massiv parallel acceler document dsimilar classifi detect web attack profil dby dassoci resili traffic profil solut internet backbon profil dby dassoci resili traffic profil solut internet backbon link homophili applic layer usag traffic classif metric forens multi dlevel approach mine volatil graph basset scalabl gatewai finder larg graph literatur search mix dmembership commun discoveri hcdf hybrid commun discoveri framework evalu statist test dnetwork classifi relat data recent number model techniqu develop data mine machin learn relat network domain instanc independ ident distribut method specif exploit statist depend instanc order improv classif accuraci focu depend affect abil draw accur conclus perform model specif complex link structur attribut depend network data violat assumpt convent statist test make difficult test assess model unbias manner work examin task network classif question algorithm learn model result significantli level perform show commonli form evalu pair test overlap network sampl result unaccept level type error show type error increas correl instanc increas size evalu set increas proport label node network decreas propos method network cross valid combin pair test produc accept level type error provid reason level statist power type ii error appli latent dirichlet alloc group discoveri larg graph paper introduc lda scalabl bayesian approach find latent group structur larg real world graph data exist bayesian approach group discoveri infinit relat model appli small graph coupl hundr node lda short latent dirichlet alloc graph util topic model algorithm find latent group structur specif modifi latent dirichlet alloc lda oper graph data text corpora modif reflect differ real world graph data text corpora node neighbor count document word count empir studi appli lda larg graph thousand node pubm scientif public repositori compar lda quantit perform link predict exist approach bayesian infinit relat model bayesian cross associ averag lda outperform irm cross associ term area roc curv demonstr lda discov qualit inform fast mine complex time dstamp event collect complex time stamp event find pattern anomali event meet person agenda item locat teleconfer public author graphit visual queri system larg graph present graphit system user visual construct queri pattern find exact approxim match subgraph larg attribut graph visual match social network person occup attribut user draw star queri find ceo interact secretari manag account structur similar graphit rai algorithm run queri user chosen data graph gain benefit high speed scalabl abil find exact match graphit toler indirect path ceo account direct path exist graphit fast algorithm estim node proxim find match enabl scale graph databas size demonstr graphit usag benefit dblp author public graph consist node edg demo video graphit download ghost edg classif spars label network address problem classif partial label network network classif observ class label spars techniqu statist relat learn shown perform network classif task exploit depend class label neighbor node relat classifi fail unlabel node label neighbor support learn train phase infer test phase situat aris realworld problem observ label spars paper propos approach network classif combin aspect statist relat learn semi supervis learn improv classif perform spars network approach work ad ghost edg network enabl flow inform label unlabel node experi real world data set demonstr approach perform rang condit exist approach collect classif semi supervis learn fail task approach improv area roc curv auc point exist approach demonstr approach run time proport number label node number edg leverag label dindepend featur classif spars label network empir studi abstract address problem network classif spars label network recent work demonstr success statist relat learn srl semi supervis learn ssl problem approach reli avail label node infer valu miss label label perform approach degrad addit approach sensit specif set node label averag perform accept perform specif task explor complimentari approach network classif base label independ li featur featur calcul valu class label previou work made li featur effect featur classif perform extens studi present empir studi order understand effect experi real world data set show li featur produc classifi sensit specif label assign lead perform improv srl ssl base classifi examin rel util individu li featur show case combin divers network base structur characterist inform collect classif network data real world applic produc network data worldwid web hypertext document connect hyperlink social network peopl connect friendship link commun network comput connect commun link biolog network protein interact network recent focu machin learn research extend tradit machin learn classif techniqu classifi node network articl provid introduct area research head pattern discoveri time devolv multi daspect data examin experiment methodolog classifi relat data experiment methodolog evalu classif algorithm relat network data complic depend relat data instanc survei literatur relat classifi examin experiment methodolog report survei reveal methodolog fall main group base distinct formul classif problem network classif network classif methodolog betweennetwork set straightforward methodolog network classif complex vari explor number variat present experiment result illustr import similar differ methodolog network classif fast deffort pattern match larg attribut graph focu larg graph node attribut social network node label person job titl set find subgraph match user queri pattern star queri find ceo strong interact manag lawyer account structur close similarli loop queri spot monei launder ring tradit sql base method recent graph index method return answer exact match exist method find exact match present user propos good order method toler indirect path ceo account sampl queri direct path exist featur scalabl gener queri nq node data graph node problem polynomi time complex nq prohibit rai graph rai method find high qualiti subgraph time linear size data graph experiment result dlbp author public graph node edg illustr effect scalabl approach result agre intuit speed excel take second averag node queri dblp graph proceed twelfth acm sigkdd intern confer knowledg discoveri data mine privaci concern applic involv microdata recent effici anonym attract research work previou method global recod map domain quasi identifi attribut gener chang valu global recod achiev effect anonym term discern queri answer accuraci anonym data anonym data analysi accept analyt applic attribut data set util analysi util attribut consid previou method paper studi problem util base anonym propos simpl framework util attribut framework cover numer categor data develop simpl effici heurist local recod method util base anonym extens perform studi real data set synthet data set show method outperform state art multidimension global recod method discern queri answer accuraci util base method boost qualiti analysi anonym data visual analysi larg heterogen social network semant structur abstract social network analysi activ area studi sociolog uncov invis relationship actor network understand social process behavior import techniqu varieti applic area web organiz studi homeland secur paper present visual analyt tool ontovi understand larg heterogen social network node link repres concept relat concept relat relat ontolog schema ontovi name inform ontolog social network semant prune larg heterogen network addit semant abstract ontovi user structur abstract import filter make larg network manag facilit analyt reason uniqu capabl ontovi illustr case studi hybrid approach multiresolut model larg dscale scientif data evolut hierarch partit algorithm larg dscale scientif data step increas complex scientif data set grow exponenti size scalabl algorithm heurist partit data increas paper describ threestep evolut hierarch partit algorithm larg scale spatio tempor scientif data set gener massiv simul version algorithm simpl top partit techniqu divid data bisect spatio tempor space shortcom algorithm lead version partit algorithm bottom approach version partit hierarchi construct systemat agglomer underli cartesian grid data final version algorithm util intrins topolog data origin scientif problem build partit hierarchi bottom fashion specif topolog heurist agglomer data level partit hierarchi grow complex algorithm version algorithm build partit hierarchi time build tree larger size data set compar previou version hierarch partit algorithm scalabl algorithm need partit tera scale data set true scientif domain size data set grown exponenti recent year describ evolut hierarch partit algorithm larg scale scientif data set specif larg scale simul program produc data set mesh format data set mesh format consist interconnect grid small zone data point store figur depict mesh produc astrophys simul star mid life mesh data vari time consist multipl dimens variabl irregular grid musick critchlow provid nice introduct scientif mesh data knowledg represent issu semant graph relationship detect import task homeland secur predict threat vulner detect relationship seemingli disjoint entiti structur task semant graph relat data graph attribut relat graph graph encod relationship type link pair type node semant graph similar semant network ai node link type relat ontolog graph schema node set attribut ag attribut node type person select type attribut node link depend human expertis subject arbitrari subject introduc bias algorithm oper semant graph rais knowledg represent issu semant graph provid solut recent develop idea field complex network concept transit evalu relev individu link semant graph detect relationship propos statist measur semant graph illustr semant measur graph construct movi terror data evolut hierarch partit algorithm larg dscale scientif data step increas complex scientif data set grow exponenti size scalabl algorithm heurist partit data increas paper describ threestep evolut hierarch partit algorithm larg scale spatio tempor scientif data set gener massiv simul version algorithm simpl top partit techniqu divid data bisect spatio tempor space shortcom algorithm lead version partit algorithm bottom approach version partit hierarchi construct systemat agglomer underli cartesian grid data final version algorithm util intrins topolog data origin scientif problem build partit hierarchi bottom fashion specif topolog heurist agglomer data level partit hierarchi grow complex algorithm version algorithm build partit hierarchi time build tree larger size data set compar previou version hierarch partit algorithm scalabl algorithm need partit tera scale data set true scientif domain size data set grown exponenti recent year describ evolut hierarch partit algorithm larg scale scientif data set specif larg scale simul program produc data set mesh format data set mesh format consist interconnect grid small zone data point store figur depict mesh produc astrophys simul star mid life mesh data vari time consist multipl dimens variabl irregular grid musick critchlow provid nice introduct scientif mesh data system build intellig agent learn retriev extract inform present system rapidli easili build instruct adapt softwar agent retriev extract inform wisconsin adapt web assist wawa construct intellig agent accept user prefer form instruct user provid instruct compil neural network respons adapt capabl intellig agent agent neural network modifi user provid system construct train exampl user creat train exampl rate web page document importantli wawa agent techniqu reinforc learn intern creat exampl user provid addit instruct life agent experiment evalu home page finder agent seminar announc extractor agent illustr instruct adapt agent retriev extract inform tina eliassi drad terenc critchlow ghaleb abdulla theori drefin approach inform extract investig appli theori refin task extract inform text theori refin partial domain knowledg incorrect supervis learner provid knowledg guid learner task learner refin discard knowledg train supervis learner knowledg base neural network initi compil prior knowledg inform extract task prior knowledg extract slot specif task approach gener test address task gener step produc candid extract intellig search space extract test step train network judg candid output exce system select threshold experi cmu seminarannounc yeast subcellularloc domain demonstr approach instruct cach memori modern medium high speed cpu hold temporarili portion content main memori believ instruct data cach memori referenc percent time requir access main memori cach memori permit 