improv verbos queri subset distribut model reformul passag analysi build semant represent person inform structur annot search queri pseudo drelev feedback build semant represent person inform geometr represent multipl document evalu verbos queri process techniqu rank multipl document type desktop search unsupervis estim dirichlet smooth paramet learn rank queri reformul queri term rank base depend pars verbos queri learn concept import weight depend model model queri concept term depend shown signific posit effect retriev perform task web search relev high rank critic previou work treat concept equal import assumpt hold longer complex queri paper show effect exist term depend model natur extend assign weight concept demonstr weight depend model train exist learn rank techniqu small number train queri studi compar effect endogen collectionbas exogen base extern sourc featur determin concept import test weight depend model perform experi publicli trec corpora proprietari web corpu experiment result model consist significantli outperform standard bag word model unweight term depend model combin endogen exogen featur gener result retriev effect queri reformul anchor text queri reformul techniqu base queri log studi method captur user intent improv retriev effect evalu techniqu primarili focus proprietari queri log select sampl queri paper suggest anchor text readili effect substitut queri log studi effect rang queri reformul techniqu includ log base stem substitut expans standard trec collect result show logbas queri reformul techniqu effect standard collect expans safer form queri modif word substitut show anchor text simul queri log effect real log techniqu retriev experi pseudo ddesktop collect onlin commun search thread structur queri evolut show xml view combin xml queri languag provid surprisingli effect solut problem repres queri evolut databas evolut content evolut schema xml histori databas relat repres natur mean tempor group data model show schema tabl histori difficult repres relat databas easili repres xml queri xqueri probabilist retriev model semistructur data refin keyword queri xml retriev combin content structur regress rank learn meet opportun descript queri practition academia industri discuss latest develop aspect onlin advertis interact inform retriev technolog hope transform patent prior dart queri dstage queri segment inform retriev find text reus web queri structur expans dstage term depend japanes web retriev unsolv problem search approach blog site search resourc select statist view bin retriev model evalu text represent retriev group document cluster dbase resampl method pseudo drelev feedback retriev model question answer archiv discov kei concept verbos queri practition academia industri discuss latest develop aspect onlin advertis interact inform retriev technolog hope local text reus detect umass trec blog distil task inform dpattern dbase approach novelti detect hierarch languag model expert find enterpris corpora enterpris corpora evid employe work automat find expert topic present gener approach repres knowledg potenti expert mixtur languag model document retriev document expert gener probabilist techniqu weight retriev document expert specif posterior distribut model expert indirectli set document exploit underli structur complex languag featur experi show method excel perform trec expert search task effect collect combin evid expertis heterogen collect measur rank list robust queri perform predict introduct special issu databas inform retriev integr proxim dbase document represent name entiti retriev web search bu opportunist connect internet open wireless access point commonli municip area vehicular network opportunist connect internet second open access point paper adapt interact process web search retriev vehicular network intermitt internet access system call thedu mobil node internet proxi collect search engin result prefetch result page mobil node download pre fetch web page proxi contribut set techniqu make aggress select prefetch practic result significantli greater number relev web result return mobil user evalu scheme deploi thedu dieselnet vehicular testb buse oper micro urban area amherst ma simul workload find user expect time respons web search queri compar thedu latenc receiv relev respons queri minut deploi semi urban region sparser distribut ap compar big citi model term associ ad dhoc retriev perform languag model framework investig retriev perform manual dbuilt topic model model text topic popular research area machin learn inform retriev ir research focus automat method hand craft topic resourc onlin paper investig retriev perform topic model construct manual base hand craft directori resourc origin queri smooth manual select topic model view ideal user context model experi topic model trec retriev task show type topic model benefit perform good relev model automat queri modif model smooth queri topic model outperform relev model subset queri automat select model queri result relev model demonstr improv relev model automat built topic model base directori resourc latent concept expans markov random field queri expans form pseudo relev feedback relev feedback common techniqu improv retriev effect previou approach import issu role featur import model term depend paper propos robust queri expans techniqu base markov random field model inform retriev techniqu call latent concept expans mechan model term depend expans arbitrari featur model power framework simpl term occurr featur implicitli expans techniqu evalu techniqu relev model state art languag model queri expans techniqu model demonstr consist signific improv retriev effect trec data set describ techniqu gener meaning multi term concept task queri suggest reformul recommend citat academ paper abstract substanti effort wast scientif circl research rediscov idea publish literatur problem allevi avail recent academ work onlin kind text search system popular todai poor handl vocabulari mismatch research word relev document order find make serendipit result approach problem literatur search unpublish manuscript queri search system approach entir text content paper search process text previou literatur citat graph connect find relev relat materi evalu techniqu manual automat evalu method find order magnitud improv averag precis compar text similar baselin match resum job base relev model comparison sentenc retriev techniqu umass trec blog distil task identifi improv retriev procedur question linear featur dbase model inform retriev abstract number linear featur base model propos inform retriev commun recent model present differ share common underli framework paper explor discuss theoret issu framework includ paramet space detail supervis train algorithm directli maxim evalu metric consider averag precis present result show train model lead significantli test set perform compar train method directli maxim metric final show linear featur base model consist significantli outperform current state art retriev model correct choic featur bag word model implicit user prefer inform retriev paper report recent work field inform retriev attempt overli simplifi approach repres document queri bag word simpl model make difficult accur model user inform queri structur dstage term depend japanes languag improv novelti detect gener topic sentenc level inform pattern detect inform document stream import compon potenti applic work novelti detect approach base identif sentenc level inform pattern propos inform pattern concept novelti detect present emphasi inform pattern gener topic queri simpli turn specif question answer specif name entiti ne elabor analysi sentenc level inform pattern data trec novelti track includ sentenc length name entiti sentenc level opinion pattern analysi guidelin appli pattern novelti detect gener topic final unifi pattern base approach present novelti detect gener specif topic method deal gener topic focu experiment result show propos approach significantli improv perform novelti detect gener topic perform topic trec novelti track rank robust framework predict queri perform boost relev model perform queri term depend lda dbase document model ad dhoc retriev abstract semant smooth incorpor synonym sens inform languag model effect potenti signific improv retriev perform previous implement semant smooth model translat model shown good experiment result model unabl incorpor contextu inform overcom limit propos context sensit semant smooth method decompos document set weight context sensit topic signatur map topic signatur queri term languag model contextsensit semant smooth refer topic signatur languag model detail implement type topic signatur depend ontolog exist applic domain ontolog base concept multiword phrase map probabl topic signatur individu term estim em algorithm document model base topic signatur map deriv smooth method evalu thetrec genom track ontolog base concept trec ad hoc track disk multiword phrase experi show signific improv stage languag model languag model contextinsensit semant smooth index term inform retriev languag model semant smooth topic signatur concept multiword phrase repres document name entiti stori link detect sld hierarch languag model expert find enterpris corpora enterpris corpora evid employe work automat find expert topic present gener approach repres knowledg potenti expert mixtur languag model document retriev document expert gener probabilist techniqu weight retriev document expert specif posterior distribut model expert indirectli set document exploit underli structur complex languag featur experi show method excel perform trec expert search task effect collect combin evid expertis heterogen collect lda dbase document model ad dhoc retriev abstract semant smooth incorpor synonym sens inform languag model effect potenti signific improv retriev perform previous implement semant smooth model translat model shown good experiment result model unabl incorpor contextu inform overcom limit propos context sensit semant smooth method decompos document set weight context sensit topic signatur map topic signatur queri term languag model contextsensit semant smooth refer topic signatur languag model detail implement type topic signatur depend ontolog exist applic domain ontolog base concept multiword phrase map probabl topic signatur individu term estim em algorithm document model base topic signatur map deriv smooth method evalu thetrec genom track ontolog base concept trec ad hoc track disk multiword phrase experi show signific improv stage languag model languag model contextinsensit semant smooth index term inform retriev languag model semant smooth topic signatur concept multiword phrase framework predict qualiti answer dtextual featur type document collect develop web servic servic provid track textual featur click count paper present framework textual featur predict qualiti document show qualiti measur successfulli incorpor languag model base retriev model test approach collect question answer pair gather commun base question answer servic peopl answer question experiment result qualiti measur show signific improv baselin repres cluster retriev cluster base retriev cbr document group cluster retriev directli entireti queri smooth document languag model cluster typic repres simpl concaten indri trec notebook lesson learn terabyt track submiss classif approach document pair relev feedback inform explor textual textual document pair featur classifi unjudg document relev relev predict rank baselin document retriev featur includ citat measur url similar featur machin learn system document rank differ score assign baselin retriev system umass trec enterpris track tabl extract answer retriev precis predict base rank list coher document qualiti model web ad hoc retriev find expert commun dbase question danswer servic document qualiti model web ad hoc retriev similar measur track inform flow text similar span spectrum broad topic similar extrem document ident intermedi level similar result summar paraphras copi stronger form topic relev applic inform flow analysi question answer task paper explor mechan measur intermedi kind similar focus task identifi piec inform origin sentenc sentenc document document comparison incorpor algorithm recap prototyp inform flow analysi tool experiment result recap mechan propos exist method identifi intermedi form similar novelti detect base sentenc level pattern detect inform document stream import compon potenti applic paper novelti detect approach base identif sentenc level pattern propos user inform pattern sentenc combin queri word name entiti phrase import relev inform singl word propos novelti detect approach focus identif previous unseen queri relat pattern sentenc specif queri preprocess repres pattern includ queri word requir answer type pattern retriev sentenc determin answer present analysi pattern sentenc perform data trec novelti track experi novelti detect carri data trec novelti track experiment result show propos pattern base approach significantli outperform baselin term precis top rank find similar question larg question answer archiv recent signific increas number commun base question answer servic web peopl answer peopl question servic rapidli build larg archiv question answer archiv valuabl linguist resourc major task question answer servic find question archiv semant similar user question enabl high qualiti answer archiv retriev remov time lag commun base system paper discuss method question retriev base similar answer archiv estim probabl translat base retriev model show model find semant similar question word overlap translat model sentenc retriev work propos translat model monolingu sentenc retriev propos method construct parallel corpu method propos lexicon learn bilingu arab english corpu align sentenc level perform significantli improv result queri likelihood baselin demonstr smooth local context sentenc improv retriev queri likelihood baselin optim strategi complex queri paper propos set strategi process class path queri base vari design decis dimens dimens spatial overlap oper area link preprocess intermix path search process area satisfi queri constraint prefilt dynam select path search propos effici techniqu implement result integr queri process strategi util technolog spatial join tree spatial overlap search spatial cluster effici path search paper conduct extens experi propos strategi evalu perform base effici experiment studi produc interest observ find preprocess spatial queri advantag smaller buffer size low area select factor preprocess cost includ intermix path search spatial queri work larger buffer size high area select factor markov random field model term depend paper develop gener formal framework model term depend markov random field model arbitrari text featur incorpor evid make featur base occurr singl term order phrase unord phrase explor full independ sequenti depend full depend variant model approach develop train model directli maxim averag precis maxim likelihood train data ad hoc retriev experi present newswir web collect includ gov collect trec terabyt track result show signific improv model depend larger web collect find semant similar question base answer larg number question answer pair collect question answer board faq page web paper propos automat method find question mean method detect semant similar question word overlap calcul question question similar answer question develop similar measur base languag model compar tradit similar measur experiment result show semant similar question pair effect found propos similar measur recap system identifi inform flow indri trec terabyt track work detail experi carri indri search engin trec terabyt track result present task includ cienci ad hoc name page find cienci run focus queri optim techniqu ad hoc run import term proxim document qualiti name page find run investig document prior document structur indri trec terabyt track work detail experi carri indri search engin trec terabyt track result present task includ cienci ad hoc name page find cienci run focus queri optim techniqu ad hoc run import term proxim document qualiti name page find run investig document prior document structur umass robust mixtur relev model queri expans paper describ umass trec robust track experi focu approach term proxim pseudo relev feedback extern collect result approach highli ectiv lowel databas research dassess analysi statist question classif fact dbase question question classif system plai import role question answer system wide rang domain goal question classif accur assign label question base expect answer type approach past reli match question hand craft rule rule requir labori ort creat su er specif statist question classif method overcom issu emploi machin learn techniqu empir show statist approach robust achiev good perform divers data set hand tune examin role di erent syntact semant featur perform find semant featur tend increas perform pure syntact featur final analyz common misclassif error provid insight wai overcom comput dassist inform retriev recherch inform se applic riao multi dagent approach peer dto dpeer base inform retriev system framework select queri expans queri expans techniqu shown improv averag retriev perform techniqu oper system fact greatli degrad perform individu queri show comparison languag model unexpand expand retriev result predict expand retriev strai origin sens queri case unexpand result expand result remain case strai detect evalu method wide varieti trec collect languag model approach metadata cross ddatabas linkag search question answer perform tabl data question answer qa tabl data challeng inform retriev task paper describ qa system tabl creat machin learn heurist tabl extract method error analyz order improv system govern statist data appli improv type tabl data set show experiment result cluster dbase retriev languag model previou research cluster base retriev inconclus bring improv retriev effect document base retriev recent develop languag model approach ir motiv examin problem retriev framework propos model cluster base retriev evalu trec collect show cluster base retriev perform consist collect realist size signific improv document base retriev obtain fulli automat manner relev inform provid human formal multipl dbernoulli model languag model languag model assum document gener sampl multinomi distribut formal method estim exist multinomi case paper reexamin languag model base multipl bernoulli distribut assumpt explor past formal present smooth languag model estim assumpt formal bayesian standpoint evalu high accuraci retriev techniqu abstract inform retriev research concernedwith improv effect search applic inform analysi specif requir exist forhigh accuraci retriev mean achiev high precis top document rank paramount paper presentwork aim achiev high accuraci ad hoc document retriev incorpor approach question answer qa focu relev result high rank list argu tradit precis recal notappropri measur evalu task reciproc rank mrr relev result eval uat method modifi queri achiev high accuraci experi trec data provid support forth approach mrr incorpor qa techniqu high accuraci ad hoc retriev task categori subject descriptor inform storag retriev system soft ware perform evalu effici effect inform storag retriev inform search andretriev queri formul answer model question answer passag retriev answer pattern shown improv perform open domain factoid qa system requir construct pattern manual develop algorithm learn automat present simpler approach extend techniqu languag model creat answer model languag model train correct answer train question show fit natur probabilist model answer passag retriev demonstr effect trec qa corpu formal multipl dbernoulli model languag model languag model assum document gener sampl multinomi distribut formal method estim exist multinomi case paper reexamin languag model base multipl bernoulli distribut assumpt explor past formal present smooth languag model estim assumpt formal bayesian standpoint automat recognit read level user queri introduct gener interact web search engin character size fit represent user prefer search context task context obviou limit current search technolog person context identifi major research challeng number workshop work describ ongo effort automat infer knowledg user context user queri report experi focu recognit context featur read level show queri user level group effect separ scientif readablil studi decad readabl indic develop determin read level text popular includ flesch kincaid smog fog test statist approach propos recent common characterist method indri trec terabyt track paper overview experi carri trec terabyt track indri search engin indri effici effect distribut search engin inqueri base infer network framework support structur queri unlik inqueri languag model probabl network ad flexibl describ approach terabyt track involv automat construct structur queri titl portion trec topic method term proxim inform html document structur addit number optim procedur effici queri process explain umass trec novelti hard trec novelti track umass particip task find relev sentenc harder year continu show mark improv baselin call sentenc relev variant tfidf success approach achiev improv baselin locat sentenc primarili similar sentenc earlier sentenc focus name entiti high accuraci retriev document hard track investig clarif form fix variabl length passag retriev metadata clarif form result passag level feedback provid improv compar user suppli relat text document evalu outperform relat text passag evalu document retriev method queri expans compon show gain relat text found displai top passag feedback outperform displai centroid passag name entiti feedback result mix perform primari find passag retriev document retriev method perform passag retriev method passag evalu metric binari prefer charact clarif form improv passag retriev retriev method explor found benefit variabl length passag fix length passag corpu geographi genr metadata result signific retriev perform combin languag model infer network approach retriev paper describ develop structur document collect user gener text numer metadata explor exploit metadata inform retriev ir collect consist document extract youtub video page basketbal gener nba nation basketbal associ set topic relev judgement addit collect user profil relat nba collect baselin ir experi report effect video metadata retriev effect result surprisingli show search video titl perform significantli search addit metadata text field video tag descript thirteenth intern work shop research issu data engin multi dlingual inform manag time dbase languag model explor relationship time relev trec ad hoc queri type queri identifi favor recent document propos time base languag model approach retriev queri show time incorpor queri likelihood model relev model carri experi compar time base languag model heurist techniqu incorpor document recenc rank result show time base model perform heurist techniqu relev queri feedback statist languag model tradit relev feedback research explor relev document feedback queri represent updat base set relev document return user work investig relev queri feedback updat document represent base set relev queri propos statist model incorpor relev queri feedback tabl extract condit random field abil find tabl extract inform compon data mine question answer inform retriev task document tabl order commun dens pack multi dimension inform tabl emploi layout pattern effici field record dimension form languag model inform retriev systemat investig approach estim paramet languag model inform retriev call parsimoni languag model parsimoni languag model explicitli address relat level languag model typic smooth fewer paramet describ data appli parsimoni model stage retriev process index time search time feedback time experiment result show build model significantli smaller standard model perform standard approach salton award lectur inform retriev comput scienc evolv relationship tabl extract condit random field abil find tabl extract inform compon data mine question answer inform retriev task document tabl order commun dens pack multi dimension inform tabl emploi layout pattern effici field record dimension form gener hierarch summari web search hierarchi provid mean organ summar access inform describ method automat gener hierarchi small collect text appli techniqu summar document retriev search engin show hierarchi provid access document simpl rank list term hierarchi summari document top tf idf weight term addit discuss formal framework techniqu techniqu new databas trec collect lowel databas research assess challeng inform retriev languag model report workshop held center intellig inform retriev passag retriev base languag model previou research shown passag level evid bring ad benefit document retriev document long span subject area recent develop languag model approach ir provid effect altern tradit retriev model stream research motiv examin passag languag model framework paper report experi passag simpl languag model relev model compar result document base retriev result inqueri search engin base languag model approach comparison test data includ heterogen homogen document collect experi show passag retriev feasibl languag model context importantli provid reliabl perform retriev base full document featur document relev task fact dorient question describ result ongo project consid question type document featur relationship retriev techniqu examin document featur top document retriev question find list faq occur document judg relev task orient question judg relev fact orient question quasm system question answer semi dstructur data paper describ system question answer semistructur metadata quasm pronounc chasm question answer system aim improv search perform provid user specif answer user scan retriev document answer goal answer factual question exploit structur inher document found world wide web www base structur document index smaller unit metadata transform tabl cell smaller unit metadata import part task addit report work improv question classif languag model domain develop system document retriev crawl www fedstat gov cross dlingual relev model propos formal model cross languag inform retriev reli queri translat document translat approach leverag recent advanc languag model directli estim accur topic model target languag start queri sourc languag model integr popular techniqu disambigu queri expans unifi formal framework describ topic model estim parallel corpu dictionari test framework construct chines topic model english queri clir task trec model achiev perform strong mono lingual baselin term averag precis initi precis model outperform monolingu baselin main contribut work unifi formal model integr techniqu essenti ectiv cross languag retriev predict queri perform develop method predict queri perform comput rel entropi queri languag model collect languag model result clariti score measur coher languag usag document model gener queri suggest clariti score measur ambigu queri respect collect document show correl posit averag precis varieti trec test set clariti score identifi in ectiv queri averag relev inform develop algorithm automat set clariti score threshold predict poorli perform queri accept queri valid trec data compar automat threshold optimum threshold check frequent result good achiev sampl experi randomli assign queri class task orient question answer introduct inform retriev techniqu treat question democrat question grammar orient process rule manipul similar wai question data user process treat request request inform topic disregard import subtleti question knowledg question ask process fact develop categori inform impli question aid retriev focu question natur question carri inform techniqu improv retriev queri expans varieti queri expans disambigu techniqu studi distinguish fact orient taskorient distinguish categori inform explicit queri understand survei index search xml document sigir proceed annual intern acm sigir confer research develop inform retriev relev feedback person languag model perspect approach person involv learn short term long term user model user model provid context queri interact inform system paper discuss languag model repres context support context base techniqu relev feedback queri disambigu overview perspect person studi inform retriev time goal person improv effect inform access adapt individu user techniqu relev feedback filter consid support person consider research user model inform retriev research essenti goal current research person build model user interest prefer time filter system relev dbase languag model long recogn primari obstacl effect perform classic model estim relev model train data propos techniqu estim model queri demonstr techniqu produc highli accur relev model experi show relev model outperform baselin languag model system trec retriev main contribut work effect formal method estim relev model train data find topic word hierarch summar probabilist techniqu phrase extract workshop languag model inform retriev incorpor stem languag model framework work suggest notion stem increas number word occurr estim probabl word includ member stem class stem view type smooth probabl estim show view stem lead simpl incorpor idea corpu base stem present gener model stem gener term variant stem gener stem class member model evalu empir di erenc form stem combin approach ir invit talk inform retriev base statist languag model paper formal approach inform retriev base statist languag model introduc simpl reformul basic languag model approach introduc notion import queri term import queri term unknown paramet explicitli model queri term gener relev document import term unimport term languag model approach shown explain number practic fact todai inform retriev system explain current state inform retriev theori includ stop word mandatori term coordin level rank retriev phrase exploratori analysi phrase text retriev phrase commerci experiment search engin larg amount work area result remain mix clear phrase improv retriev effect paper examin phrase properti independ specif retriev approach explor phrase usag text corpora relev pattern relat phrase usag result understand phrase method phrase phrase techniqu evalu method directli determin phrase formul inform retriev introduct phrase inform retriev aid effect name multiword featur phrase term nomin compound implement retriev experi phrase inconclus experi phrase yield small improv inqueri trec improv effect inform retriev local context analysi techniqu automat queri expans extens studi inform retriev research mean address word mismatch queri document techniqu categor global local global techniqu reli analysi collect discov word relationship local techniqu emphas analysi top rank document retriev queri type techniqu advantag limit paper propos techniqu call local context analysi combin advantag global techniqu call phrasefind local techniqu local feedback experi number collect english english show local context analysi offer effect consist retriev result gener languag model inform retriev statist languag model successfulli speech recognit part speech tag syntact pars recent appli inform retriev paradigm document view languag sampl queri gener process retriev document rank base probabl produc queri languag model document paper present languag model inform retriev base rang data smooth techniqu includ good ture estim curv fit function model combin model conceptu simpl intuit easili extend incorpor probabl phrase word pair word tripl experi wall street journal trec data set show perform model compar inqueri languag model inform retriev word pair shown improv retriev perform deriv concept hierarchi text cluster dbase languag model distribut retriev effect retriev distribut environ import difficult problem lack effect appear collect select base word histogram heterogen collect relev document scatter collect search collect miss relev document exist collect select metric lack sound theoret justif tune problem propos approach distribut retriev base document cluster languag model document cluster organ collect topic languag model properli repres topic effect select topic queri base idea method propos suit environ show method improv effect distribut retriev gener languag model inform retriev poster abstract pic matric computation tractabl class probabilist queri oper paper present class queri oper reason explain call pic oper pic oper compris subclass gener class queri oper express infer network framework subclass chosen satisfi import design criteria oper belong subclass computation tractabl intuit plausibl candid model queri oper wide varieti queri oper defin pic oper work report princip motiv search effect boolean oper approach model boolean queri oper infer network goal gener classic strict proposit logic interpret queri oper object gener twofold hand graduat input boolean oper represent document term vector boolean characterist extend vector featur weight gener real valu oper output dichotom relev judgment replac document rank respons user queri notabl success pursuit report salton fox wu salton al salton al ground geometr metaphor vector space model defin gener class pnorm oper extend tradit oper natur experiment result achiev posit recent experiment shown system perform improv replac infer network boolean oper calcul inqueri retriev syst languag model approach inform retriev model document index document retriev extens studi integr class model goal research difficult problem argu reason lack adequ index model suggest index model solv problem feel make unwarr parametr assumpt lead retriev perform make prior assumpt similar document warrant propos approach retriev base probabilist languag model estim model document individu approach model parametr integr document index document retriev singl model advantag approach collect statist heurist retriev model integr part model resolv ambigu cross dlanguag retriev main hurdl improv clir effect resolv ambigu translat avail resourc problem present techniqu base occurr statist unlink corpora reduc ambigu phrasal term translat combin method techniqu reduc ambigu achiev monolingu effect final compar occurr method parallel corpu machin translat techniqu show good retriev effect achiev complex resourc corpu dbase stem cooccurr word cariant sort search user dinterfac framework text search current user interfac textual databas search leav desir individu confus group inconsist propos phase framework user interfac design framework common structur terminolog search preserv distinct featur individu collect search mechan paper appear cacm april pp earlier detail version appear tech report ir seri onlin lib magazin http www dlib org dlib januari content html materi base work support part nation scienc foundat librari congress depart commerc cooper agreement number eec nrad contract number nasa contract nag nsf contract iri opinion find conclus recommend express author reflect sponsor tintin system retriev text tabl tabl form import kind data element text retriev gist entir new articl exposit concis captur tabular form paper examin util exploit inform kei word digit document provid user flexibl power queri capabl specif exploit structur inform document identifi tabl compon field user queri base field empir result demonstr heurist method base tabl extract compon tag perform effect effici experi retriev tintin system strongli structur decomposit facilit represent user inform effect retriev tabl introduct current literatur inform retriev deal wor text segment topic investig problem text segment topic applic task includ topic track broadcast speech data topic identif full text databas research tackl similar problem goal studi focus data small segment size segment sentenc word common make problem challeng present method segment make queri expans techniqu find common featur topic segment experi techniqu show effect introduct recent increas interest track topic inform feed feed continu stream text produc speech recognit broadcast new text environ mark topic boundari delin sentenc text ters word repe probabilist retriev ocr degrad text dgram retriev ocr degrad text gram formul probabilist retriev system examin paper direct retriev document gram databas gram gram result improv retriev perform standard word base queri data level percent degrad wors achiev method gram identifi match match term queri expans perform standard queri method effect direct gram queri formul improv altern queri compon weight scheme measur term similar final web base retriev applic gram retriev ocr text displai queri term highlight sourc document imag introduct major problem retriev ocr text imag data inevitabl computation tractabl probabilist model boolean oper infer network model inform retriev probabilist interpret boolean queri oper prior work shown oper perform pnorm oper develop context vector space model design altern oper infer network framework contend issu comput tractabl defin flexibl class link matric natur candid implement boolean oper algorithm comput probabl involv link matric class present experiment result indic boolean oper implement term link matric class perform pnorm oper introduct present class oper model boolean queri ffl intuit plausibl ffl probabilist motiv ffl computation tractabl reason phrasal translat queri expans techniqu cross dlanguag inform retriev dictionari method cross languag inform retriev give perform mono lingual retriev failur translat multi term phrase shown factor respons error dictionari method studi import phrasal translat approach explor role phrase queri expans local context analysi local feedback show significantli reduc error automat dictionari translat introduct develop ir system languag english focus build mono lingual system increas avail line text languag english increas multi nation collabor motiv research cross languag inform retriev clir develop system perform retriev languag main approach clir translat machin inqueri battl trec dictionari method cross dlingual inform retriev multi lingual inform retriev ir larg limit develop system specif foreign languag explos avail electron media languag english make develop ir system cross languag boundari increasingli import paper present experi analyz factor affect dictionari base method cross lingual retriev present method dramat reduc error approach make introduct recent year amount onlin inform govern scientif busi commun risen dramat respons work develop system provid effect effici access electron media divers inform sourc explos growth internet worldwid compel evid ir system cross languag boundari increas int index handwrit word match histor manuscript written singl hand index exampl includ duboi collect univers massachusett earli presidenti librari librari congress standard techniqu index document scan convert machin readabl form ascii optic charact recognit ocr index text retriev engin ocr work handwrit altern scheme propos index text page document segment word imag word match creat equival class equival class multipl instanc word user ascii equival top equival class current paper deal match aspect process due variat singl person handwrit expect applic multilingu text retriev recent enorm increas network inform access line databas led databas languag english center intellig inform retriev ciir univers massachusett involv varieti industri govern digit librari applic multilingu text retriev inform retriev research evalu english databas queri advanc statist techniqu incorpor rank term weight perform languag describ experi rang project involv text retriev spanish japanes chines issu cover project includ document represent techniqu morpholog segment queri formul expans techniqu relev feedback comparison retriev effect english combin classifi text categor type classifi investig context text categor problem medic domain automat assign icd code dictat inpati discharg summari nearest neighbor relev feedback bayesian independ classif appli individu combin combin classifi produc result singl type classifi specif medic categor problem queri formul weight method nearest neighbor classifi improv perform introduct past research inform retriev shown improv retriev effect multipl represent index queri formul multipl search strategi work investig attain similar improv domain text categor combin represent classif queri expans local global document analysi automat queri expans long suggest techniqu deal fundament issu word mismatch inform retriev number approach expans studi recent attent focus techniqu analyz corpu discov word relationship global techniqu analyz document retriev initi queri local feedback paper compar effect approach show global analysi advantag local analysi gener effect show global analysi techniqu word context phrase structur local set document produc result effect predict simpl local feedback introduct problem word mismatch fundament inform retriev simpli state mean peopl word describ concept queri auth inqueri trec paper focus relev differ previous publish algorithm ad hoc experi uncertainti inform retriev system paper survei method repres reason imperfect inform open attempt classifi type imperfect pervad data discuss sourc imperfect classif framework work explicitli concern represent imperfect inform relat work imperfect inform basi reason work survei drawn field databas field artifici intellig area long concern problem caus imperfect inform paper stress relationship approach develop provid govern inform internet experi thoma thoma system design make legisl inform gener public internet regard prototyp govern digit librari joint project librari congress univers massachusett thoma signific opportun studi retriev interfac techniqu need support effect access complex inform internet base environ preliminari experi thoma signific interest type inform queri tend shorter studi retriev evalu trec techniqu queri process queri expans morpholog process incorpor improv keyword inform retriev internet queri process introduct mid decemb librari congress request newli elect leadership hous repres develo machin learn inform retriev abstract search distribut collect infer network inform retriev system network environ rais set issu receiv attent issu includ rank document collect relev queri select set collect rank list merg document rank return set collect paper describ method address issu infer network model discuss implement inqueri system present experiment result demonstr effect integr ir rdbm cooper index recent experi inqueri paper focus relev differ previous publish algorithm descript ad hoc experi nsf center intellig inform retriev recent explos network inform resourc attract attent problem autom method gather inform respons queri user literatur deal locat gather select respons queri multitud respons sourc paper deal model respons queri singl sourc inform complet respons queri necessit piec mutual relat partial respons dispar possibl heterogen sourc present system cooper retriev composit case case distribut agent multi agent system gestalt perspect good case deriv summat subcas agent partial view result local case match base local view local case assembl result case term global measur propos negoti driven case retriev algorithm approach dynam resolv inconsist case piec retriev process work report support part nsf center intellig inform retriev arpa contract offic naval research contract nation scienc foundat contract cda content inform necessarili reflect posit polici govern offici endors infer offic autom intellig workflow system effect text retriev base combin evid corpu user trec tipster experi inqueri inqueri probablist inform retriev system base bayesian infer network model paper describ recent improv system result particip tipster project trec confer improv includ transform form base specif inform complex structur queri automat queri expans automat recognit featur document relev feedback simul document rout experi gigabyt document collect inform process manag introduct effect inform retriev ir system depend represent match system repres inform repres document determin inform match document approach improv represent document text queri framework infer combin automat manual index represent probabilist retriev result research inform retriev suggest signific improv retriev effect obtain combin result multipl index represent queri formul search strategi infer net model retriev design point view treat inform retriev evidenti reason process multipl sourc evid document queri content combin estim relev probabl paper system base model studi retriev effect benefit combin type document queri inform found typic commerci databas inform servic result substanti real benefit lab report special section univers massachusett center intellig inform retriv proceed annual intern acm dsigir confer research develop inform retriev dublin support full dtext inform retriev persist object store full text inform retriev system unusu challeng data manag requir attempt made satisfi requir tradit relat databas manag system attempt produc discourag result inform retriev system typic custom data manag facil requir signific develop effort provid servic standard databas manag system advanc data manag system object orient databas manag system persist object store offer reason altern previou approach exist inform retriev system inqueri substitut persist object store mneme portion custom data manag system manag invert file index result improv perform signific opportun inform document retriev rout inqueri system eval word phrase sequenc word broken word subsequ crude oil price trend crude oil oil price price trend modif origin phrase extract procedur elimin number special process step appli nounphras extract process section weight section queri titl descript narr number term section accomplish simpli duplic titl time descript sum narr term phrasefind phrasefind techniqu corpu base queri expans phrasal concept identifi part speech tagger text context concept word close proxim integr text databas system possibl challeng document retriev rout inqueri system eval word phrase sequenc word broken word subsequ crude oil price trend crude oil oil price price trend modif origin phrase extract procedur elimin number special process step appli nounphras extract process section weight section queri titl descript narr number term section accomplish simpli duplic titl time descript sum narr term phrasefind phrasefind techniqu corpu base queri expans phrasal concept identifi part speech tagger text context concept word close proxim fast increment index full dtext inform retriev full text inform retriev system tradition design archiv environ provid support ad document exist document collect requir entir collect index modern applic inform filter oper dynam environ requir frequent addit document collect provid abil tradit invert file index built top persist object store data manag facil persist object store produc effici increment updat invert list describ system present experiment result show superior increment index competit queri process perform keyword full text document retriev increment index persist object store perform introduct full text inform retriev ir system establish tool satisfi user inf relev feedback infer network relev feedback modifi queri judgement relev highli rank document histor import method increas perform inform retriev system paper extend infer network model introduc turtl croft includ relev feedback techniqu differ relev feedback text abstract full text collect studi preliminari result relev feedback structur queri support infer net model report introduct relev feedback method inform retriev attempt improv perform queri modifi queri base user reaction initi retriev document specif user judgement relev relev document retriev add term queri reweight queri term document user comparison index techniqu japanes text retriev effect multipl queri represent inform retriev system perform independ gener boolean queri formul ten trec topic produc ten expert onlin searcher formul group group combin search trec test collect inqueri probabilist infer network retriev engin result show progress combin queri formul lead progress improv retriev perform result compar perform inqueri natur languag base queri combin issu recal perform measur larg databas rais overlap search conduct studi trec search smaller expect trec rout ad dhoc retriev evalu inqueri system hasiz retriev base combin evid text represent word phrase paragraph manual assign keyword version queri natur languag boolean combin consist probabilist framework type data fusion effect inform retriev context number year primari motiv develop infer net approach featur infer net approach abil captur complex structur network repres inform queri practic consequ complex boolean queri evalu easili natur languag queri produc rank output repres rule base concept base queri probabilist framework led concentr automat analysi queri techniqu enhanc guest editor introduct knowledg dbase statist approach text retriev retriev strategi hypertext inqueri retriev system larger heterogen text databas inform retriev research depend develop power effici flexibl retriev engin paper describ retriev system inqueri base probabilist retriev model support sophist index complex queri formul inqueri successfulli databas document introduct increas interest sophist inform retriev ir techniqu led number larg text databas research size databas term number document length document typic full text present signific challeng ir research experi thousand document abstract order carri research type text represent retriev model learni retriev complex object tradit inform retriev ir consid document atom unit paper show retriev compon document satisfi inform finer granular eas brows retriev result approach support multimedia network ir multimedia document compos object network combin collect compris document gain unifi view network databas multimedia document complex object retriev heterogen document corpu model appropri present probabilist retriev function initi estim probabilist paramet base logic structur document retriev process probabilist logic infer probabilist paramet retriev process repres probabilist datalog program execut hyspirit system pro loos dcoupl integr text retriev system object dorient databas system modern system develop environ deploi object manag facil call repositori store document creat maintain system develop pcte iso ecma standard public tool interfac open repositori paper present document retriev extens oqlori queri languag pcte extens propos cover pattern match term base document retriev automat gener document descript vector flexibl definit address document agiven queri integr facil case tool integr pattern match facil queri languag address author main contribut approach homogen integr term base document retriev flexibl definit document introduct repositori base applic wide spread experi larg document collect panel tipster panel univers massachusett tipster project lexic ambigu inform retriev lexic ambigu pervas problem natur languag process quantit inform extent problem impact inform retriev system report analysi lexic ambigu inform retriev test collect experi determin util word mean separ relev relev document experi show consider ambigu special databas word sens inform filter inform retriev side coin side coin comparison text retriev model univers massachusett tipster project search engin critic tool intellig analysi number innov search introduc research emphasi analyst began tipster project inqueri search engin introduc support specif complex queri probabilist infer network framework recent research languag model led develop indri search engin combin featur infer net languag model architectur design larg scale applic paper describ indri system show queri languag design support modern languag technolog present result demonstr indri effect effici summari databas research activ univers massachusett effici probabilist infer text retriev phrase structur queri inform retriev editori call amp calendareditori evalu infer network dbase retriev model infer network support document retriev introduc network base retriev model compar convent probabilist boolean model perform retriev system base infer network model evalu compar perform convent retriev model hypertext inform retriev fundament concept panel infer network document retriev infer network support document retriev introduc network basead retriev model compar convent probabili tic boolean model experi queri acquisit document retriev system im recent experiment document retriev system emphasi acquisit detail model inform interact user argu enhanc queri combin relev feedback improv retriev perform paper describ studi aim evalu easili enhanc queri acquir user effect addit knowledg retriev result signific effect benefit obtain acquisit domain concept relat queri concept level import inform term cluster syntact phrase term cluster syntact phrase format method transform natur languag text mix success strategi improv qualiti text represent document retriev strength method complementari explor combin produc superior represent paper discuss implement syntact phrase gener preliminari experi produc phrase cluster experi show small improv retriev effect result phrase cluster clear corpora larger standard inform retriev test collect requir evalu techniqu visual interact end user goal base system intellig inform retriev overview ir research mass interpret nomin compound inform retriev chapter applic text mine techniqu sociolog present chapter techniqu text mine base extract inform document titl chapter hierarch dictionari text mine task discuss interact retriev complex document servic journal chemic inform comput scienc bibliographi mackinlai jock automat design graphic present stanford univers dissert technic report stan cs manber udi amp sun wu glimps tool search entir file system proceed winter usenix confer san francisco ca mann william amp sandra thompson rhetor structur theori theori text organ technic report isi rs isi man christoph automat acquisit larg subcategor dictionari corpora proceed st annual meet associ comput lingusit columbu marchionini gari peter liebscher amp xia lin author hyperdocu design interact interfac inform retriev onlin system ed martin dillon york ny greenwood press markei chairman messag retriev model incorpor hypertext link inform cope problem theori typolog approach compet time analog architectur depend percept devot theoret approach physic properti architectur case architect solut sourc inspir design process present model represent architectur design case face problem formal ill structur domain knowledg consist theoret base investig relationship architectur design inform languag repres inform theoret model languag game abstract lga present mean studi relationship base model propos inform structur support intellig reason retriev case materi design goal definit similar desi word sens disambigu machin dreadabl dictionari paper describ mean automat build larg neural network vlnn definit text machin readabl dictionari demonstrato network word sens disambigu method bring earlier independ approach word sens disambigu machin readabl dictionari spread activ model automat construct vlnn enabl real size experi neural network natur languageprocess turn insight behavior design lead improv retriev document plausibl infer experiment studi support brows intellig text retriev system text relev signatur proceed relat human knowledg task requir plan librari chairman messag retriev document plausibl infer preliminari studi implement rank strategi text signatur chairman messag reason except plan execut monitor approach natur languag process document retriev document retriev system restrict natur task techniqu larg number document broad domain effect techniqu develop base statist word occurr text paper describ approach natur languag process nlp techniqu essenti natur languag problem comparison request text text document titl abstract propos nlp techniqu develop request model base conceptu case frame compar model text candid document request model provid inform statist search techniqu identifi candid document part preliminari evalu approach case frame represent set request cacm collect construct statist search carri depend rel import support offic document architectur constrain type approach intellig inform retriev built hybrid case base reason cbr inform retriev ir system gener queri ir system inform deriv cbr analysi problem situat queri automat form submit text form set highli relev case base cbr analysi modifi version inqueri relev feedback modul approach extend reach cbr retriev purpos larger corpora inject knowledg base techniqu tradit ir introduct strength case base reason cbr system abil reason problem case perform highli intellig problem solv gener legal argument detail oper plan cbr system core abil retriev highli relev case cbr system limit avail case repres case base current case base reason syst approach intellig inform retriev built hybrid case base reason cbr inform retriev ir system gener queri ir system inform deriv cbr analysi problem situat queri automat form submit text form set highli relev case base cbr analysi modifi version inqueri relev feedback modul approach extend reach cbr retriev purpos larger corpora inject knowledg base techniqu tradit ir introduct strength case base reason cbr system abil reason problem case perform highli intellig problem solv gener legal argument detail oper plan cbr system core abil retriev highli relev case cbr system limit avail case repres case base current case base reason syst distribut expert dbase inform system interdisciplinari approach user dspecifi domain knowledg document retriev expert assist document retriev system knowledg dbase approach data manag intellig user interfac task manag intellig interfac decemb vol quarterli bulletin ieee comput societi comparison network structur databas system document retriev adapt mechan select search strategi document retriev system task support offic system role context adapt user interfac network organ document retriev network organ implement document retriev system propos organ signific advantag term rang search compar invert cluster file organ algorithm gener maintain network experi design test effici effect introduct physic group document main type file organlzd tha retriev queri nave propos implement paper investig network document retrlev system rare invert file organizatlon combin lll advantag uster file salt invert cluster file organ applic inform retriev techniqu offic implement document retriev system network organ implement document retriev system propos organ signific advantag term rang search compar invert cluster file organ algorithm gener maintain network experi design test effici effect introduct physic group document main type file organlzd tha retriev queri nave propos implement paper investig network document retrlev system rare invert file organizatlon combin lll advantag uster file salt invert cluster file organ experi automat text file retriev offic environ incorpor search model document retriev system model cluster search base classif implement model document retriev file organ cluster dbase retriev evalu goffman indirect retriev method document cluster evalu experi cranfield collect 