topic model link document updat rule estim latent topic model plai import role unsupervis learn corpu probabilist interpret corpu term latent topic space underpin assumpt topic model base document assum independ assumpt hold true realiti relat document wai citat relat research paper address limit paper present bernoulli process topic bpt model interdepend document model random bernoulli process bpt model document model distribut topic mixtur distribut relat document bpt aim obtain document model incorpor relat document appli applic includ detect topic corpora cluster document appli bpt model document collect experiment comparison state art approach demonstr promis perform smooth optim effect multipl kernel learn facetcub framework incorpor prior knowledg dneg tensor factor unsupervis learn link document latent topic model plai import role unsupervis learn corpu probabilist interpret corpu term latent topic space underpin assumpt topic model base document assum independ assumpt hold true realiti relat document wai citat relat research paper address limit paper present bernoulli process topic bpt model interdepend document model random bernoulli process bpt model document model distribut topic mixtur distribut relat document bpt aim obtain document model incorpor relat document appli applic includ detect topic corpora cluster document appli bpt model document collect experiment comparison state art approach demonstr promis perform direct network commun detect popular product link model paper problem commun detect direct network probabilist model exist probabilist model commun detect symmetr incom link outgo link treat equal condit type incom outgo link model present probabilist model direct network commun detect aim model incom link outgo link simultan differenti introduc latent variabl node product node popular explicitli captur outgo link incom link demonstr gener propos framework show symmetr model condit model commun detect deriv propos framework special case lead understand exist model deriv effici em algorithm comput maximum likelihood solut propos model extens empir studi verifi effect model insight obtain unifi framework featur select gene express model dbase entropi abstract gene express data larg number gene small number sampl featur select gene express data aim find set gene discrimin biolog sampl type machin learn techniqu tradit gene select base empir mutual inform suffer data spars issu due small number sampl overcom spars issu propos model base approach estim entropi class variabl model data multivari normal distribut fit data multivari normal distribut maximum entropi real valu distribut standard deviat wide approxim distribut data follow multivari normal distribut condit distribut class variabl select featur normal distribut entropi comput log determin covari matrix larg number gene comput log determin effici propos algorithm larg reduc comput cost experi gene dataset comparison approach show accuraci multivari gaussian gener model featur select effici algorithm resolut enhanc base learn spars associ imag patch compar document summar discrimin sentenc select collect document group quick question differ group paper studi problem summar differ document group discrimin sentenc select method propos extract discrimin sentenc repres specif characterist document group experi real world data set demonstr effect propos method resolut dinvari imag represent applic present resolut invari imag represent iir framework paper iir framework includ method build set multi resolut base train imag estim optim spars resolut invari represent imag reconstruct miss patch resolut level propos iir framework potenti resolut enhanc applic discuss imag magnif applic paper applic appli iir framework perform multi scale imag magnif introduc train strategi built compact iir set applic iir framework extend conduct continu imag scale base resolut level resolut dinvari imag represent applic present resolut invari imag represent iir framework paper iir framework includ method build set multi resolut base train imag estim optim spars resolut invari represent imag reconstruct miss patch resolut level propos iir framework potenti resolut enhanc applic discuss imag magnif applic paper applic appli iir framework perform multi scale imag magnif introduc train strategi built compact iir set applic iir framework extend conduct continu imag scale base resolut level knowledg discoveri citat network abstract knowledg discoveri scientif articl receiv increas attent recent huge repositori made develop internet digit databas corpu scientif articl digit librari document connect citat document plai role corpu document citat document exist topic model effort made differenti role topic distribut role relat paper propos bernoulli process topic bpt model model corpu level document level citat level bpt model document represent latent topic space role multilevel hierarch structur citat network captur gener process involv bernoulli process distribut paramet bpt model estim variat approxim approach addit conduct experiment evalu document model task appli bpt model scientif corpu discov latent topic comparison state ofth art method demonstr promis perform keyword unsupervis learn latent model text mine knowledg discoveri citat network abstract knowledg discoveri scientif articl receiv increas attent recent huge repositori made develop internet digit databas corpu scientif articl digit librari document connect citat document plai role corpu document citat document exist topic model effort made differenti role topic distribut role relat paper propos bernoulli process topic bpt model model corpu level document level citat level bpt model document represent latent topic space role multilevel hierarch structur citat network captur gener process involv bernoulli process distribut paramet bpt model estim variat approxim approach addit conduct experiment evalu document model task appli bpt model scientif corpu discov latent topic comparison state ofth art method demonstr promis perform keyword unsupervis learn latent model text mine resolut dinvari imag represent content dbase zoom paper present resolut invari imag represent iir framework appli content base zoom cbz applic explain gener multi resolut base set learn imag represent resolut invari kei technolog support continu imag scale task cbz applic exist base resolut enhanc approach handel simpli imag interpol algorithm give satisfactori imag qualiti discuss cluster base method construct base set experiment result show method give good imag qualiti propos iir framework outperform exist method aspect larg dscale collabor predict nonparametr random effect model nonparametr model introduc multipl relat regress task input common data space tradit transfer learn model inappropri depend output fulli resolv inputspecif task specif predictor propos model treat output respons condition independ predictor unobserv random effect model nonparametr sens dimension random effect priori determin data approach estim model present em algorithm effici larg scale collabor predict problem obtain predict accuraci competit state art result order innov laggard product categor recommend buyer exhibit purchas behavior rush purchas product tend cautiou wait review peopl trust market analysi group buyer refer innov earli adopt group refer laggard adopt behavior dynam featur user vari group product innov literatur innov electron adopt order user dynam featur product predict futur potenti buyer dynam featur unavail descript product paper studi user behavior onlin review websit epinion propos model user adopt behavior creat total order user rate product categori develop greedi algorithm markov chain base algorithm comput categori total order show user behavior inform accur predict categori product predict user follow epinion trust network evid demonstr total order group user commun close resembl trust network adopt order featur recommend system combin link content commun detect discrimin approach order innov laggard product categor recommend buyer exhibit purchas behavior rush purchas product tend cautiou wait review peopl trust market analysi group buyer refer innov earli adopt group refer laggard adopt behavior dynam featur user vari group product innov literatur innov electron adopt order user dynam featur product predict futur potenti buyer dynam featur unavail descript product paper studi user behavior onlin review websit epinion propos model user adopt behavior creat total order user rate product categori develop greedi algorithm markov chain base algorithm comput categori total order show user behavior inform accur predict categori product predict user follow epinion trust network evid demonstr total order group user commun close resembl trust network adopt order featur recommend system bayesian approach find commun evolut dynam social network fast nonparametr matrix factor larg dscale collabor filter sheer growth onlin user data challeng develop prefer learn algorithm suffici flexibl model afford comput paper develop nonparametr matrix factor method allow latent factor low rank matrix factor method singular decomposit svd probabilist princip compon analysi ppca data driven dimension increas data size show formul nonparametr model similar optim share similar procedur compar tradit parametr low rank method nonparametr model appeal flexibl model complex data depend model advantag comput price highli challeng scale larg scale problem hamper applic applic collabor filter paper introduc optim algorithm simpl implement learn nonparametr matrix factor model highli effici larg scale problem experi eachmovi netflix largest public benchmark date demonstr nonparametr model make accur predict user rate computation compar faster train comparison previou state art parametr matrix factor model latent topic model link document visual dqualiti optim super resolut paper propos robust imag super resolut algorithm aim maxim visual qualiti super resolut result good super resolut algorithm fidel preserv imag detail enhanc smooth defin percept base measur visual qualiti base qualiti measur formul imag super resolut optim aim maxim qualiti qualiti measur quadrat optim solv effici experi larg imag set subject user studi demonstr effect percept base qualiti measur robust effici propos method categori subject descriptor acm cc comput graphic pictur imag gener displai algorithm imag process comput vision enhanc sharpen deblur noisi video super dresolut low qualiti video limit resolut suffer nois directli sampl video nois deterior visual qualiti due magnifi nois paper address problem unifi framework achiev simultan de nois super resolut framework formul noisi video super resolut optim problem aim maxim visual qualiti result good qualiti result fidel preserv detailpreserv smooth propos measur qualiti scenario de nois superresolut experi varieti noisi video demonstr effect present algorithm integr cluster multi ddocument summar improv document understand document understand techniqu document cluster multi document summar receiv attent recent year current document cluster method repres document term document matrix perform cluster algorithm cluster method group document satisfactorili hard peopl captur mean document satisfactori interpret document cluster paper propos languag model simultan cluster summar document util mutual influenc document cluster summar method make document cluster method meaning interpret document summar method take document context inform consider probabilist polyad factor applic person recommend multipl dimension polyad data exist applic person recommend multipledimension data summar analyz dimens polyad data principl challeng research problem exist method separ analyz margin relationship pairwis dimens combin result motiv fact dimens polyad data jointli affect propos probabilist polyad factor approach directli model dimens simultan unifi framework show connect probabilist polyad factor neg version tucker tensor factor provid detail theoret analysi model framework discuss implement techniqu model propos extens basic framework appli propos model applic person recommend extens experi social bookmark dataset delici paper citat dataset cites demonstr effect propos model stochast relat model larg dscale dyadic data mcmc stochast relat model srm provid rich famili choic learn predict dyadic data set entiti model gener matrix factor supervis learn problem util attribut entiti hierarch bayesian framework previous variat bay infer appli srm scalabl size entiti set grow ten thousand paper introduc markov chain mont carlo mcmc algorithm equival model srm order scale comput larg dyadic data set superior scalabl predict accuraci demonstr collabor filter problem involv ten thousand user half million item stochast relat model stochast relat model srm gener gaussian process gp model relat domain observ dyadic datum index pair entiti model dyadic data multipl interact gaussian process prior multi ddocument summar sentenc dlevel semant analysi symmetr matrix factor multi document summar aim creat compress summari retain main characterist origin set document approach statist machin learn techniqu extract sentenc document paper propos multi document summar framework base sentenc level semant analysi symmetr neg matrix factor calcul sentenc sentenc similar semant analysi construct similar matrix symmetr matrix factor shown equival normal spectral cluster group sentenc cluster final inform sentenc select group form summari experiment result duc duc data set demonstr improv propos framework implement exist summar system studi factor benefit high perform conduct trnon dgreedi activ learn text categor convex ansduct experiment design learn multipl graph document recommend web offer rich relat data semant paper address problem document recommend digit librari document question network citat entiti relat due sparsiti singl graph nois graph construct propos method combin multipl graph measur document similar factor strategi base natur graph method seek singl low dimension embed document captur rel similar latent space base obtain embed recommend framework develop semisupervis learn graph addit address scalabl issu propos increment algorithm increment method significantli improv effici calcul embed incom document batch increment method evalu real world dataset prepar cites experi demonstr signific qualiti improv batch method signific effici improv toler qualiti loss increment method facetnet framework analyz commun evolut dynam network discov commun social network data analyz commun evolut commun inher characterist human interact onlin social network paper citat network commun evolv time due individu role social statu network individu research interest present innov algorithm deviat tradit step approach analyz commun evolut tradit approach commun detect time slice compar determin correspond argu approach inappropri applic noisi data paper propos facetnet analyz commun evolut robust unifi text categor gener discrimin analysi dynam activ probe helpdesk databas helpdesk databas store past interact custom compani improv custom servic qualiti common scenario helpdesk databas find recommend exist problem custom custom provid incomplet inaccur inform manual prepar list clarif question work larg databas paper investig problem automat gener minim number question reach recommend paper propos dynam activ probe method compar altern decis tree case base reason method distinct featur activ probe custom inform reach recommend inform provid custom immedi method dynam gener question probe featur ensur inform custom method base probabilist model data augment method avoid overfit estim probabl model featur ensur method robust databas incomplet error experiment result verifi effect approach kdd workshop report dmmt data mine matric tensor effici video object segment graph dcut segment video object background popular comput vision problem import applic exist method computation expens requir manual initi static camera rigid scene previou work propos joint spatio tempor linear regress algorithm automat cluster spars edg corner pixel video frame obtain motion model object background label rest pixel object segment paper propos model residu error color intens residu error tempor label consist featur color edg orient consist constrain graph appli graph cut algorithm minim energi graph obtain optim segment motion layer boundari final object layer identifi simpl heurist experiment segment result video webcam promis structur tempor analysi blogospher commun factor blogospher uniqu structur tempor properti blog typic commun media human individu paper propos techniqu captur structur tempor dynam blog commun framework commun set blog commun trigger event new articl commun repres structur tempor dynam commun graph blog commun commun intens activ level commun vari time method commun factor extract commun blogospher commun blog observ set subgraph thread discuss commun extract formul factor problem framework constrain optim object explain observ interact blogospher time provid scalabl algorithm comput solut constrain optim problem extens experiment studi synthet real blog data demonstr techniqu discov meaning commun detect tradit method predict matrix dvariat model increasingli import learn partial observ random matrix predict miss element assum entir matrix singl sampl drawn matrix variat distribut suggest matrixvari model mvtm predict miss element show mvtm gener rang probabilist model automat perform model select encourag spars predict model due conjugaci prior difficult make predict comput mode posterior distribut suggest optim method sequenti minim convex upper bound log likelihood effici scalabl experi toi data eachmovi dataset show good predict accuraci model combin content link classif matrix factor world wide web rich textual content interconnect complex hyperlink huge databas violat assumpt held convent statist method web page consid independ ident sampl difficult appli tradit mine learn method solv web mine problem web page classif exploit content link structur research direct recent receiv consider attent earli stage method exploit link structur content inform combin author inform content inform decompos link structur hub author featur appli addit document featur practic attract great simplic paper aim design algorithm exploit content linkag inform carri joint factor linkag adjac matrix document term matrix deriv represent web page low dimension factor space explicitli separ content hub author factor analysi perform base compact represent web page experi propos method compar state art method demonstr excel accuraci hypertext classif webkb cora benchmark captur user interest exploit explor hierarch document classif automat gener hierarchi integr featur sourc music inform retriev empir studi multi dlabel classif trend analysi larg document stream power comput technolog inspir peopl investig inform hidden huge amount document interest document rel time order call document stream exampl includ tv new forum email compani project call center telephon log insight document stream detect event document stream time sensit dirichlet process mixtur model find event document stream time sensit dirichlet process mixtur model gener model potenti infinit number mixtur compon dirichlet compound multinomi model model distribut word document time sensit dirichlet process mixtur model exponenti decai kernel model polynomi decai function kernel dirichlet process model slide window kernel model experi tdt dataset shown time sensit model perform term accuraci dirichlet process mixtur model slide window kernel polynomi kernel promis detect event themeriv provid visual event time axi themeriv peopl easili pictur event evolv themeriv investig top word high level summar event experi result tdt dataset suggest slide window kernel choic term captur trend event express empir studi multi dlabel classif multi dlabel classif maximum entropi method classif problem requir classifi assign singl document categori call multilabel classif categori problem condition independ mutual exclus trivial directli emploi state theart classif algorithm lose inform relat categori paper explor correl categori maximum entropi method deriv classif algorithm multi label document experi show method significantli outperform combin singl label approach identifi earli buyer purchas data market research shown consum exhibit varieti di erent purchas behavior specif tend purchas product earlier consum identifi earli buyer person market strategi potenti improv ectiv paper present parametr approach problem identifi earli buyer purchas data formul take input detail purchas inform consum construct weight direct graph node correspond consum edg correspond purchas consum common edg weight frequent consum purchas product earlier consum effici multi dwai text categor gener discrimin analysi discrimin analysi multi dclass classif distribut data mine model base similar distribut data mine ddm activ enjoi grow amount attent incept current ddm techniqu regard distribut data set singl virtual tabl assum exist global model gener data combin central paper propos similaritybas distribut data mine sbddm framework explicitli differ distribut sourc consider similar measur introduc effect evalu valid paper illustr limit current ddm techniqu concret case studi final distribut cluster sbddm framework discuss gene function classif semisupervis learn heterogen data framework individu dfocus teleconferenc ift effici reproject techniqu topic hierarchi gener linear discrimin project associ dbase similar test applic paper propos similar measur basket dataset base associ measur calcul support count formula inspir inform entropi experi real synthet dataset show effect measur paper investig applic similar measur studi problem find map categor databas attribut set similar measur gener approach identifi map propos approach implement base similar measur propos paper perform evalu valid paper explor applic similar measur mine distribut dataset algorithm cluster high dimension distribut data cluster problem identifi distribut pattern intrins correl larg data set partit data point similar class cluster problem wide studi machin learn databas statist paper studi problem cluster high dimension data paper propos algorithm call cofd algorithm distanc base cluster algorithm high dimension space base estim joint probabl margin paper explor implic approxim concept base bayesian decis procedur plausibl unif fuzzi set rough set approach approxim concept show concept approxim set result cut fuzzi set theori obtain hand concept approxim set deriv algebra probabilist rough set approxim base principl maximum minimum entropi give interpret fuzzi intersect union result enhanc understand broaden applic fuzzi rough set cofd algorithm ddistanc base cluster high dimension space improv medic fbiolog data classif perform wavelet preprocess real world dataset nois nois degrad perform learn algorithm motiv success wavelet denois techniqu imag data explor gener solut allevi eect noisi data wavelet preprocess medic biolog data classi cation experi divid categori dierent classi cation algorithm speci databas ecoli speci classi cation algorithm decis tree dierent databas experi result show wavelet denois noisi data improv accuraci classi cation method local attribut strong attent object spot integr multimod input intellig human comput interfac expect comput work user cooper manner achiev goal comput awar user attent provid assist explicit user request cognit studi ey movement suggest accomplish learn task perform focu attent lock ongo work ey movement close relat object manipul task light develop attent object spot system integr multimod data consist ey posit head posit video person perspect detect user focu attent model ey gaze head movement hidden markov model hmm represent attent point time object user interest automat extract recogn report result experi find attent object natur task make peanut butter sandwich document cluster cluster refin model select capabl paper propos document cluster method strive achiev high accuraci document cluster capabl estim number cluster document corpu model select capabl accur cluster document corpu emploi richer featur set repres document gaussian mixtur model gmm expect maxim em algorithm conduct initi document cluster initi result identifi set discrimin featur cluster refin initi obtain document cluster vote cluster label document discrimin featur set refin process discrimin featur identif cluster label vote iter appli converg document cluster hand model select capabl achiev introduc random cluster initi stage discov number cluster run document cluster process fix number time yield suffici similar result perform evalu exhibit clear superior propos method improv document cluster model select accuraci evalu demonstr featur cluster refin process contribut document cluster accuraci survei wavelet applic data mine recent signific develop wavelet method data mine process written comprehens survei topic goal paper fill void paper present high level data mine framework reduc process smaller compon applic wavelet compon reviewd paper conclud discuss impact wavelet data mine research outlin potenti futur research direct applic 