select diversifi web search result exploit queri reformul web search result diversif web user underli inform initi queri effect approach diversifi result retriev queri paper introduc probabilist framework web search result diversif explicitli account aspect underspecifi queri diversifi document rank estim document satisfi uncov aspect extent aspect satisfi rank evalu framework context divers task trec web track exploit queri reformul provid major web search engin wse mean uncov queri aspect result attest effect framework compar state art diversif approach literatur addition simul upper bound queri reformul mechan offici trec data draw insight effect queri reformul gener wse promot divers find term retriev explor semant concept astronom vocabulari blog track research trec trec blog track aim explor inform seek behaviour blogospher build reusabl test collect blog relat search task advent trec blog track led research grow field encapsul cross pollin natur languag process research paper recap task address trec blog track cover period describ corpora task address track result publish research influenc document rank expert search retriev effect underli document search compon expert search engin import impact effect gener expert search result larg scale studi perform experi context document search expert search task trec enterpris track measur influenc perform document rank rank candid expert show real simul document rank expert search system perform relat relev retriev document surprisingli case increas document search effect increas expert search perform studi select collect enrich enterpris search find good feedback document integr proxim subject sentenc blog opinion retriev select applic queri dindepend featur web inform retriev studi queri expans effect rdb rdf tool feasibili expos larg scienc archiv data integr brows recommend base intertempor choic model predict us collect enrich enterpris search fit score distribut blog opinion retriev spam issu opinion blog post search perfect document rank expert search singl dpass index mapreduc invers document frequenc invers document frequenc popular measur word import idf invari appear host heurist measur inform retriev idf heurist paper show idf optim principl sens show idf optim weight word respect minim kullback leibler distanc suitabl gener nonneg function probabl distribut optim problem close relat maximum entropi problem show idf optim weight associ at word featur inform retriev set treat document queri retriev idf optim doc ument retriev invers document frequenc invers document frequenc popular measur word import idf invari appear host heurist measur inform retriev idf heurist paper show idf optim principl sens show idf optim weight word respect minim kullback leibler distanc suitabl gener nonneg function probabl distribut optim problem close relat maximum entropi problem show idf optim weight associ at word featur inform retriev set treat document queri retriev idf optim doc ument retriev web inform retriev model abstract research intend boost relev web search result classifi websnippet social construct hierarch search concept comprehens human edit knowledg structur open directori project odp semant aspect search concept categori social construct hierarch knowledg repositori extract textual inform contribut societi textual inform explor analyz construct categori document set subsequ emploi repres semant social construct search concept simpl api xml sax compon jaxp java api xml process util read analyz rdf format odp data file structur rdf content rdf knn train construct categori document set categor web search result categor web search result ontolog filter base interact web inform seeker initi experiment result demonstr propos approach improv precis search expertis experi vote model search expertis experi vote model advanc inform retriev paper captiv reader present overview topic detect track tdt problem purpos organ broadcast new stori real world event discuss paper describ research problem consid phase tdt focus solut develop ciir problem detect problem identifi topic appear new stream approach discuss practic technic interest present algorithm modif exist ir techniqu tdt problem find data resourc virtual observatori sko vocabulari kei blog distil rank aggreg effect statist approach blog post opinion retriev high qualiti expertis evid expert search expert search evalu support document analysi link graph compress techniqu trec blog track rise blog grassroot publish medium interest peculiar characteris blog compar genr document open interest research area inform retriev field blog track introduc part renown text retriev confer trec evalu forum drive research blogospher facilit experiment evalu blog search techniqu paper report year blog track trec describ blog search task investig trec discuss main lesson learnt track conclud paper discuss broader implic blog track lesson direct futur aim uncov explor rich inform blogospher retriev sensit train measur rank opinion blog post opinionfind limit opinion dfind baselin system automat document prior featur select web retriev univers glasgow trec experi blog vote techniqu expert search trec particip task blog enterpris track continu experi terrier modular scalabl inform retriev ir platform diverg random dfr framework blog track opinion find task propos statist term weight approach identifi opinion document altern approach base opinion identif tool utilis improv opinion baselin observ appli statist term weight approach expert search task enterpris track investig proxim queri term candid occurr document syntact dbase queri reformul techniqu inform retriev vote techniqu expert search trec particip task blog enterpris track continu experi terrier modular scalabl inform retriev ir platform diverg random dfr framework blog track opinion find task propos statist term weight approach identifi opinion document altern approach base opinion identif tool utilis improv opinion baselin observ appli statist term weight approach expert search task enterpris track investig proxim queri term candid occurr document european confer inform retriev ecir paramet sensit probabilist model ad dhoc retriev expertis drift queri expans expert search perform comparison cluster replic inform retriev system multinomi random model retriev document field abstract document field titl head document offer structur document retriev propos approach literatur emploi linear combin score assign field linear combin frequenc term frequenc normalis compon context diverg random framework sound opportun integr document field probabilist random model paper introduc probabilist model incorpor field retriev process multinomi random model inform theoret approxim evalu result experi conduct standard trec web test collect show propos model perform state art field base weight model time theoret found extens current field base model relev feedback expert search abstract enterpris set expert search consid import task search task user expertis instanc requir assist topic interest expert search system assist user expertis suggest peopl relev expertis topic interest work appli expert search approach explicitli rank candid respons queri implicitli rank candid take account rank document respect queri topic pseudo relev feedback aka queri expans shown improv retriev perform adhoc search task work investig extent queri expans appli expert search task improv accuraci gener rank candid defin approach queri expans base initi rank document queri topic approach base final rank candid aim paper fold firstli determin queri expans successfulli appli expert search task ascertain form queri expans provid robust improv retriev perform perform evalu contrast queri expans approach context trec enterpris track set dfield normalis hyper dparamet name dpage find search task combin document prior web inform retriev queri independ featur call document prior number incom link document page rank type url successfulli integr web inform retriev system order enhanc retriev effect combin document prior enhanc retriev perform current combin prior approach base heurist ignor depend document prior paper present robust method condition combin document prior principl approach adjust distribut document prior sourc evid distribut document prior sourc evid investig retriev perform attain combin prior method comparison singl prior heurist combin document prior method assum document prior independ investig sensit propos method train data standard web test collect includ larg scale gov test collect find document prior experi consider high correl suggest depend document prior account extens experi larg scale collect observ propos condit combin method effect robust light syntact dbase index prune inform retriev abstract index prune techniqu elimin term index basi contribut term content document present syntact base index prune techniqu exclus shallow syntact evid decid term prune type evid document independ base assumpt gener collect document exist approxim proport relat frequenc content block part speech po block po block fix length sequenc noun verb part speech extract corpu remov index term correspond low frequenc po block strategi low frequenc po block correspond sequenc content poor word ii low frequenc po block content bear part speech preposit correspond sequenc contentpoor word experi trec test collect statist weight model full indic baselin show syntact base index prune enhanc retriev perform term averag earli precis light prune level reduc size index low cost techniqu perform similarli relat work document specif inform gener combin document prior web inform retriev queri independ featur call document prior number incom link document page rank type url successfulli integr web inform retriev system order enhanc retriev effect combin document prior enhanc retriev perform current combin prior approach base heurist ignor depend document prior paper present robust method condition combin document prior principl approach adjust distribut document prior sourc evid distribut document prior sourc evid investig retriev perform attain combin prior method comparison singl prior heurist combin document prior method assum document prior independ investig sensit propos method train data standard web test collect includ larg scale gov test collect find document prior experi consider high correl suggest depend document prior account extens experi larg scale collect observ propos condit combin method effect robust incorpor term depend dfr framework term depend occurr studi languag model instanc metzler amp croft show retriev perform significantli enhanc term depend inform work show term depend model diverg random dfr framework evalu term depend model adhoc retriev task trec gov terabyt collect examin effect vari term depend window size retriev perform propos model experi show term depend successfulli incorpor dfr framework extend weight model term qualiti measur abstract weight model lexic statist term frequenc deriv term weight estim relev document queri remov stopword consider qualiti word weight assum term frequenc good indic decis made relev document queri intuit raw term frequenc enhanc discrimin term propos lexic featur predict qualiti word weight retriev specif show part speech noun verb estim inform word gener relev queri document experiment result standard trec collect show integr propos term qualiti establish weight model enhanc retriev perform baselin origin weight model time overview trec blog track univers glasgow trec experi blog enterpris track terrier trec particip task blog enterpris track continu experi terrier modular scalabl inform retriev ir platform diverg random dfr framework blog track opinion find task propos statist term weight approach identifi opinion document altern approach base opinion identif tool utilis improv opinion baselin observ appli statist term weight approach expert search task enterpris track investig proxim queri term candid occurr document perform analysi distribut inform retriev architectur improv network simul model abstract studi present analysi interconnect network distribut inform retriev ir system simul switch network versu share access network result show switch network improv perform replic system switch network prevent satur network larg number queri server combin field queri expans adapt queri expans examin content load part speech block inform retriev vote candid adapt data fusion techniqu expert search task combin field ditem email search email exampl structur document field field exploit enhanc retriev effect inform retriev ir system search mail list archiv recent experi trec enterpris track field appli vari degre success particip work field base weight model investig retriev perform attain field examin field evid combin search expertis terrier platform overview trec blog track rise internet blog creation journal web page log creat highli dynam subset world wide web evolv respond real world event blog weblog recent emerg grassroot publish medium call blogospher collect blog univers glasgow trec experi terabyt enterpris track terrier trec particip task terabyt enterpris track continu experi terrier modular scalabl inform retriev ir platform further research diverg random dfr framework weight model introduc effect low cost model combin evid document structur captur term depend proxim addition terabyt track improv queri expans mechan field present trec refin techniqu combin evid linear uniform introduc low cost syntacticallybas nois reduct techniqu flexibl appli queri index name page find task present techniqu combin queri independ evid form prior probabl enterpris track test vote model expert search experi focu candid length normalis retriev perform enhanc appli retriev techniqu underli rank document queri reformul automat gener queri concept document space decis mechan select combin evid topic distil combin evid increas retriev effect paper investig effect decis mechan select combin evid web inform retriev topic distil introduc measur queri broad select combin evid queri result experi show statist signific associ output decis mechan rel effect combin evid show propos methodolog appli oper set relev inform set decis mechan threshold automat queri perform predict present framework queri perform predict task estim effect search perform respons queri lack relev judgment approach base statist decis theori estim util document rank respect inform express queri address uncertainti infer inform estim util expect similar rank induc relev model impact relev model base presum repres inform specif queri perform predictor instanti framework substanti outperform state art predictor trec corpora appli light natur languag process ad dhoc cross languag inform retriev univers glasgow webclef experi dfield normalis languag specif stem term frequenc normalis tune bm dfr model abstract term frequenc normalis paramet tune crucial issu inform retriev ir import impact retriev perform classic pivot normalis approach suffer collect depend problem consequ requir relev assess collect obtain optim paramet set paper tackl collect depend problem propos tune method measur normalis effect propos method refin extend methodolog experi evalu propos tune method trec collect normalis diverg random dfr model bm normalis method result show normalis method tune method significantli outperform robust empir obtain baselin divers trec collect margin comput cost terrier inform retriev platform increas interest improv robust ir system effect difficult queri system robust achiev high averag precis map entir set topic signific map worst topic map fact queri expans qe increas global map hurt perform worst topic select applic qe natur answer obtain robust retriev system network analysi distribut inform retriev architectur abstract studi present analysi interconnect network distribut inform retriev ir system simul switch network versu share access network result show switch network improv perform replic system switch network prevent satur network larg number queri server studi dirichlet prior term frequenc normalis appli smooth techniqu languag model approach paper appli dirichlet prior term frequenc normalis classic bm probabilist model diverg random pl model contribut paper twofold extens experi trec collect show newli gener model dirichlet prior normalis appli provid robust effect perform propos theoreticallydriven approach automat paramet tune dirichlet prior normalis experi show tune approach optimis retriev perform newli gener dirichlet prior base weight model univers glasgow trec experi terabyt enterpris track terrier particip trec continu experi terrier modular scalabl inform retriev ir framework task terabyt enterpris track terabyt track investig diverg random weight model queri expans approach account document field content titl anchor text addit test select queri expans mechan determin appropri queri expans queri basi statist inform low cost queri perform predictor enterpris track investig combin document field evid inform occur enterpris set email item task investig tempor thread prior suitabl email search expert search task candid gener profil expertis evid collect propos model rank candid profil respons queri case studi distribut inform retriev architectur index terabyt text dempster dshafer theori queri dbias combin evid web abstract paper report larg scale experi evalu formal queri bias combin evid mechan dempster shafer theori evid combin optim result obtain content link analys web queri bias mechan base queri scope measur queri specif queri scope defin probabilist propag mechan top hierarch structur concept provid wordnet standard web test collect link analysi approach result show propos approach improv retriev effect keyword combin content link analysi queri bias combin evid dempster shafer theori evid web inform retriev queri scope static absorb model web work paper review di erent idea link base analysi search result static rank web page base call randomsurf model review present unifi framework topic base hub author model discret compon method variant ica pca develop illustr page english languag wikipedia collect propos present commun link rank consortium extract web intellig paper opportun open sourc search acm sigir workshop mathemat fformal method inform retriev mf fir gaug student understand interact lectur univers glasgow clef french monolingu inform retriev terrier perform analysi distribut architectur index terabyt text abstract simul architectur distribut inform retriev system larg web collect order work optim set set resourc analys effect distribut replic cluster architectur variabl number workstat collect approxim million document terabyt text test perform architectur show pure distribut architectur broker bottleneck due high number local answer set sort replic system network bottleneck due high number queri server continu data interchang broker final demonstr cluster system outperform replic system larg number queri server due reduct network load construct queri concept document space base data mine techniqu select combin evid topic distil document aggreg dlevel inform combin evid inform retriev studi extens order increas effect paper studi select applic retriev approach queri basi web inform retriev methodolog base assumpt queri benefit uniform retriev approach order select retriev approach queri evid hyperlink structur retriev document distribut aggreg group document domain experiment result show obtain import improv retriev effect simpl statist decis mechan queri basi queri dbase pre dretriev model select approach inform retriev paper propos queri base pre retriev approach model select problem automat select perform retriev model retriev process take place approach queri cluster statist bestperform retriev model cluster queri assign closest cluster queri appli model cluster evalu model select approach disk amp trec collect result show model select approach achiev stabl perform outperform optim retriev model indi erent queri result show interestingli retriev model consist perform queri belong cluster us hyperlink structur queri dbias topic distil paper introduc inform theoret method estim us hyperlink structur induc set retriev document evalu effect method context optim bayesian decis mechan select retriev approach perqueri basi trec task estim hyperlink structur us stabl weight scheme emploi sampl document reduc comput overhead evalu effect hyperlink structur us realist set set threshold decis mechan automat result show improv baselin obtain discoveri aggreg usag profil base cluster inform univers glasgow trec experi web particip trec test terrier modular scalabl inform retriev framework track mix queri task web track emploi decis mechan select retriev approach queri basi robust track order cope poorlyperform queri pre retriev perform predictor weight function recommend mechan test train approach automat tune term frequenc normalis paramet terabyt track emploi distribut version terrier test effect techniqu anchor text pseudo queri expans select weight model queri infer queri perform pre dretriev predictor abstract predict queri perform interest import issu inform retriev ir current predictor involv relev score time consum comput current predictor suitabl practic applic paper studi set predictor queri perform gener prior retriev process linear parametr correl predictor queri perform assess trec disk disk minu cr collect result propos predictor signific correl queri perform show predictor infer queri perform practic applic univers glasgow trec experi web particip trec test terrier modular scalabl inform retriev framework track mix queri task web track emploi decis mechan select retriev approach queri basi robust track order cope poorlyperform queri pre retriev perform predictor weight function recommend mechan test train approach automat tune term frequenc normalis paramet terabyt track emploi distribut version terrier test effect techniqu anchor text pseudo queri expans select weight model queri distribut relev document domain dlevel aggreg topic distil paper studi distribut relev document aggreg form group retriev document domain aggreg account size measur correl incom outgo hyperlink report preliminari experi trec topic distil task find larger aggreg aggreg correl hyperlink relev document result show distribut domain level aggreg potenti find relev document studi paramet tune term frequenc normal weight scheme star dgraph inform theoret queri context detect web inform retriev util dorient hyperlink analysi model web univers glasgow robust track queri dbase model select approach poorli dperform queri univers glasgow web track dynam applic hyperlink analysi queri scope year particip web track aim combin dynam evid content hyperlink analysi end introduc decis mechan base call queri scope concept topic distil task find anchor text increas precis significantli contentonli retriev result contrast trec find queri scope show select applic hyperlink analysi url base score effect gener queri improv precis fact effect run decis mechan outperform significantli content anchor text retriev item task emploi queri scope order distinguish name page queri home page queri obtain result close content anchor text baselin report acm sigir workshop semant web swir build bilingu dictionari parallel web document paper describ system automat construct bilingu dictionari cross languag inform retriev applic describ automat target candid parallel document filter candid document process creat parallel sentenc parallel sentenc automat translat adapt emim techniqu dictionari translat term creat evalu dictionari human expert evalu show system perform addit result obtain automat creat corpora compar obtain manual creat corpora parallel document compar techniqu approach advantag simpl uniform easi implement provid encourag result univers glasgow web track trec aim particip topic distil name page find task web track evalu found modular probabilist framework web inform retriev integr content link analys link analysi compon framework emploi probabilist approach call absorb model calcul measur popular document induc web graph hierarch cluster summaris approach web retriev glasgow trec interact track introduct motiv current search engin ax typifi lack precis coupl elong rank list style result present combin factor make relev data extract increasingli complex main investig particip interact track trec assess effect visualis techniqu displai result search engin current system provision name huddlesearch newli develop cluster algorithm dynam organis relev document travers hierarchi gener specif cluster categori extend trec summaris tool summaris multipl document summari paint caricatur content cluster individu document allow user provision judg cluster relev prior view content interact evalu summaris technolog task orient approach paper present task orient approach evalu automat text summaris system evalu system tradition troublesom area summaris research propos scheme evalu exist system determin rel effect interact search task condit approxim intend system user dcenter imag retriev system imag rank task involv paramet depend intrins characterist imag index languag repres semant content develop weight model combin paramet gener scheme find balanc paramet straightforward paramet combin lead rank accept user paper choos set test queri present impact paramet rank imag combin discuss combin evalu follow user orient approach compar rank provid paramet combin rank human judgment step user center imag retriev system dynam adapt user profil prefer introduct imag constitut complex type medi rank digit imag combin evid conceptu graph order logic paper conceptu graph consist convivi graphic represent logic flexibl weight scheme multimedia document effect effici relat queri process conceptu graph recent paper ouni amp huiber work logic relat framework inform retriev present emphas import relat accur index provid relat treatment effect retriev knowledg represent formal support relat index major fulli relat treatment conceptu graph offer rich need relat index complex highli structur document provid relat base process paper follow relat framework case conceptu graph account relat properti retriev process approach lead sound extens preserv semant formal implement import goal obtain workabl system model index retriev imag conceptu graph deal complex imag part index process keyword suffici obtain index faith represent imag content propos conceptu graph formal index languag keyword relat obtain index precis retriev effect improv paper present system comput assist imag index process perform formal imag model result index process set conceptu graph organ improv retriev execut time imag retriev system call relief implement object orient dbm web ensur manag imag test collect good result respect execut time qualiti answer introduct precis orient promis retriev algorithm system base conceptu graph formal relief combin express rapid singl system paper constitut propos effici effect logic inform retriev system relat index approach opinion necess cope emerg applic base multimedia conceptu graph formal index languag choic relat index support captur properti logic inform retriev model workabl system order logic standard inform retriev techniqu combin effect obtain express system accur handl complex document improv retriev effect achiev good time perform experiment imag test collect system web provid illustr role logic futur develop inform retriev system introduct emerg applic base fast precis dorient imag retriev system logic relat approach inform retriev index relat index approach farradan work inform carri fix set relationship type underli set term idea essenc mean inform encapsul relationship term import relationship wide recogn field relat databas knowledg represent formal field substanti improv understand relationship problem involv formal relationship correctli repres formal field exploit object concret oper inform retriev previou attempt manag relationship address structur aspect exclud manipul index express relat oper paper suggest prime relat properti extend invert file approach inform retriev conceptu graph multifacet logic model inform retriev van rijsbergen promot principl match queri document found form uncertain logic infer formul deliber abstract attempt made principl implement effect logic semant classic logic infer logic implic strict match process overcom problem world semant good formal uncertain infer need inform retriev system nie initi work basic assumpt consider document world refin approach adopt complex inform consid base notion facet document point view conceptu graph about paper present inform retriev model base conceptu graph name elen elen conceptu graph repres inform inher document queri graph oper determin graph generalis graph case inform carri graph inform carri generalis graph articl compar conceptu graph about about deriv inform retriev model about deriv map situat logic framework framework axiom drive retriev process filter retriev mechan compar approach axiom govern method show elen close relat classic about deriv aspect 