ptm probabilist topic map model mine parallel document collect medic case dbase retriev leverag medic ontolog physician feedback uiuc dibm imageclef abstract paper report experi result uiuc ibm team particip medic case retriev task imageclef experi multipl method leverag medic ontolog user physician feedback work achiev retriev perform submiss local supervis metric learn tempor physiolog data social action track nois toler time dvari factor graph user behavior action inasocialnetworkareinfluencedbyvari factor person interest social influenc global trend public systematicallystudi social action evolv dynam social network towhat extent factor affect user action paper propos nois toler time vari factor graph model ntt fgm model predict social action ntt fgm simultan model social network structur user attribut user action histori predict user futur action specif user action time gener latent state influenc attribut herownlatentstateattimet andherneighbor state attimetandt basedonthisintuit weformalizeth social action track problem ntt fgm model present effici algorithm learn model combin idea continu linear system markov random field final present case studi model predict futur social action valid model type ofreal worlddataset qualit ourmodelcandiscov interestingpatternsofthesocialdynam quantit experiment resultsshowthattheproposedmethodoutperformsseveralbaselin method social actionpredict categori andsubject descriptor tempor recommend graph long short dterm prefer fusion contextour contextu contour analysi dynam multi drelat cluster contextour contextu contour analysi dynam multi drelat cluster facetatla multifacet visual rich text corpora smallblu social network analysi expertis search collect intellig head metric activ learn applic servic classif metafac commun discoveri relat hypergraph factor social influenc analysi larg dscale network himap adapt visual larg dscale onlin social network visual larg scale onlin social network challeng essenti task paper present himap system visual cluster graph hierarch group summar himap emploi adapt data load techniqu accur control visual densiti graph view optim layout algorithm kind edg bundl method effect avoid visual clutter commonli found previou social network visual tool himap integr suit interact user easili navig social map smooth coher view transit momentum final confirm effect himap algorithm graph traves base evalu multivi content dbase social network explor multi dwai visual analysi extract commun structur relat hypergraph cluster subset select applic servic metric head pattern discoveri time devolv multi daspect data hierarch paramet dfree commun discoveri abstract larg bipartit graph document term userproduct graph find meaning commun quickli automat propos commun hierarchi commun commun propos method context specif cluster tree cct find commun multipl level user intervent base inform theoret principl mdl specif partit graph progress refin subgraph allow user quickli navig global coars structur graph focus local pattern fring benefit addit indic qualiti achiev compress typic hierarch method demonstr scalabl effect real larg graph scalabl tensor decomposit multi daspect data mine disco distribut dcluster map dreduc case studi petabyt dscale end dto dend mine huge dataset preval research routin work dataset terabyt size interest real world applic produc huge volum messi data mine process involv step start pre process raw data estim final model data abund scalabl easyto tool distribut process emerg map reduc wide embrac academia industri databas term map reduc simpl power execut engin complement data storag manag compon paper describ experi find appli map reduc raw data final model import mine task focu cluster studi applic text mine collabor filter bio informat graph mine propos distribut cluster disco framework introduc practic approach distribut data pre process cluster develop disco hadoop open sourc map reduc implement show disco scale effici process analyz extrem larg dataset hundr gigabyt commod hardwar scoop autom social recommend enterpris process manag bayesian tensor analysi head pattern discoveri time devolv multi daspect data colibri fast mine larg static dynam graph low rank approxim adjac matrix graph essenti find pattern commun detect anomali addition desir track low rank structur graph evolv time effici limit storag real graph typic thousand million node spars standard decomposit svd preserv sparsiti led develop method cur cmd seek nonorthogon basi sampl column row spars matrix approach typic produc overcomplet base wast space time paper propos famili colibri method deal challeng version static graph colibri iter find nonredund basi prove loss accuraci compar competitor cur cmd achiev signific save space time real data colibri requir space order magnitud faster proport squar number redund column addition propos effici updat algorithm dynam time evolv graph colibri evalu larg real network traffic dataset show colibri time faster publish competitor cmd head pattern discoveri time devolv multi daspect data bayesian tensor approach dd face model increment pattern discoveri stream graph tensor ii auspic depart energi lawrenc livermor spars graph mine compact matrix decomposit increment tensor analysi theori applic hide crowd privaci preserv evolv stream correl track address problem preserv privaci stream receiv surprisingli limit attent static data studi wide approach base random perturb data valu stream pose addit challeng analysi data perform increment limit process time buffer space make batch approach unsuit characterist stream evolv time approach base global analysi data adequ show effici effect track correl autocorrel structur multivari stream leverag add nois maxim preserv privaci sens hard remov techniqu achiev result previou static global approach requir limit process time memori provid mathemat analysi experiment evalu real data valid correct effici effect algorithm dynam mixtur model multipl time dseri tradit probabilist mixtur model latent dirichlet alloc impli data record document fulli exchang data natur collect time obei order time paper present dynam mixtur model dmm onlin pattern discoveri multipl time seri dmm notic drawback svd base method data stream neg valu hidden variabl produc nonneg input appli dmm model real world dataset achiev significantli result intuit interpret graphscop paramet dfree mine larg time devolv graph compact matrix decomposit larg spars graph larg spars graph find pattern anomali import applic model larg spars graph network traffic monitor research citat network analysi social network analysi regulatori network gene low rank decomposit svd cur power techniqu reveal latent hidden variabl pattern high dimension data method ignor sparsiti properti graph incur high memori comput cost practic propos method compact matrix decomposit cmd comput spars low rank approxim cmd dramat reduc comput cost space requir exist decomposit method svd cur cmd kei build block propos procedur effici construct analyz dynam graph real time applic data provid theoret guarante method present result real larg dataset network flow data gb trace host month dblp mb year show cmd order magnitud effici state art svd cur faster requir space reconstruct accuraci final demonstr cmd detect anomali monitor timeevolv graph successfulli detect worm hierarch scan pattern real network data mine larg graph stream matrix tensor tool intellig system monitor larg cluster intemon prototyp monitor mine system larg cluster monitor host prototyp data center cmu snmp protocol store monitor data mysql databas visual time seri data jsp web base frontend interfac user set cluster monitor system abil automat analyz monitor data real time alert user potenti anomali state art stream mine method sophist definit anomali broken correl input stream pinpoint reason anomali intemon user friendli gui user perform interact mine task fulli oper window dbase tensor analysi high ddimension multi daspect stream data stream valu multipl aspect environment sensor type temperatur humid locat timestamp type locat addit aspect model stream simultan find pattern multipl aspect increment stream fashion paper problem address gener data model tensor stream effect algorithm framework window base tensor analysi wta variat wta independentwindow tensor analysi iw move window tensor analysi mw present evalu extens real dataset final illustr import applic multi aspect correl analysi maca wta demonstr effect environment monitor applic local correl track time seri address problem captur track local correl time evolv time seri approach base compar local auto covari matric spectral decomposit seri gener notion linear cross correl concis captur wide varieti local pattern trend method produc gener similar score evolv time accur reflect chang relationship final estim increment stream set demonstr us robust effici wide rang real dataset distribut pattern discoveri multipl stream conclus recommend express materi author necessarili reflect view nation scienc foundat fund parti stream graph dynam tensor analysi find pattern author keyword associ evolv time datacub product branchcustom sale inform matrix decomposit princip compon analysi pca variant invalu tool mine dimension reduct featur select rule identif numer set stream data text graph social network order author keyword propos envis higher order data tensor tap vast literatur topic method necessarili scale oper semi infinit stream introduc dynam tensor analysi dta method variant dta compact summari high order high dimension data reveal hidden correl algorithm design dta carefulli scalabl space effici store past fulli automat user defin paramet propos sta stream tensor analysi method fast stream approxim dta implement method appli real set anomali detect multi latent semant index real larg dataset network flow data gb month dblp mb year experi show method fast accur find interest pattern outlier real dataset distribut pattern discoveri multipl stream conclus recommend express materi author necessarili reflect view nation scienc foundat fund parti intemon intellig system monitor larg cluster intemon prototyp monitor mine system larg cluster monitor host prototyp data center cmu snmp protocol store monitor data mysql databas visual time seri data jsp web base frontend interfac user set cluster monitor system abil automat analyz monitor data real time alert user potenti anomali state art stream mine method sophist definit anomali broken correl input stream pinpoint reason anomali intemon user friendli gui user perform interact mine task fulli oper spatio dtempor join select move object databas manag spatial object continu chang posit extent time recent found larg interest databas commun queri move object interest tempor topolog relationship evolv spatial object concept spatio tempor predic propos describ relationship goal paper design effici algorithm spatiotempor join select paper propos design algorithm predic individu emploi gener algorithm scheme cover present futur predic definit intemon continu mine sensor data larg dscale dinfrastructur modern data center larg number compon monitor includ server switch router environment control system paper describ intemon prototyp monitor mine system data center snmp protocol monitor data center carnegi mellon store monitor data mysql databas allow visual time seri data jsp web base frontend interfac system administr set intemon cluster monitor system abil automat analyz correl monitor data real time alert administr potenti anomali effici state art stream mine method report broken correl input stream method intellig compress histor data avoid administr configur threshold base monitor band onlin latent variabl detect sensor network neighborhood format anomali detect bipartit graph real applic model bipartit graph user file system trader stock financi trade system confer author scientif public network introduc oper bipartit graph identifi similar node neighborhood format find abnorm node anomali detect propos algorithm comput neighborhood node random walk restart graph partit propos algorithm identifi abnorm node neighborhood inform evalu qualiti neighborhood base semant dataset measur perform anomali detect algorithm manual inject anomali effect effici method confirm experi real dataset stream pattern discoveri multipl time dseri paper introduc spirit stream pattern discoveri multipl timeseri numer data stream valu observ time tick spirit increment find correl hidden variabl summaris kei trend entir stream collect relev search anomali detect bipartit graph real applic model bipartit graph user file system trader stock financi trade system confer author scientif public network introduc oper bipartit graph identifi similar node relev search find node connect irrelev node anomali detect propos algorithm comput relev score node random walk restart graph partit propos algorithm identifi anomali relev score evalu qualiti relev search base semant dataset measur perform anomali detect algorithm manual inject anomali effect effici method confirm experi real dataset queri past present futur spatio dtempor move object vehicl road network continu gener larg amount spatio tempor inform form data stream effici manag stream challeng goal due highli dynam natur data fast line comput paper present approach approxim queri process present past futur spatio tempor databas propos increment updat multi dimension histogram present time queri develop gener architectur maintain queri histor data implement stochast approach predict result queri refer futur final experiment prove effect effici techniqu realist simul queri past present futur spatio dtempor move object vehicl road network continu gener larg amount spatio tempor inform form data stream effici manag stream challeng goal due highli dynam natur data fast line comput paper present approach approxim queri process present past futur spatio tempor databas propos increment updat multi dimension histogram present time queri develop gener architectur maintain queri histor data implement stochast approach predict result queri refer futur final experiment prove effect effici techniqu realist simul select estim predict spatio dtempor queri paper propos cost model select estim predict spatio tempor window queri initi focu uniform data propos formula captur point rectangl type object queri mobil combin dynam object dynam queri appli model uniform dataset introduc spatio tempor histogram addit spatial veloc distribut partit advantag techniqu high accuraci order magnitud lower error previou techniqu ii abil handl queri type iii effici handl updat tpr dtree optim spatio dtempor access method predict queri predict spatio tempor queri retriev set move object intersect queri window futur time interv access method process queri practic tpr tree paper perform analysi determin factor affect perform predict queri show factor consid tpr tree insert delet algorithm tree design static data motiv propos index structur call tpr tree take account uniqu featur dynam object set improv construct algorithm addit provid cost model determin optim perform achiev data partit spatio tempor access method experiment comparison illustr tpr tree optim significantli outperform tpr tree condit analysi predict spatio dtempor queri paper present probabilist cost model estim select spatio tempor window queri join expect distanc queri nearest neighbor model captur queri object mobil combin move queri move object data type point rectangl arbitrari dimension addit develop special spatio tempor histogram account locat veloc inform increment maintain extens perform evalu verifi propos techniqu produc highli accur estim uniform uniform data index retriev histor aggreg inform move object spatio tempor databas store inform posit individu object time applic traffic supervis mobil commun system summar data averag number car area specif period phone servic cell dai requir inform obtain oper databas comput expens render onlin process inapplic vital solut construct spatiotempor data warehous paper describ framework support olap oper spatiotempor data argu spatial tempor dimens model combin dimens data cube present data structur integr spatiotempor index pre aggreg materi techniqu requir priori knowledg group hierarchi develop method util propos structur effici execut ad hoc group by techniqu static dynam dimens 