machin learn knowledg discoveri databas knowledg discoveri databas nontrivi extract implicit previous unknown potenti inform databas argu contrast machin learn knowledg discoveri databas appli real world databas real world databas larg rais problem access real world databas access databas manag system number access reduc minimum properti forc standard set orient interfac relat databas manag system order appli method knowledg discoveri databas present system discov data depend build set orient interfac point main effort put discoveri restrict unari inclus function depend relat databas system embodi infer rel machin learn knowledg discoveri databas knowledg discoveri databas nontrivi extract implicit previous unknown potenti inform databas argu contrast machin learn knowledg discoveri databas appli real world databas real world databas larg rais problem access real world databas access databas manag system number access reduc minimum properti forc standard set orient interfac relat databas manag system order appli method knowledg discoveri databas present system discov data depend build set orient interfac point main effort put discoveri restrict unari inclus function depend relat databas system embodi infer rel machin learn knowledg discoveri databas knowledg discoveri databas nontrivi extract implicit previous unknown potenti inform databas argu contrast machin learn knowledg discoveri databas appli real world databas real world databas larg rais problem access real world databas access databas manag system number access reduc minimum properti forc standard set orient interfac relat databas manag system order appli method knowledg discoveri databas present system discov data depend build set orient interfac point main effort put discoveri restrict unari inclus function depend relat databas system embodi infer rel commun dsearch problem plan success cocktail parti evalu queri result signific databas random queri similar project queri dflow graph defin measur similar queri interest difficult problem reliabl queri similar measur varieti applic queri recommend queri expans advertis paper exploit inform present queri log order develop measur semant similar queri approach reli concept queri flow graph graph base represent queri log queri flow graph aggreg queri reformul user node graph repres queri queri connect part search goal queri similar measur obtain project graph subgraph extract low dimension euclidean space experi show measur obtain captur notion semant similar queri diversifi queri recommend optim framework queri recommend queri recommend integr part modern search engin goal queri recommend facilit user search inform queri recommend user explor concept relat inform paper present formal treatment problem queri recommend framework model queri behavior user probabilist reformul graph queri flow graph boldi al cikm sequenc queri submit user path graph assign score valu queri defin suitabl util function expect util achiev reformul path queri flow graph provid recommend ad shortcut queri flow graph nudg reformul path user user follow path larger expect util discuss detail import question aris propos framework provid exampl meaning util function optim discuss estim effect recommend reformul probabl address complex optim problem suggest effici algorithm solut valid model algorithm extens experiment techniqu appli scenario user behavior model markov process high densiti leadership nuclei endors social network guest editor introduct special issu select paper ecml pkdd learn predict evolut social network special issu ecml pkdd guest editor introduct dnearest neighbor uncertain graph complex network biolog social commun network entail uncertainti model probabilist graph similar problem similar search standard graph fundament problem probabilist graph effici answer nearest neighbor queri nn problem comput closest node specif node paper introduc framework process nn queri probabilist graph propos distanc function extend graph concept shortest path order comput probabilist graph design algorithm base sampl nn queri process effici prune search space techniqu experi distanc function outperform previous altern identifi true neighbor real world biolog data demonstr algorithm scale graph ten million edg effici algorithm larg dscale local triangl count articl studi problem approxim local triangl count larg graph larg graph estim accur number triangl incid node graph question undirect direct graph problem comput global number triangl graph consid knowledg contribut address problem approxim local triangl count focu effici issu aris massiv graph consid direct case distribut local number triangl relat local cluster coeffici interest applic show measur comput detect presenc spam activ largescal web graph provid featur content qualiti assess social network comput local number triangl undirect direct propos approxim algorithm base idea min wise independ permut broder al algorithm oper semi stream fashion space main memori perform log sequenti scan edg graph algorithm fast shortest path distanc estim larg network mine graph evolut rule taxonomi ddriven lump sequenc mine abstract taxonomi event dataset sequenc event studi problem find effici effect wai produc compact represent sequenc model sequenc markov model state correspond node provid taxonomi state repres event subtre node lump observ event state correspond intern node taxonomi compact model easier understand visual expens decreas data likelihood formal defin character problem propos scalabl search method find good trade conflict goal maxim data likelihood minim model complex implement idea taxomo taxonomi driven model appli domain queri log mine mine move object trajectori empir evalu confirm feasibl us approach keyword data mine sequenc analysi markov model queri log analysi spatial data analysi queri signific databas random taxonomi ddriven lump sequenc mine abstract taxonomi event dataset sequenc event studi problem find effici effect wai produc compact represent sequenc model sequenc markov model state correspond node provid taxonomi state repres event subtre node lump observ event state correspond intern node taxonomi compact model easier understand visual expens decreas data likelihood formal defin character problem propos scalabl search method find good trade conflict goal maxim data likelihood minim model complex implement idea taxomo taxonomi driven model appli domain queri log mine mine move object trajectori empir evalu confirm feasibl us approach keyword data mine sequenc analysi markov model queri log analysi spatial data analysi danonym minim loss inform random approxim algorithm comput bucket order www elsevi locat ipl random approxim algorithm comput bucket order queri dlog mine detect spam dai million user search inform web search engin provid implicit feedback result shown queri click feedback encod form queri log consist sequenc search action user queri describ inform term compos queri ii document return search engin iii document click iv rank document list result date time search action click vi anonym identifi session work investig idea character document queri belong queri log goal improv algorithm detect spam document level queri level dr searcher mr browser unifi hyperlink dclick graph introduc unifi graph represent web includ structur usag inform model graph simpl union web hyperlink click graph hyperlink graph express link structur web page click graph bipartit graph queri document denot user search behavior extract search engin queri log import motiv model unifi main activ user web search brows time analyz effect random walk graph intuit task measur combin link structur usag data provid addit inform contain structur independ experiment result show hyperlink click graph strength weak stationari distribut score rank web page evalu unifi graph gener consist robust score follow close result obtain individu graph appli noisi data belief unifi web graph properti improv current web document rank gener rank stationari distribut score deriv random walk combin graph indic structur usag data reliabl situat search wikipedia contextu inform propos framework search wikipedia contextu inform framework extend typic keyword search queri type set term classic web search sourc wikipedia document queri term repres inform user interest find document context queri task rank document wikipedia respect relev queri term context document associ context queri term search result search initi page made relev suggest number featur extend classic querysearch model context document consid ranksvm joachim learn weight individu featur suitabl construct train data document rank queri time product featur weight vector experi propos method consider improv result obtain tradit approach context account queri dflow graph model applic queri log record queri action user search engin valuabl inform interest prefer behavior user implicit feedback search engin result mine wealth inform queri log import applic includ querylog analysi user profil person advertis queri recommend paper introduc queri flow graph graph represent interest knowledg latent queri behavior intuit queri flow graph direct edg queri qi queri qj mean queri part search mission path queri flow graph search behavior likelihood strength edg path queri flow graph outcom queri log mine time tool propos methodolog build graph mine time textual inform aggreg queri user approach build real world queryflow graph larg scale queri log demonstr util concret applic find logic session queri recommend us queri flow graph applic danonym revisit concept anonym recent literatur formal evalu privaci preserv publish tabl introduc mine larg network subgraph count effici semi dstream algorithm local triangl count massiv graph topic queri decomposit introduc problem queri decomposit queri document retriev system produc small set queri union result document correspond approxim origin queri ideal queri repres coher conceptu separ topic provid abstract formul queri decomposit problem tackl perspect show problem instanti specif variant set cover problem provid effici greedi algorithm show problem constrain cluster problem kind constraint cluster predefin cluster develop phase algorithm base hierarch agglom cluster dynam program experi conduct set actual queri web scale search engin confirm effect propos solut find high dqualiti content social media abstract qualiti user gener content vari drastic excel abus spam avail content increas task identifi high qualiti content site base user contribut social media site increasingli import social media gener exhibit rich varieti inform sourc addit content wide arrai content inform link item explicit qualiti rate member commun paper investig method exploit commun feedback automat identifi high qualiti content test case focu yahoo answer larg commun question answer portal rich amount type content social interact introduc gener classif framework combin evid sourc inform tune automat social media type qualiti definit commun question answer domain show system separ high qualiti item rest accuraci close human discret basi problem discret basi problem collect boolean vector find collect boolean basi vector origin vector repres disjunct basi vector show decis version problem np complet optim version approxim finit ratio studi variat problem boolean basi vector mutual otrhogon show variat close relat metric median problem boolean space solv problem algorithm present design variat mention sole base solv median problem heurist intend solv gener discret basi problem studi result extens experi made algorithm synthet real world data result twofold synthet data algorithm real world data result good algorithm unimod segment applic unimod detect algorithm unimod segment applic unimod detect optim segment tree model sequenc data abund applic area comput biologi environment scienc telecommun real life sequenc strong segment structur segment complex paper studi descript sequenc segment variabl length markov chain vlmc tree model discov segment boundari sequenc time obtain vlmc segment context tree probabl distribut vector captur essenti featur segment bayesian inform criterion bic krichevski trofimov probabl kt select number segment sequenc dna data method select segment close correspond annot region gene design trade doff search engin cach danonym minim loss inform recurr predict model sequenc segment segment rearrang abstract comput genom involv compar sequenc base amp quot similar amp quot detect evolutionari function relat ship recent portion human genom sequenc speci fairli short spars mostsequenc effort focus gene short unit similar sequenc measur base charact level differ enc advent genom sequenc technolog emerg consensu measur similar long genom sequenc captur rearrang larg segmentsfound abund human genom paper abstract gener problem comput sequenc similar presenc segment rearrang problem isclos relat comput smallest grammar string block edit distanc string problem otherproblem np hard main result simpl factor approxim algorithm problem contrast approxi mation relat problem factor omega log optim algorithm work linear time pass prove sult relat sequenc similar measur base segment rearrang tight constant factor introduct similar comparison biomolecular sequenc plai import rolein comput genom due premis sequenc similar impact cach search engin neighbor web spam detect web topolog web spam significantli deterior qualiti search engin result larg incent commerci search engin detect spam page effici accur paper present spam detect system topolog web graph exploit link depend web page content page find link host tend belong class spam spam demonstr method incorpor web graph topolog predict obtain base classifi cluster host graph assign label host cluster major vote ii propag predict label neighbor host iii predict label neighbor host featur retrain classifi result accur system detect web spam appli practic larg scale web data estim number citat author reput abstract studi problem predict popular item dynam environ author post continu item provid feedback exist item problem appli predict popular blog post rank photograph photo share system predict citat scientif articl author inform monitor item interest short period time creation case studi show estim number citat academ paper inform past articl written author paper citat inform short period time obtain predict correl actual baselin predict perform system improv predict ad featur extract past publish histori author increas correl actual predict valu price valid dynam network cluster aggreg assess data mine result swap random problem assess signific data mine result high dimension data set studi extens literatur problem mine frequent set find correl signific test chi squar test method result test depend specif attribut dataset test difficult appli set pattern complex result data mine paper simpl random techniqu deal shortcom approach consist produc random dataset row column margin dataset comput result interest random instanc compar result actual data random techniqu assess result type data mine algorithm frequent set cluster rank gener random dataset margin variat markov chain approach base simpl swap oper give theoret result effici random method appli swap random method wellknown dataset result dataset structur discov data mine algorithm random artifact dataset discov structur convei meaning inform optim segment tree model sequenc data abund applic area comput biologi environment scienc telecommun real life sequenc strong segment structur segment complex paper studi descript sequenc segment variabl length markov chain vlmc tree model discov segment boundari sequenc time obtain vlmc segment context tree probabl distribut vector captur essenti featur segment bayesian inform criterion bic krichevski trofimov probabl kt select number segment sequenc dna data method select segment close correspond annot region gene assess data mine result swap random problem assess signific data mine result high dimension data set studi extens literatur problem mine frequent set find correl signific test chi squar test method result test depend specif attribut dataset test difficult appli set pattern complex result data mine paper simpl random techniqu deal shortcom approach consist produc random dataset row column margin dataset comput result interest random instanc compar result actual data random techniqu assess result type data mine algorithm frequent set cluster rank gener random dataset margin variat markov chain approach base simpl swap oper give theoret result effici random method appli swap random method wellknown dataset result dataset structur discov data mine algorithm random artifact dataset discov structur convei meaning inform dimens binari data dataset larg number variabl spars depend structur variabl simpler number variabl suggest defin effect dimension dataset nontrivi problem problem defin robust measur dimens dataset show basic idea fractal dimens adapt binari data fractal dimens difficult interpret introduc concept normal fractal dimens dataset normal fractal dimens count number independ column need achiev unnorm fractal dimens normal fractal dimens measur degre depend structur data studi properti normal fractal dimens discuss comput give empir result normal fractal dimens compar pca assess data mine result swap random problem assess signific data mine result high dimension data set studi extens literatur problem mine frequent set find correl signific test chi squar test method result test depend specif attribut dataset test difficult appli set pattern complex result data mine paper simpl random techniqu deal shortcom approach consist produc random dataset row column margin dataset comput result interest random instanc compar result actual data random techniqu assess result type data mine algorithm frequent set cluster rank gener random dataset margin variat markov chain approach base simpl swap oper give theoret result effici random method appli swap random method wellknown dataset result dataset structur discov data mine algorithm random artifact dataset discov structur convei meaning inform algorithm discov bucket order data order rank item type import task applic queri process scientif data mine total order item mislead group item practic equal rank bucket order total order ti captur essenti order inform overfit data form concept class total order arbitrari partial order address question find bucket order set item pairwis preced inform item discuss method comput pairwis preced data describ simpl effici algorithm find good bucket order algorithm provabl approxim guarante scale larg dataset provid experiment result artifici real data show us bucket order demonstr accuraci effici algorithm discret basi problem discret basi problem collect boolean vector find collect boolean basi vector origin vector repres disjunct basi vector show decis version problem np complet optim version approxim finit ratio studi variat problem boolean basi vector mutual otrhogon show variat close relat metric median problem boolean space solv problem algorithm present design variat mention sole base solv median problem heurist intend solv gener discret basi problem studi result extens experi made algorithm synthet real world data result twofold synthet data algorithm real world data result good segment dimension reduct sequenc segment dimension reduct method studi high dimension sequenc reduc complex represent origin data paper studi interplai techniqu formul problem segment sequenc model basi small size essenti reduc dimens input sequenc give algorithm problem combin exist method sequenc segment dimension reduct propos algorithm prove guarante qualiti solut obtain describ experiment result synthet real dataset includ data exchang rate genom sequenc experi show algorithm discov underli structur data includ segment structur interdepend dimens keyword segment multidimension data pca time seri cluster aggreg paramet dfree spatial data mine mdl spatial data consist set binari featur take valu collect spatial extent grid cell propos method simultan find spatial correl featur occurr pattern paramet emploi minimum descript length mdl principl coupl natur compress region defin good mean featur occurr pattern good help compress set locat featur convers spatial correl good help compress set featur region approach scalabl larg dataset number locat featur evalu method real synthet dataset mine chain relat tradit data mine applic problem mine singl relat attribut scientif bibliographi databas author relat paper interest discov associ rule author real life multipl attribut relat chain relat author write paper paper concern topic mine relat chain pose addit challeng paper problem chain relat find selector object project relat satisfi specif properti motiv approach properti hold dataset hold project data selector set discuss algorithm examin condit apriori techniqu experiment demonstr effect method dimens induc cluster dimens induc cluster unimod segment sequenc approxim collect frequent set studi problem data mine comput collect frequent item set larg transact databas obstacl applic frequent set mine size output collect larg carefulli examin understood user restrict output border frequent item set collect allevi problem geometr combinatori tile data paper introduc simpl probabilist model hierarch tile data basic tile specifi subset row subset column data rectangl probabl occurr cell hierarch tile addition set except tile probabl subrectangl origin rectangl row column order consist consecut element order tile geometr combinatori give simpl random algorithm find good geometr tile main result show spectral order techniqu find good order turn combinatori tile geometr tile give empir result perform method price valid dynam network autom rank databas queri result investig problem rank answer databas queri tupl return adapt appli principl probabilist model inform retriev structur data propos solut domain independ leverag data workload statist correl rank function custom applic present result preliminari experi demonstr effici qualiti rank system fragment order high dimension collect data occur applic attribut data set typic consid unord case natur total partial order underli variabl data set exampl variabl order exist includ term document cours enrol data paleontolog site fossil data collect observ applic flat unord set data set respect underli order variabl variabl respect underli order variabl observ nois level variabl appear observ similarli longer sequenc variabl expect observ indic lt lt ak occur observ find recurr sourc sequenc time seri displai tremend variabl reason assum sequenc gener assembl small number sourc contribut segment sequenc hidden sourc sequenc written concaten piec stem sourc defin segment problem show np hard gener case give approxim algorithm achiev approxim ratio error measur error measur gener result higher dimens give empir result real chromosom artifici data show method work practic xtract learn document type descriptor xml document collect dtd infer xml document xtract approach xml rapidli emerg standard data represent exchang web document type descriptor dtd valuabl inform structur xml document crucial role effici storag queri xml data import dtd mandatori document xml databas accompani dtd paper present overview xtract system infer dtd schema databas xml document dtd syntax incorpor full express power regular express naiv approach typic fail produc concis intuit dtd xtract infer algorithm emploi sequenc sophist step involv find pattern input sequenc replac regular express gener gener candid dtd factor candid dtd adapt algorithm logic optim literatur appli minimum descript length mdl principl find dtd candid combinatori algorithm max csp maintain stream statist slide window extend abstract mayur datar aristid gioni piotr indyk rajeev motwani abstract problem maintain aggreg statist data stream respect data element refer model slide window model basic problem stream bit maintain count number element stream show ffl log bit memori estim number factor ffl give match lower bound omega gamma ffl log memori bit determinist random algorithm extend scheme maintain sum posit integ provid match upper lower bound gener problem appli techniqu obtain effici algorithm lp norm vector slide window model algorithm basic count problem adapt techniqu work slide window model multipl overhead ffl log memori ffl factor loss accuraci includ maintain approxim histogram hash tabl statist aggreg sum averag evalu strategi similar search web find page web similar queri page relat page import compon modern search engin varieti strategi propos answer relat page queri compar evalu user studi expens larg strategi space search tune paramet present techniqu automat evalu strategi web hierarchi open directori place user feedback appli evalu methodolog mix document represent strategi includ text anchor text link discuss rel advantag disadvantag approach examin final describ ecient construct similar index chosen strategi provid sampl result index maintain stream statist slide window mayur datar aristid gioni piotr indyk rajeev motwani abstract problem maintain aggreg statist data stream respect data element refer model slide window model basic problem stream bit maintain count number element stream show ffl log bit memori estim number factor ffl give match lower bound omega gamma ffl log memori bit determinist random algorithm extend scheme maintain sum posit integ provid match upper lower bound gener problem appli techniqu obtain effici algorithm lp norm vector slide window model algorithm basic count problem adapt techniqu work slide window model multipl overhead ffl log memori ffl factor loss accuraci includ maintain approxim histogram hash tabl statist aggreg sum averag effici tunabl similar set retriev model complex data set modern object relat system support set attribut queri capabl paper initi formal studi index techniqu set attribut base similar suitabl defin notion similar set techniqu modern applic recommend collabor filter autom advertis techniqu probabilist approxim natur design principl creat structur make wide data structur techniqu mean eas integr exist infrastructur find interest associ support prune abstract associ rule mine heretofor reli condit high support work effici priori algorithm effect rule interest relationship occur frequent number applic data mine identif similar web document cluster collabor filter rule interest compar instanc data case highli correl item possibl causal relationship infrequ item develop famili algorithm solv problem emploi combin random sampl hash techniqu provid analysi algorithm develop conduct experi real synthet data obtain compar perform analysi find interest associ support prune abstract associ rule mine heretofor reli condit high support work effici priori algorithm effect rule interest relationship occur frequent number applic data mine identif similar web document cluster collabor filter rule interest compar instanc data case highli correl item possibl causal relationship infrequ item develop famili algorithm solv problem emploi combin random sampl hash techniqu provid analysi algorithm develop conduct experi real synthet data obtain compar perform analysi scalabl techniqu cluster web cluster crucial techniqu deal massiv amount inform present web cluster perform offlin independ search queri perform onlin result search queri offlin approach aim effici cluster similar page web techniqu local sensit hash lsh web page hash similar page higher probabl collis dissimilar page preliminari experi stanford webbas shown hash base scheme scale million url xtract system extract document type descriptor xml document scalabl techniqu cluster web cluster crucial techniqu deal massiv amount inform present web cluster perform offlin independ search queri perform onlin result search queri offlin approach aim effici cluster similar page web techniqu local sensit hash lsh web page hash similar page higher probabl collis dissimilar page preliminari experi stanford webbas shown hash base scheme scale million url similar search high dimens hash nearest neighbor queri problem aris larg varieti databas applic context similar search late increas interest build search index structur perform similar search high dimension data imag databas document collect time seri databas genom databas techniqu solv problem fall prei curs dimension amp quot data structur scale poorli data dimension fact number dimens exce search tree relat structur involv inspect larg fraction databas brute forc linear search suggest select featur choic distanc metric typic applic heurist determin approxim nearest neighbor su ce practic purpos paper examin scheme approxim similar search base hash basic idea hash point surfac reflect recoveri point light illumin paper algorithm colour recoveri present assum geometri scene spectral power distribut point illumin sourc respons function sensor calibr jointli algorithm colour recoveri part integr system develop laboratori environment model geometri scene recov laser stripe rang finder inform exploit colour recoveri algorithm point light sourc attach system illumin scene order confin undesir side effect ambient light joint spectral power distribut point light sourc respons function camera obtain line calibr introduct object paper present algorithm obtain surfac colour inform control enviro 