{"id":"https://openalex.org/W7159614210","doi":"https://doi.org/10.48550/arxiv.2604.27600","title":"Purifying Multimodal Retrieval: Fragment-Level Evidence Selection for RAG","display_name":"Purifying Multimodal Retrieval: Fragment-Level Evidence Selection for RAG","publication_year":2026,"publication_date":"2026-04-30","ids":{"openalex":"https://openalex.org/W7159614210","doi":"https://doi.org/10.48550/arxiv.2604.27600"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.27600","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.27600","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":null,"license_id":null,"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.2604.27600","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5134770534","display_name":"Xihang Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Xihang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134961607","display_name":"Zihan Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Zihan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134983538","display_name":"Chengkai Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Chengkai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134932084","display_name":"Cao Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Cao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134947648","display_name":"Ke Zeng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zeng, Ke","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134965873","display_name":"Quan Z. Sheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sheng, Quan Z.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5134930925","display_name":"Lina Yao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yao, Lina","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":0.7699000239372253,"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":0.7699000239372253,"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/T10028","display_name":"Topic Modeling","score":0.1324000060558319,"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/T10601","display_name":"Handwritten Text Recognition Techniques","score":0.0066999997943639755,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/selection","display_name":"Selection (genetic algorithm)","score":0.6980999708175659},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.5871999859809875},{"id":"https://openalex.org/keywords/fragment","display_name":"Fragment (logic)","score":0.54339998960495},{"id":"https://openalex.org/keywords/metric","display_name":"Metric (unit)","score":0.5288000106811523},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5268999934196472},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.4925999939441681},{"id":"https://openalex.org/keywords/margin","display_name":"Margin (machine learning)","score":0.40059998631477356},{"id":"https://openalex.org/keywords/identification","display_name":"Identification (biology)","score":0.38519999384880066}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7484999895095825},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.6980999708175659},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5871999859809875},{"id":"https://openalex.org/C2776235265","wikidata":"https://www.wikidata.org/wiki/Q18392052","display_name":"Fragment (logic)","level":2,"score":0.54339998960495},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5293999910354614},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.5288000106811523},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5268999934196472},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4941999912261963},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.4925999939441681},{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.40059998631477356},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.38519999384880066},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.3418999910354614},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3057999908924103},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.29899999499320984},{"id":"https://openalex.org/C197947376","wikidata":"https://www.wikidata.org/wiki/Q5155608","display_name":"Comparability","level":2,"score":0.29100000858306885},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2840999960899353},{"id":"https://openalex.org/C187029079","wikidata":"https://www.wikidata.org/wiki/Q958679","display_name":"Cognitive reframing","level":2,"score":0.274399995803833},{"id":"https://openalex.org/C2985684807","wikidata":"https://www.wikidata.org/wiki/Q1513879","display_name":"Text generation","level":2,"score":0.2703999876976013},{"id":"https://openalex.org/C149629883","wikidata":"https://www.wikidata.org/wiki/Q660926","display_name":"Fraction (chemistry)","level":2,"score":0.2685999870300293},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2606000006198883},{"id":"https://openalex.org/C2780898871","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Performance metric","level":2,"score":0.25999999046325684},{"id":"https://openalex.org/C23123220","wikidata":"https://www.wikidata.org/wiki/Q816826","display_name":"Information retrieval","level":1,"score":0.2533000111579895},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.25270000100135803}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.27600","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.27600","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.27600","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.27600","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"score":0.4271901845932007,"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Multimodal":[0,8],"Retrieval-Augmented":[1],"Generation":[2],"(MRAG)":[3],"is":[4,40,54],"widely":[5],"adopted":[6],"for":[7,92],"Large":[9],"Language":[10],"Models":[11],"(MLLMs)":[12],"with":[13,177],"external":[14],"evidence":[15,28,84,126,175],"to":[16,56,70,152,197],"reduce":[17],"hallucinations.":[18],"Despite":[19],"its":[20],"success,":[21],"most":[22],"existing":[23],"MRAG":[24,80,193],"frameworks":[25],"treat":[26],"retrieved":[27,111],"as":[29,81,106],"indivisible":[30],"documents,":[31],"implicitly":[32],"assuming":[33],"that":[34,67,97,127,144,187],"all":[35],"content":[36,63],"within":[37],"a":[38,48,52,57,82,95,141,166,170],"document":[39,53],"equally":[41],"informative.":[42],"In":[43],"practice,":[44],"however,":[45],"sometimes":[46],"only":[47],"small":[49],"fraction":[50],"of":[51,125,149],"relevant":[55],"given":[58],"query,":[59],"while":[60,217],"the":[61,146,153,183],"remaining":[62],"introduces":[64],"substantial":[65],"noise":[66],"may":[68],"lead":[69],"performance":[71],"degradation.":[72],"We":[73,87],"address":[74],"this":[75],"fundamental":[76],"limitation":[77],"by":[78],"reframing":[79],"fine-grained":[83],"selection":[85,176],"problem.":[86],"propose":[88],"Fragment-level":[89],"Evidence":[90],"Selection":[91],"RAG":[93],"(FES-RAG),":[94],"framework":[96],"selects":[98],"atomic":[99],"multimodal":[100,112],"fragments":[101,117],"rather":[102],"than":[103],"entire":[104],"documents":[105,113],"grounding":[107],"evidence.":[108],"FES-RAG":[109,188],"decomposes":[110],"into":[114,169],"sentence-level":[115],"textual":[116],"and":[118,221],"region-level":[119],"visual":[120],"fragments,":[121,210],"enabling":[122],"precise":[123],"identification":[124],"directly":[128],"supports":[129],"generation.":[130],"To":[131],"guide":[132],"fragment":[133,151],"selection,":[134],"we":[135,160],"introduce":[136],"Fragment":[137],"Information":[138],"Gain":[139],"(FIG),":[140],"principled":[142],"metric":[143],"measures":[145],"marginal":[147],"contribution":[148],"each":[150],"MLLM's":[154],"generation":[155,222],"confidence.":[156],"Based":[157],"on":[158,182],"FIG,":[159],"distill":[161],"fragment-level":[162],"utility":[163],"judgments":[164],"from":[165],"high-capacity":[167],"MLLM":[168],"lightweight":[171],"selector,":[172],"achieving":[173,195],"accurate":[174],"low":[178],"inference":[179],"overhead.":[180],"Experiments":[181],"M2RAG":[184],"benchmark":[185],"show":[186],"consistently":[189],"outperforms":[190],"state-of-the-art":[191],"document-level":[192],"methods,":[194],"up":[196],"27":[198],"percent":[199],"relative":[200],"improvement":[201],"in":[202],"CIDEr.":[203],"By":[204],"selecting":[205],"fewer":[206],"yet":[207],"more":[208],"informative":[209],"our":[211],"approach":[212],"substantially":[213],"reduces":[214],"context":[215],"length":[216],"improving":[218],"factual":[219],"accuracy":[220],"coherence.":[223]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-02T00:00:00"}