{"id":"https://openalex.org/W7140350757","doi":null,"title":"Enhancing Text-to-SQL Systems Through LLM Routing and Effective Evaluation","display_name":"Enhancing Text-to-SQL Systems Through LLM Routing and Effective Evaluation","publication_year":2025,"publication_date":"2025-12-01","ids":{"openalex":"https://openalex.org/W7140350757"},"language":"en","primary_location":{"id":"pmh:oai:publications.polymtl.ca:71174","doi":null,"is_oa":true,"landing_page_url":"https://publications.polymtl.ca/71174/","pdf_url":"https://publications.polymtl.ca/71174/36/2025_MohammadhosseinMalekpour.pdf","source":{"id":"https://openalex.org/S4306401013","display_name":"PolyPublie (\u00c9cole Polytechnique de Montr\u00e9al)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I45683168","host_organization_name":"Polytechnique Montr\u00e9al","host_organization_lineage":["https://openalex.org/I45683168"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Malekpour, M. (2025). Enhancing Text-to-SQL Systems Through LLM Routing and Effective Evaluation [Master's thesis, Polytechnique Montr\u00e9al].","raw_type":"m\u00e9moire de ma\u00eetrise"},"volume":null,"type":"dissertation","is_traditional":true,"indexed_in":[],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://publications.polymtl.ca/71174/36/2025_MohammadhosseinMalekpour.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5109656446","display_name":"Mohammadhossein Malekpour","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Malekpour, Mohammadhossein","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5109656446"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.3043999969959259,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.3043999969959259,"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/T10028","display_name":"Topic Modeling","score":0.2890999913215637,"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/T10317","display_name":"Advanced Database Systems and Queries","score":0.2207999974489212,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/text-to-sql","display_name":"text-to-SQL","score":0.9769999980926514},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.777999997138977},{"id":"https://openalex.org/keywords/llm-routing","display_name":"LLM routing","score":0.9319999814033508},{"id":"https://openalex.org/keywords/cost-aware-routing","display_name":"cost-aware routing","score":0.675000011920929},{"id":"https://openalex.org/keywords/query-routing","display_name":"query routing","score":0.3059999942779541},{"id":"https://openalex.org/keywords/evaluation-indicators","display_name":"evaluation metrics","score":0.4959999918937683},{"id":"https://openalex.org/keywords/execution-accuracy","display_name":"execution accuracy","score":0.39800000190734863},{"id":"https://openalex.org/keywords/decision-support","display_name":"decision support","score":0.3959999978542328}],"concepts":[{"id":"https://openalex.org/C15708023","wikidata":"https://www.wikidata.org/wiki/Q80083","display_name":"Humanities","level":1,"score":0.7613000273704529},{"id":"https://openalex.org/C17744445","wikidata":"https://www.wikidata.org/wiki/Q36442","display_name":"Political science","level":0,"score":0.6524999737739563},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.24480000138282776},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.21289999783039093},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.1793999969959259},{"id":"https://openalex.org/C144024400","wikidata":"https://www.wikidata.org/wiki/Q21201","display_name":"Sociology","level":0,"score":0.13950000703334808},{"id":"https://openalex.org/C142362112","wikidata":"https://www.wikidata.org/wiki/Q735","display_name":"Art","level":0,"score":0.13899999856948853},{"id":"https://openalex.org/C2777230681","wikidata":"https://www.wikidata.org/wiki/Q7923820","display_name":"Vetting","level":2,"score":0.11550000309944153},{"id":"https://openalex.org/C76155785","wikidata":"https://www.wikidata.org/wiki/Q418","display_name":"Telecommunications","level":1,"score":0.10440000146627426},{"id":"https://openalex.org/C2983435990","wikidata":"https://www.wikidata.org/wiki/Q22725","display_name":"Network routing","level":3,"score":0.10350000113248825}],"mesh":[],"locations_count":1,"locations":[{"id":"pmh:oai:publications.polymtl.ca:71174","doi":null,"is_oa":true,"landing_page_url":"https://publications.polymtl.ca/71174/","pdf_url":"https://publications.polymtl.ca/71174/36/2025_MohammadhosseinMalekpour.pdf","source":{"id":"https://openalex.org/S4306401013","display_name":"PolyPublie (\u00c9cole Polytechnique de Montr\u00e9al)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I45683168","host_organization_name":"Polytechnique Montr\u00e9al","host_organization_lineage":["https://openalex.org/I45683168"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Malekpour, M. (2025). Enhancing Text-to-SQL Systems Through LLM Routing and Effective Evaluation [Master's thesis, Polytechnique Montr\u00e9al].","raw_type":"m\u00e9moire de ma\u00eetrise"}],"best_oa_location":{"id":"pmh:oai:publications.polymtl.ca:71174","doi":null,"is_oa":true,"landing_page_url":"https://publications.polymtl.ca/71174/","pdf_url":"https://publications.polymtl.ca/71174/36/2025_MohammadhosseinMalekpour.pdf","source":{"id":"https://openalex.org/S4306401013","display_name":"PolyPublie (\u00c9cole Polytechnique de Montr\u00e9al)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I45683168","host_organization_name":"Polytechnique Montr\u00e9al","host_organization_lineage":["https://openalex.org/I45683168"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Malekpour, M. (2025). Enhancing Text-to-SQL Systems Through LLM Routing and Effective Evaluation [Master's thesis, Polytechnique Montr\u00e9al].","raw_type":"m\u00e9moire de ma\u00eetrise"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7140350757.pdf","grobid_xml":"https://content.openalex.org/works/W7140350757.grobid-xml"},"referenced_works_count":29,"referenced_works":["https://openalex.org/W2145618437","https://openalex.org/W2269738476","https://openalex.org/W2751448157","https://openalex.org/W2768409085","https://openalex.org/W2965373594","https://openalex.org/W2972571786","https://openalex.org/W3093819145","https://openalex.org/W3134835171","https://openalex.org/W3168867926","https://openalex.org/W3198002980","https://openalex.org/W4210716649","https://openalex.org/W4226053975","https://openalex.org/W4304195432","https://openalex.org/W4313532427","https://openalex.org/W4317716303","https://openalex.org/W4317940064","https://openalex.org/W4377372223","https://openalex.org/W4382202531","https://openalex.org/W4392453936","https://openalex.org/W4396571402","https://openalex.org/W4398234583","https://openalex.org/W4399175046","https://openalex.org/W4401857375","https://openalex.org/W4402670291","https://openalex.org/W4404344612","https://openalex.org/W4404406411","https://openalex.org/W4407123926","https://openalex.org/W4415063137","https://openalex.org/W4417287252"],"related_works":["https://openalex.org/W2798663534","https://openalex.org/W4407953095","https://openalex.org/W7128657185","https://openalex.org/W7131791502","https://openalex.org/W7212968308","https://openalex.org/W4403883262","https://openalex.org/W4396571402","https://openalex.org/W4406072772","https://openalex.org/W7204902635","https://openalex.org/W3093923671"],"abstract_inverted_index":{"R\u00c9SUM\u00c9:":[0],"L'exploration":[1],"des":[2,62,74,87,115,120,161,171,250,324,327,332,336,415,440,444,456,470,477,500,510,524],"donn\u00e9es":[3,328],"par":[4,262,267,436,443],"requ\u00eates":[5,125,229],"constitue":[6],"un":[7,138,187,212,221,279,406],"goulot":[8],"d'\u00e9tranglement":[9],"op\u00e9rationnel":[10],"dans":[11,523],"la":[12,254,259,318,351,354,364,367,385,390,393,419,452,463,487,497,513],"prise":[13],"de":[14,25,36,38,90,100,134,141,180,189,191,210,217,224,238,243,245,256,271,290,322,344,353,366,369,392,429,446,459,461,472,475,491,519,528],"d\u00e9cision":[15,246],"en":[16,50,72,76,80,98,102,298,376,383,434,438,450,531],"entreprise.":[17,532],"Les":[18,64],"utilisateurs":[19],"m\u00e9tier":[20],"disposent":[21],"souvent":[22],"du":[23,292,320],"contexte":[24],"domaine":[26],"n\u00e9cessaire,":[27],"mais":[28],"ma\u00eetrisent":[29],"mal":[30],"le":[31,112,197,204,206,263,284,294,300,304,411],"SQL":[32,81,213,273,556,620,733],"et":[33,57,83,143,152,242,252,326,372,388,431,479,496,517],"les":[34,42,48,84,95,124,126,132,165,228,342,427,505],"sch\u00e9mas":[35,166],"bases":[37],"donn\u00e9es,":[39,416],"tandis":[40],"que":[41,131,155,159,258,283,397],"\u00e9quipes":[43],"data":[44,535,562,847],"centralis\u00e9es,":[45],"qui":[46,60,199,303,357,424],"traduisent":[47],"demandes":[49],"requ\u00eates,":[51],"fonctionnent":[52],"comme":[53],"une":[54,145,149,287,378,466,481],"ressource":[55],"partag\u00e9e":[56],"contrainte,":[58],"ce":[59,302],"introduit":[61],"d\u00e9lais.":[63],"syst\u00e8mes":[65,457,506],"Text-to-SQL":[66,117,458,507,577,681,877,906],"visent":[67],"\u00e0":[68,178,220,236,269,310,317,455,512],"r\u00e9duire":[69],"cette":[70],"friction":[71,583],"traduisant":[73],"questions":[75,588],"langage":[77,91],"naturel":[78],"(NL)":[79,587],"ex\u00e9cutable,":[82],"progr\u00e8s":[85],"r\u00e9cents":[86],"grands":[88],"mod\u00e8les":[89],"(LLM)":[92],"ont":[93],"rendu":[94],"premiers":[96],"d\u00e9ploiements":[97,309],"entreprise":[99],"plus":[101,103,127,295,408,483,515,521,526],"envisageables.":[104],"Malgr\u00e9":[105],"ces":[106,176],"avanc\u00e9es,":[107],"deux":[108,181],"d\u00e9fis":[109,177],"persistent:":[110],"(i)":[111,613],"co\u00fbt":[113,195],"\u00e9lev\u00e9":[114],"pipelines":[116,621],"fortement":[118],"d\u00e9pendants":[119],"LLM":[121,192,511,677],"pour":[122,196,201,226,410,503],"traiter":[123],"complexes,":[128],"alors":[129],"m\u00eame":[130],"charges":[133],"travail":[135],"r\u00e9elles":[136],"couvrent":[137],"large":[139,596],"\u00e9ventail":[140],"complexit\u00e9;":[142],"(ii)":[144,640],"\u00e9valuation":[146,482],"reposant":[147],"sur":[148,160,249,253,278,341,486,509],"m\u00e9trique":[150],"grossi\u00e8re":[151],"binaire,":[153],"telle":[154],"l'exactitude":[156,398],"d'ex\u00e9cution,":[157],"ainsi":[158,405],"benchmarks":[162,655],"publics":[163],"dont":[164],"ne":[167,401],"refl\u00e8tent":[168],"pas":[169],"ceux":[170,476],"entreprises.":[172],"Cette":[173],"th\u00e8se":[174],"traite":[175],"travers":[179],"contributions":[182],"compl\u00e9mentaires.":[183],"Premi\u00e8rement,":[184],"nous":[185,314,334,417],"introduisons":[186,418],"cadre":[188,490],"routage":[190,264,285],"sensible":[193],"au":[194,215],"Text-to-SQL,":[198],"s\u00e9lectionne,":[200],"chaque":[202],"requ\u00eate,":[203],"mod\u00e8le":[205,223,293],"moins":[207],"co\u00fbteux":[208,241],"susceptible":[209],"g\u00e9n\u00e9rer":[211],"exact,":[214],"lieu":[216],"recourir":[218],"syst\u00e9matiquement":[219,426],"unique":[222],"r\u00e9f\u00e9rence":[225,370],"toutes":[227],"NL.":[230],"Notre":[231,489],"routeur":[232],"est":[233,358],"l\u00e9ger,":[234],"construit":[235],"partir":[237],"signaux":[239],"peu":[240],"r\u00e8gles":[244],"simples":[247],"(bas\u00e9es":[248],"scores":[251],"classification),":[255],"sorte":[257],"surcharge":[260],"induite":[261],"reste":[265],"n\u00e9gligeable":[266],"rapport":[268],"l'\u00e9tape":[270],"g\u00e9n\u00e9ration":[272],"qu'il":[274],"optimise.":[275],"Des":[276],"exp\u00e9riences":[277],"benchmark":[280,742],"r\u00e9cent":[281],"montrent":[282],"conserve":[286],"exactitude":[288],"proche":[289],"celle":[291],"performant":[296],"tout":[297,449],"r\u00e9duisant":[299],"co\u00fbt,":[301],"rend":[305],"particuli\u00e8rement":[306],"adapt\u00e9":[307],"aux":[308],"fort":[311],"volume.":[312],"Deuxi\u00e8mement,":[313],"renfor\u00e7ons":[315],"l'\u00e9valuation":[316],"fois":[319,514],"point":[321],"vue":[323],"m\u00e9triques":[325,381,494],"d'\u00e9valuation.":[329],"Du":[330,413],"c\u00f4t\u00e9":[331,414],"m\u00e9triques,":[333],"proposons":[335],"mesures":[337],"d'ex\u00e9cution":[338,399],"fines":[339,495],"op\u00e9rant":[340],"tables":[343],"r\u00e9sultats":[345],"ex\u00e9cut\u00e9es:":[346],"l'Execution":[347,360],"Precision":[348,792],"(EXP)":[349,793],"quantifie":[350,363],"part":[352,365],"sortie":[355,368],"pr\u00e9dite":[356],"correcte,":[359],"Recall":[361,804],"(EXR)":[362,805],"r\u00e9cup\u00e9r\u00e9e,":[371],"leur":[373],"score":[374,818],"F1":[375,817],"fournit":[377],"synth\u00e8se.":[379],"Ces":[380],"mettent":[382],"\u00e9vidence":[384],"correction":[386],"partielle":[387],"distinguent":[389],"sur-pr\u00e9diction":[391],"sous-pr\u00e9diction":[394],"d'une":[395],"mani\u00e8re":[396,520],"binaire":[400],"permet":[402,480],"pas,":[403],"offrant":[404],"signal":[407,842],"diagnostique":[409],"d\u00e9bogage.":[412],"Textual":[420,851],"Query":[421,852],"Augmentation":[422,853],"(TQA),":[423,854],"d\u00e9-naturalise":[425],"identifiants":[428],"sch\u00e9ma":[430],"leurs":[432],"mentions":[433],"NL,":[435],"exemple":[437],"rempla\u00e7ant":[439],"noms":[441],"descriptifs":[442],"abr\u00e9viations":[445],"style":[447],"entreprise,":[448],"pr\u00e9servant":[451],"s\u00e9mantique.":[453],"Appliqu\u00e9e":[454],"l'\u00e9tat":[460],"l'art,":[462],"TQA":[464,498,879,899],"r\u00e9v\u00e8le":[465],"fragilit\u00e9":[467],"importante":[468],"sous":[469],"r\u00e9gimes":[471],"nommage":[473],"proches":[474,527],"entreprises":[478],"r\u00e9aliste,":[484],"centr\u00e9e":[485],"robustesse.":[488],"routage,":[492],"nos":[493],"constituent":[499],"approches":[501],"pratiques":[502],"rendre":[504],"fond\u00e9s":[508],"efficaces":[516],"\u00e9valu\u00e9s":[518],"fiable,":[522],"conditions":[525,916],"l'usage":[529],"r\u00e9el":[530],"ABSTRACT:":[533],"Exploratory":[534],"querying":[536],"is":[537,711,801,813],"an":[538],"operational":[539],"bottleneck":[540],"in":[541,595,831],"enterprise":[542,603,661,920],"decision-making.":[543],"Business":[544],"users":[545],"often":[546],"have":[547,600],"the":[548,614,624,688,732,752,771,777,798,810,846],"necessary":[549],"domain":[550],"context":[551],"but":[552],"lack":[553],"familiarity":[554],"with":[555,868],"and":[557,592,639,652,717,720,773,815,826,860,887,898,911],"database":[558],"technology,":[559],"while":[560,755,871],"centralized":[561],"teams":[563],"that":[564,682,725,744,750,785,833],"translate":[565],"requests":[566],"into":[567,589],"queries":[568],"operate":[569,786],"as":[570,649],"a":[571,633,644,675,685,701,740,839],"constrained":[572],"shared":[573],"resource,":[574],"introducing":[575],"delays.":[576],"systems":[578,907],"aim":[579],"to":[580,622,659,693,700,731,749,875,918],"reduce":[581],"this":[582,608],"by":[584],"translating":[585],"natural-language":[586],"executable":[590],"SQL,":[591,696],"recent":[593],"advances":[594],"language":[597],"models":[598],"(LLMs)":[599],"made":[601],"early":[602],"deployments":[604],"increasingly":[605],"feasible.":[606],"Despite":[607],"progress,":[609],"two":[610,669],"challenges":[611,667],"persist:":[612],"high":[615],"cost":[616],"of":[617,636,751,797,809],"LLM-heavy":[618],"Text-to-":[619],"tackle":[623],"most":[625],"complex":[626],"queries,":[627],"even":[628],"though":[629],"real":[630,919],"workloads":[631],"span":[632],"wide":[634],"range":[635],"query":[637],"complexities;":[638],"evaluation":[641,768],"relying":[642],"on":[643,653,684,739,787],"coarse,":[645],"binary":[646,834],"metric":[647,772,778],"such":[648],"execution":[650,783,835],"accuracy":[651,747,836],"public":[654],"whose":[656],"schemas":[657],"fail":[658],"reflect":[660],"ones.":[662],"This":[663],"thesis":[664],"addresses":[665],"these":[666],"through":[668],"complementary":[670],"contributions.":[671],"First,":[672],"we":[673,766,780,849],"introduce":[674,850],"cost-aware":[676],"routing":[678,726,745,894],"framework":[679],"for":[680,705,762,843,903],"selects,":[683],"per-query":[686],"basis,":[687],"least":[689],"expensive":[690],"model":[691,704,754],"expected":[692],"generate":[694],"accurate":[695],"rather":[697],"than":[698],"defaulting":[699],"single":[702],"strongest":[703,753],"all":[706],"NL":[707,862],"queries.":[708],"Our":[709,893],"router":[710],"lightweight,":[712],"built":[713],"from":[714,829],"inexpensive":[715],"signals":[716],"simple":[718],"score-based":[719],"classification-based":[721],"decision":[722],"rules,":[723],"so":[724],"overhead":[727],"remains":[728],"negligible":[729],"relative":[730],"generation":[734],"stage":[735],"it":[736,759],"optimizes.":[737],"Experiments":[738],"modern":[741],"demonstrate":[743],"preserves":[746],"close":[748],"reducing":[756],"cost,":[757],"making":[758,904],"especially":[760],"well-suited":[761],"high-volume":[763],"deployments.":[764],"Second,":[765],"strengthen":[767],"along":[769],"both":[770,908],"dataset":[774],"dimensions.":[775],"On":[776,845],"side,":[779,848],"propose":[781],"fine-grained":[782,896],"measures":[784],"executed":[788],"result":[789],"tables:":[790],"Execution":[791,803],"quantifies":[794,806],"how":[795,807],"much":[796,808],"predicted":[799],"output":[800,812],"correct,":[802],"ground-truth":[811],"recovered,":[814],"their":[816,861],"summarizes":[819],"both.":[820],"These":[821],"metrics":[822],"expose":[823],"partial":[824],"correctness":[825],"distinguish":[827],"over-":[828],"under-prediction":[830],"ways":[832],"cannot,":[837],"providing":[838],"more":[840,889,909,912],"diagnostic":[841],"debugging.":[844],"which":[855],"systematically":[856],"de-naturalizes":[857],"schema":[858],"identifiers":[859],"mentions,":[863],"e.g.,":[864],"replacing":[865],"descriptive":[866],"names":[867],"enterprise-style":[869],"abbreviations,":[870],"preserving":[872],"semantics.":[873],"Applied":[874],"state-of-the-art":[876],"systems,":[878],"reveals":[880],"substantial":[881],"brittleness":[882],"under":[883,915],"enterprise-like":[884],"naming":[885],"regimes":[886],"enables":[888],"realistic,":[890],"robustness-aware":[891],"evaluation.":[892],"framework,":[895],"metrics,":[897],"provide":[900],"practical":[901],"approaches":[902],"LLM-based":[905],"efficient":[910],"reliably":[913],"evaluated":[914],"closer":[917],"use.":[921]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-03-26T00:00:00"}