944137573},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.3095000088214874},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.29179999232292175},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.28700000047683716},{"id":"https://openalex.org/C2777508537","wikidata":"https://www.wikidata.org/wiki/Q7936620","display_name":"Visual reasoning","level":2,"score":0.2773999869823456},{"id":"https://openalex.org/C2778712577","wikidata":"https://www.wikidata.org/wiki/Q3505966","display_name":"Retraining","level":2,"score":0.2741999924182892},{"id":"https://openalex.org/C2780665704","wikidata":"https://www.wikidata.org/wiki/Q959298","display_name":"Intervention (counseling)","level":2,"score":0.26969999074935913},{"id":"https://openalex.org/C62354387","wikidata":"https://www.wikidata.org/wiki/Q875399","display_name":"Boundary (topology)","level":2,"score":0.2556999921798706},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.25110000371932983}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.05594","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.05594","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":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.05594","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.05594","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":"Preprint"},"sustainable_development_goals":[{"display_name":"Peace, Justice and strong institutions","score":0.4386718273162842,"id":"https://metadata.un.org/sdg/16"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"While":[0],"Multimodal":[1],"Large":[2],"Language":[3],"Models":[4],"(MLLMs)":[5],"are":[6,123],"increasingly":[7],"integrated":[8],"with":[9,136],"Retrieval-Augmented":[10],"Generation":[11],"(RAG)":[12],"to":[13,51,102,134,164],"mitigate":[14],"hallucinations,":[15],"the":[16,27,34,39,81,100,128,137],"introduction":[17,40],"of":[18,36,41,61,84,114],"external":[19],"documents":[20],"can":[21],"conceal":[22],"severe":[23],"failure":[24],"modes":[25],"at":[26],"instance":[28],"level.":[29],"We":[30],"identify":[31],"and":[32,89,92,160,172,180],"formalize":[33],"phenomenon":[35],"recorruption,":[37],"where":[38,127],"even":[42],"perfectly":[43],"accurate":[44],"\"oracle\"":[45],"context":[46],"causes":[47],"a":[48,58,72,94,152],"capable":[49],"model":[50,101,186],"abandon":[52],"an":[53,112],"initially":[54],"correct":[55,120],"prediction.":[56],"Through":[57],"mechanistic":[59],"diagnosis":[60],"internal":[62],"attention":[63,86],"matrices,":[64],"we":[65,144],"show":[66],"that":[67,98,117,156],"recorruption":[68],"is":[69],"driven":[70],"by":[71,80],"two-fold":[73],"attentional":[74],"collapse:":[75],"(1)":[76],"visual":[77,85,158],"blindness,":[78],"characterized":[79],"systemic":[82],"suppression":[83],"mass":[87],"($M_{vis}$)":[88],"sharpness":[90],"($S_{vis}$),":[91],"(2)":[93],"structural":[95],"positional":[96,125],"bias":[97,132],"forces":[99],"prioritize":[103],"boundary":[104],"tokens":[105],"over":[106],"semantic":[107],"relevance.":[108],"Our":[109],"analysis":[110],"reveals":[111],"Illusion":[113],"Success,":[115],"demonstrating":[116],"many":[118],"seemingly":[119],"RAG":[121],"outcomes":[122],"merely":[124],"coincidences":[126],"model's":[129],"textual":[130,165],"copying":[131],"happens":[133],"align":[135],"ground-truth":[138],"location.":[139],"To":[140],"address":[141],"these":[142],"vulnerabilities,":[143],"propose":[145],"Bottleneck":[146],"Attention":[147],"Intervention":[148],"for":[149],"Recovery":[150],"(BAIR),":[151],"parameter-free,":[153],"inference-time":[154],"framework":[155],"restores":[157,177],"saliency":[159],"applies":[161],"position-aware":[162],"penalties":[163],"distractors.":[166],"Across":[167],"medical":[168],"factuality,":[169],"social":[170],"fairness,":[171],"geospatial":[173],"benchmarks,":[174],"BAIR":[175],"successfully":[176],"multimodal":[178],"grounding":[179],"improves":[181],"diagnostic":[182],"reliability":[183],"without":[184],"requiring":[185],"retraining":[187],"or":[188],"fine-tuning.":[189]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-09T00:00:00"}