model extract deep dweb queri interfac discov topic structur databas increas complex enterpris databas preval lack document incur signific cost understand integr databas exist solut address mine kei foreign kei paid attent high level structur databas paper problem discov topic structur databas support semant brows larg scale data integr describ idisc discoveri system base multi strategi learn framework idisc exploit vari evid databas schema instanc valu construct multipl kind databas represent emploi set base cluster discov preliminari topic cluster tabl databas represent aggreg final cluster meta cluster improv accuraci extend idisc multipl level aggreg cluster boost techniqu introduc measur tabl import propos approach discov cluster repres facilit semant brows import featur framework highli extens addit databas represent base cluster easili incorpor theframework wehaveextensivelyevaluatedidisc larg real world databas result show discov topic structur high degre accuraci webiq learn web match deep dweb queri interfac integr deep web sourc requir highli accur semant match attribut sourc queri interfac match establish compar similar attribut label instanc attribut queri interfac data instanc pervas lack instanc reduc accuraci current match techniqu address problem describ webiq solut learn surfac web deep web automat discov instanc interfac attribut webiq extend question answer techniqu commonli ai commun purpos describ incorpor webiq current interfac match system extens experi realworld domain show util webiq result show acquir instanc improv match accuraci modest runtim overhead merg sourc queri interfac onweb databas merg interfac schema deep web cluster aggreg problem integr larg number interfac schema deep web scale problem divers sourc present challeng convent manual rule base approach schema integr address challeng propos formul schema integr optim problem object maxim satisfi constraint individu schema optim problem shown np complet develop approxim algorithm lmax build unifi schema recurs applic cluster aggreg extend lmax handl irregular frequent occur interfac schema extens evalu realworld data set show effect approach merg interfac schema deep web cluster aggreg problem integr larg number interfac schema deep web scale problem divers sourc present challeng convent manual rule base approach schema integr address challeng propos formul schema integr optim problem object maxim satisfi constraint individu schema optim problem shown np complet develop approxim algorithm lmax build unifi schema recurs applic cluster aggreg extend lmax handl irregular frequent occur interfac schema extens evalu realworld data set show effect approach bootstrap domain ontolog semant web servic sourc web site abstract vision semant web servic promis network interoper web servic sourc major challeng realiz vision lack autom mean acquir domain ontolog mark web servic paper propos deepmin system learn domain ontolog sourc web site set sourc domain interest deepmin learn base ontolog queri interfac grow current ontolog probe sourc discov addit concept instanc data page retriev sourc evalu deepmin real world domain preliminari result deepmin discov concept instanc high accuraci interact cluster dbase approach integr sourc queri interfac deep web increas number data sourc web content access queri interfac domain interest exist sourc vari coverag queri capabl import step integr sourc integr queri interfac specif focu crucial step integr accur match interfac integr queri interfac receiv attent recent current approach suffici gener model interfac flat schema map field interfac perform integr blackbox fashion process restart scratch wrong requir labori paramet tune paper propos interact cluster base approach match queri interfac hierarch natur interfac captur order tree vari type complex map field examin approach propos effect identifi map put human integr back loop propos approach interact learn paramet resolut uncertain map extens experi conduct result show approach highli effect uic trec web track year member databas inform system lab dbi univers illinoi chicago uic particip trec particip task web track topic distil name page find linkag inform document content inform document submit run statist method estim us text databas search desir data internet common wai internet singl search engin capabl search data internet approach interfac invok multipl search engin user queri potenti satisfi user number search engin interfac larg invok search engin queri cost effect creat unnecessari network traffic send queri larg number useless search engin search useless search engin wast local resourc problem overcom us search engin respect queri predict paper present statist method estim us search engin queri queri us search engin paper defin combin number document search engin suffici similar queri averag similar document experiment result estim method accur exist method effici effect metasearch text databas incorpor linkag document linkag document signific impact import document argu import document point document import document metasearch engin facilit ordinari user retriev inform multipl local sourc text databas larg scale metasearch engin content local databas repres repres user queri evalu set repres databas order determin databas search previou work linkag inform document util determin databas search paper inform emploi determin degre relev document respect queri inform store databas repres facilit select databas queri establish find similar document multipl text databas paper present methodolog find similar document multipl text databas queri posit integ methodolog consist step databas rank order document retriev databas order databas similar document queri rank ahead databas methodolog guarante retriev similar document queri statist method provid identifi databas estim similar document number strategi present retriev document identifi databas experiment result illustr rel perform strategi introduct internet vast inform sourc recent year consid effici effect metasearch larg number text databas metasearch engin facilit ordinari user retriev inform multipl local sourc text databas metasearch engin content local databas repres repres user queri evalu set repres databas order determin databas search number databas larg order ten thousand tradit metasearch engin ineffici queri evalu databas repres storag requir site metasearch engin larg paper propos hierarchi databas repres improv effici provid algorithm search hierarchi show retriev effect algorithm evalu user queri databas repres al estim us search engin paper present statist method estim us search engin queri estim metasearch engin choos local search engin invok queri us search engin paper defin combin number document search engin suffici similar queri averag similar document experiment result propos estim method accur introduct search engin creat internet ordinari user find desir data search engin databas defin set document search search engin index document databas creat store search engin speed queri process amount data internet huge believ end mil 