proceed acm confer inform knowledg manag phrase dbase method hierarch cluster web snippet learn concept drift data stream unlabel data automat categor algorithm web servic doubl dwindow dbase classif algorithm concept drift data stream pattern match flexibl wildcard recur charact onlin stream featur select studi interest challeng problem onlin stream featur select size featur set unknown featur learn leav number observ constant problem candid featur arriv time learner task select set featur stream featur standard featur select method perform scenario present framework base featur relev framework promis altern method onlin stream featur select osf present onlin select strongli relev redund featur addit osf faster fast osf algorithm propos improv select effici experiment result show algorithm achiev compact accuraci exist stream featur select algorithm dataset mine distribut chang stock order stream onlin stream featur select studi interest challeng problem onlin stream featur select size featur set unknown featur learn leav number observ constant problem candid featur arriv time learner task select set featur stream featur standard featur select method perform scenario present framework base featur relev framework promis altern method onlin stream featur select osf present onlin select strongli relev redund featur addit osf faster fast osf algorithm propos improv select effici experiment result show algorithm achiev compact accuraci exist stream featur select algorithm dataset sequenti pattern mine wildcard nettre approxim maxim pattern match gap doff constraint ensembl prune individu contribut order ensembl set learn model make decis collect ensembl accur singl learner exist ensembl method tend construct unnecessarili larg ensembl increas memori consumpt comput cost ensembl prune tackl problem select subset ensembl member form subensembl subject resourc consumpt respons time accuraci similar origin ensembl paper analyz accuraci divers trade prove classifi accur make predict minor group import subensembl construct base gain insight heurist metric consid accuraci divers propos explicitli evalu individu classifi contribut ensembl incorpor ensembl member decreas order contribut subensembl form user select top percent ensembl member depend resourc avail toler wait time predict experiment result uci data set show subensembl form propos epic ensembl prune individu contribut order algorithm outperform origin ensembl state ofth art ensembl prune method orient order oo matrix modular svm robust imbalanc data effici visual concept detect rule synthes multipl relat databas fast orthogon nonneg matrix tri dfactor simultan cluster oasnet optim alloc approach influenc maxim modular social network adapt inform sampl activ learn approach activ learn involv period train classifi choos data point lowest confid design confid measur nontrivi altern approach period choos data instanc maxim disagr label predict ensembl classifi classifi underli structur fit framework ensembl suitabl data set question aris find suitabl ensembl data set work introduc method begin heterogen ensembl compos multipl instanc classifi type call adapt inform sampl algorithm period add data point train set adapt ratio classifi type heterogen ensembl favor classifi type optim classifi ensembl stochast method experiment result show propos method perform consist homogen ensembl comparison random sampl uncertainti sampl show algorithm effect draw inform data point train discret method machin learn neural inspir algorithm limit pure form work nomin data real world problem provis made support process continu valu data paper present empir result obtain discret method preprocessor supervis learner real world problem discret techniqu outperform discret preprocess step case found inferior direct handl continu valu data result suggest machin learn algorithm design directli handl continu valu data reli preprocess ad hoc techniqu random decis tree ensembl mine concept drift noisi data stream exploratori analysi protein translat regulatori network hierarch random graph manifold elast net unifi framework spars dimens reduct nesvm fast gradient method support vector machin new filter summar web integr induct deduct noisi data mine special issu select paper ieee dmf modifi semi dsupervis learn algorithm laplacian eigenmap nonneg matrix factor orthogon subspac icdm abstract assist domain expert activ learn select construct fewer exampl request label build accur classifi previou work activ learn gener specif queri real world applic domain expert oracl readili answer gener queri don care attribut power gener queri gener queri equival specif overli gener queri good answer domain expert oracl highli uncertain make learn difficult paper propos activ learn algorithm ask good gener queri extend algorithm construct hierarch featur nomin numer attribut demonstr experiment method ask significantli fewer queri compar previou work activ learn initi label dataset small oracl inaccur class probabl estim method readili deploi real world data mine task obtain label exampl costli extens paper publish ieee icdm keyword activ learn domain expert gener queri icdm workshop hybrid artifici intellig system joint warfar system jwar equip command model cm perform situat assess action coa select command behavior model cbm bia decis command leadership style cm hybrid artifici intellig system model doctrin fuzzi rule set tree base lookahead algorithm strategi cbm emploi behaviorbas fuzzi rule set augment cm assess situat bias coa select criteria extend myer brigg person trait cbm link person trait militari attitud consequ valu emploi fuzzi rule set result set valu combin select specif coa audit trail user abil modifi input paramet underli rule cm cbm applic decis multipl echelon mine frequent pattern gap doff condit relat extract method chines name entiti base locat semant featur local semi dlinear regress river runoff forecast relat extract method chines name entiti base locat semant featur ubiquit person inform process wildcard activ learn adapt heterogen ensembl abstract common approach activ learn iter train singl classifi choos data point base uncertainti nontrivi design uncertainti measur unbias choic classifi queri committe suggest ensembl divers accur classifi inform data point maxim disagr predict ensembl member method find ensembl data set remain open question paper random subspac method combin activ learn creat multipl instanc classifi type algorithm introduc adapt ratio classifi type ensembl accuraci show propos algorithm outperform uncertainti sampl naiv bay uncertainti sampl bag boost random subspac method random sampl knowledg work adapt ratio classifi heterogen ensembl activ learn vagu dclass learn data stream multi dsphere support vector data descript outlier detect multi ddistribut data approxim repeat pattern mine gap requir error detect uncertainti model imprecis data cluster web servic automat categor pattern match wildcard base kei charact locat aggreg ensembl mine concept drift data stream nois paramet estimd semi drandom decis tree ensembl stream data compar studi pattern match algorithm sequenc bpbm algorithm string match wildcard length constraint mine global interest pattern multipl databas kernel estim agent dbase hybrid system microarrai data analysi optim servic system base applic dlevel qo optim servic system base applic dlevel qo geometr subspac select ubiquit mine interact data mine agent region categori pars undirect graphic model mine concept ddrift data stream multipl semi drandom decis tree intrus detect model base mine data stream descript cluster algorithm base nonneg matrix factor data mine algorithm problem busi process autom technolog increasingli compani improv effici intern process servic offer custom order satisfi custom employe busi process execut high predict qualiti crucial organ meet servic level agreement sla stipul custom forese earli risk miss sla order set expect correct action paper focu critic issu busi process qualiti analyz predict prevent occurr except deviat desir accept behavior character problem propos solut base data wareh mine techniqu describ architectur implement tool suit enabl except analysi predict prevent final show experiment result obtain tool suit analyz intern hp process cleans noisi data stream multilay chang dpoint detect stock order flow wavelet transform keyword extract base lexic chain word doccurr chines new web page gene express analys genet algorithm base hybrid approach bayesian tensor analysi keyphras extract chines new web page base semant relat keyphras extract chines new web page base semant relat automat recognit new web page error dtoler data mine data mine conceptu equival contrast mine classif learn learn occur compar classif learn order compar data group exist method compar raw instanc learn classif rule paper take approach conceptu equival group equival underli concept equival instanc space necessarili overlap rule set necessarili present appear methodolog compar propos learn represent group underli concept crossexam group instanc group concept represent innov fold quantifi degre conceptu equival group retrac sourc discrep level abstract level underli concept specif level instanc appli numer data categor data fourth circumv direct comparison possibl alarg number rule demand substanti effort reduc depend accuraci emploi classif algorithm empir evid suggest methodolog effect simpl scenario nois cleans concept chang learn bayesian tensor approach dd face model tkde decemb eic editori state transact editori tkde editori board mine nois knowledg error dawar data mine top algorithm data mine abstract paper present top data mine algorithm identifi ieee intern confer data mine icdm decemb mean svm apriori em pagerank adaboost knn naiv bay cart top algorithm influenti data mine algorithm research commun algorithm provid descript algorithm discuss impact algorithm review current research algorithm algorithm cover classif tensor rank discrimin analysi converg method discrimin multilinear subspac select proceed acm sigkdd intern confer knowledg discoveri data mine present famili algorithm uncov tribe group individu share unusu sequenc affili work infer commun structur describ larg scale trend search small group tightli link individu behav anomal respect trend appli algorithm larg tempor relat data set consist million employ record nation associ secur dealer result tribe individu higher risk fraud homogen respect risk score geograph mobil signific level compar random set individu share affili measur uncertainti differ contrast group sail dapprox effici dline algorithm approxim pattern match wildcard length constraint mine nois knowledg error awar data mine mine frequent itemset prune concept lattic discov relat pattern multipl databas probabilist tensor analysi akaik bayesian inform criteria gener averag diverg analysi nois model associ corrupt rule probabilist tensor analysi akaik bayesian inform criteria empir studi nois impact cost dsensit learn mine complex pattern sequenc gap requir mine frequent pattern wildcard biolog sequenc vari densiti spatial cluster base hierarch tree parallel algorithm learn bayesian network class nois attribut nois impact detect cleans algorithm attribut reduct decis tabl eic editori tkde editori board tkde guidelin survei paper gener tensor discrimin analysi gabor featur gait recognit real dtime textur synthesi dtile set semi drandom multipl decis dtree algorithm mine data stream mine maxim frequent itemset data stream supervis tensor learn paper aim gener tensor input supervis learn supervis tensor learn stl framework establish convex optim base learn techniqu support vector machin svm minimax probabl machin mpm stl framework convent learn machin gener order tensor input studi applic tensor learn machin design featur extract linear discrimin analysi lda method tensor base featur extract name tenor rank discrimin analysi tr da gener algorithm advantag reduc curs dimens problem machin learn data mine avoid failur converg achiev separ categori sampl gener mpm stl version name tensor mpm tmpm tmpm learn seri tensor project iter evalu origin mpm experi binari classif problem show tmpm significantli outperform origin mpm mine sequenti pattern time sequenc gener object recognit region statist model layer joint boost human carri statu visual surveil person gait carri object bag suitcas rucksack result human identif track made difficult averag gait imag simpl repres carri statu paper introduc set gabor base human gait appear model gabor function similar recept field profil mammalian cortic simpl cell high dimension featur space make train difficult order solv problem propos gener tensor discrimin analysi gtda seamlessli incorpor object gabor base human gait appear model structur inform natur constraint gtda differ previou tensor base discrimin analysi method train converg exist method fail converg train stage make unsuit practic task experi carri usf baselin data set recogn human id gait silhouett propos gabor gait incorpor gtda demonstr significantli outperform exist appear base method user dcenter agent structur inform locat web error awar data mine correct classif classifi ensembl correct divers base learner scalabl repres instanc select rank ontolog dbase featur weight biomed literatur classif triangul bayesian network adapt genet algorithm identifi bridg rule conceptu cluster svm dod svm method detect outlier mine anticip concept chang proactiv dreactiv predict data stream abstract predict stream data import activ modern societi major challeng pose data stream data grow limit difficult retain long histori raw data underli concept data chang time novelti paper fold measur conceptu equival organ data histori histori concept contrast common practic recent raw data concept histori compact retain essenti inform learn learn concept transit pattern concept histori anticip concept case concept chang proactiv prepar predict model futur chang contrast convent methodolog passiv wait chang incorpor proactiv reactiv predict anticip turn correct proper predict model launch instantli eic editori tkde editori board logic framework identifi qualiti knowledg data sourc eic editori state transact class nois handl effect cost dsensit learn cost dguid iter classif filter eic editori tkde editori board asymmetr bag random subspac support vector machin dbase relev feedback imag retriev challeng problem data mine research statist machin learn techniqu data classif assum entiti independ ident distribut real world entiti interconnect explicit implicit relationship form complex network graph base classif method emerg recent year suitabl complex network degre distribut network consider paper propos techniqu modular kernel effect exploit latent commun structur network entiti classif number experi hypertext dataset show propos approach lead excel classif perform comparison state art method effici string match wildcard length constraint supervis tensor learn paper aim gener tensor input supervis learn supervis tensor learn stl framework establish convex optim base learn techniqu support vector machin svm minimax probabl machin mpm stl framework convent learn machin gener order tensor input studi applic tensor learn machin design featur extract linear discrimin analysi lda method tensor base featur extract name tenor rank discrimin analysi tr da gener algorithm advantag reduc curs dimens problem machin learn data mine avoid failur converg achiev separ categori sampl gener mpm stl version name tensor mpm tmpm tmpm learn seri tensor project iter evalu origin mpm experi binari classif problem show tmpm significantli outperform origin mpm effect classif noisi data stream attribut dorient dynam classifi select abstract recent mine data stream import challeng task real world applic credit card fraud protect sensor network popular solut separ stream data chunk learn base classifi chunk integr base classifi effect classif paper propos dynam classifi select dc mechan integr base classifi effect mine data stream propos algorithm dynam select singl classifi classifi test instanc run time scheme statist inform attribut valu attribut partit evalu set disjoint subset procedur evalu classif accuraci base classifi subset test instanc attribut valu determin subset similar instanc evalu set construct classifi highest classif accuraci subset select classifi test instanc experiment result compar studi demonstr effici efficaci method dc scheme appear promis mine data stream dramat concept drift signific amount nois base classifi conflict low confid decrement algorithm maintain frequent itemset dynam databas abstract data mine machin learn confront problem pattern mainten data updat fundament oper data manag exist data mine algorithm assum databas static databas updat requir rediscov pattern scan entir data effici mine techniqu data addit databas paper propos decrement algorithm pattern discoveri data delet databas conduct extens experi evalu approach illustr propos algorithm model captur interact data data decreas supervis tensor learn paper aim gener tensor input supervis learn supervis tensor learn stl framework establish convex optim base learn techniqu support vector machin svm minimax probabl machin mpm stl framework convent learn machin gener order tensor input studi applic tensor learn machin design featur extract linear discrimin analysi lda method tensor base featur extract name tenor rank discrimin analysi tr da gener algorithm advantag reduc curs dimens problem machin learn data mine avoid failur converg achiev separ categori sampl gener mpm stl version name tensor mpm tmpm tmpm learn seri tensor project iter evalu origin mpm experi binari classif problem show tmpm significantli outperform origin mpm ac aggress classifi ensembl error detect scalabl induct learn partit data combin proactiv reactiv predict data stream mine data stream import scienc commerc major challeng data grow limit difficult retain long histori underli concept data chang time common practic recent raw data paper measur conceptu equival organ data histori histori concept journei concept chang identifi concept appear learn transit pattern concept predict convent methodolog passiv wait concept paper incorpor proactiv reactiv predict proactiv mode anticip concept futur concept chang take place prepar predict strategi advanc anticip turn correct proper predict model launch instantli concept chang promptli resort reactiv mode adapt predict model data system repro propos implement idea experi compar system repres exist predict method benchmark data set repres diversifi scenario concept chang empir evid demonstr propos methodolog effect effici solut predict data stream rearrang data object effici stabl cluster discret method machin learn neural inspir algorithm limit pure form work nomin data real world problem provis made support process continu valu data paper present empir result obtain discret method preprocessor supervis learner real world problem discret techniqu outperform discret preprocess step case found inferior direct handl continu valu data result suggest machin learn algorithm design directli handl continu valu data reli preprocess ad hoc techniqu eic editori tkde editori board video data mine semant index event detect associ perspect appendic video data mine semant index event detect associ perspect abstract advanc media entertain industri includ stream audio digit tv present challeng manag access larg audio visual collect current content manag system support retriev low level featur motion color textur low level featur mean naiv user prefer identifi content high level semant concept creat gap system user bridg system effect end paper present knowledg base video index content manag framework domain specif video basketbal video provid solut explor video knowledg mine associ video data explicit definit evalu measur tempor support confid video associ propos integr inher featur video data approach video process techniqu find visual audio cue court field camera motion activ applaus introduc multilevel sequenti associ mine explor associ audio visual cue classifi associ assign class label appear video construct video indic experiment result demonstr perform propos approach index term video mine multimedia system databas manag knowledg base system support vector machin base dmean cluster real dtime busi intellig system editori tkde topic area revis support vector machin base dmean cluster real dtime busi intellig system databas classif multi ddatabas mine mine frequent pattern pattern tree error detect impact dsensit instanc rank noisi dataset noisi dataset locat erron instanc attribut rank suspici instanc base impact system perform interest import research issu provid paper error detect impact sensit instanc rank edir mechan address problem noisi dataset train benchmark classifi instanc effect classifi treat suspici forward subset attribut switch class label train classifi ap instanc ap benchmark classifi locat erron attribut quantit rank instanc defin impact measur base inform gain ratio ir calcul ir attribut ir impact sensit weight sum impact sensit weight locat erron attribut total impact experiment result demonstr effect strategi intellig digit librari system biologist aid research obtain organ manag biolog data develop sophist digit librari system util advanc data mine techniqu stone al digit librari system implement central ee web applic link publicli access data repositori internet digit librari base framework convent librari objectori paradigm person user center servic base user area interest prefer make person servic user profil repres prefer individu user construct base user past activ goal user option util user profil system make relev inform user form amount level detail minim user effort core project agent architectur advanc servic combin data mine capabl domain knowledg form semant network stone al semant network impart knowledg structur system reason draw conclus biolog data object feder view dispar databas interest biologist develop semant network includ concept establish control vocabulari chief data mine artifici intellig data analysi cost dguid class nois handl effect cost dsensit learn recent research machin learn data mine relat area produc wide varieti algorithm costsensit cs classif maxim classif accuraci minim misclassif cost object method assum train set signific nois rare case real world environ paper systemat studi impact class nois cs learn propos cost guid class nois handl algorithm identifi nois effect cs learn call cost guid iter classif filter cicf seamlessli integr cost exist classif filter nois identif put equal weight handl nois class exist effort cicf put emphasi expens class make success deal dataset larg cost ratio experiment result compar studi real world dataset exist nois corrupt perform cs classifi adopt propos cicf algorithm significantli reduc misclassif cost cs classifi noisi environ dynam classifi select effect mine noisi data stream recent mine data stream import challeng task real world applic credit card fraud protect sensor network popular solut separ stream data chunk learn base classifi chunk integr base classifi effect classif paper propos dynam classifi select dc mechan integr base classifi effect mine data stream propos algorithm dynam select singl classifi classifi test instanc run time scheme statist inform attribut valu attribut partit evalu set disjoint subset procedur evalu classif accuraci base classifi subset test instanc attribut valu determin subset similar instanc evalu set construct classifi highest classif accuraci subset select classifi test instanc experiment result compar studi demonstr effici efficaci method dc scheme appear promis mine data stream dramat concept drift signific amount nois base classifi conflict low confid data acquisit activ impact dsensit instanc select oidm onlin interact data mine deal predict dbut dunpredict attribut noisi data sourc abstract attribut nois affect classif learn previou work handl attribut nois focus predict attribut predict class attribut attribut predict unpredict predict essenti classif learn import handl nois unpredict requir strategi predict attribut paper present studi identifi cleans measur nois predict unpredict attribut strategi propos theoret analysi empir evid suggest strategi effect effici previou altern data mine artifici intellig data analysi effici mine posit neg associ rule class nois attribut nois quantit studi data mine ai perspect paramet tune induct dalgorithm dorient featur elimin guest editor introduct mine action knowledg web editori introduc editor din dchief explor video content structur hierarch summar elimin class nois larg dataset paper present approach identifi elimin mislabel instanc larg distribut dataset partit dataset subset small process induct algorithm time construct good rule subset good rule evalu dataset instanc ik error count variabl count number time identifi nois subset instanc higher error valu higher probabl mislabel threshold scheme major object identifi nois experiment result compar studi real world dataset report evalu effect effici propos approach mine video associ effici databas manag support effici video databas manag paper explor concept video associ mine associ pattern character sequenti video shot cluster inform continu video sequenc video shot segment mechan introduc pars discret shot cluster shot visual distinct group construct shot cluster sequenc class label shot associ mine scheme design mine sequenti cluster sequenc detect associ convei valuabl knowledg video databas manag experiment result demonstr effect design multi ddatabas mine multi databas mine import research area urgent analyz data sourc essenti differ mono multi databas mine limit exist multi databas mine effort paper design multidatabas mine process research issu involv mine multi databas includ databas cluster local pattern analysi discuss synthes high dfrequenc rule data sourc onlin public reproduc transmit form mean permiss data mine research meet practic develop ieee intern confer data mine san jose california novemb decemb panel discuss data mine research meet practic develop motiv organ panel discuss provid advic industri peopl explor direct data mine develop hierarch video content descript summar unifi semant visual similar vide increasingli medium choic varieti commun channeca renneca primarili incr kw le vex ofneh ork multimest system kew ourhesw vide se provid summari tractabl format many xistingapproach limit xplore import low le ve lfe hxl hxl unit summar unfortunatew sew wstpk contew andstructur vide corr stp low le ve lfennww direww ve close stplxpk sce dese stp audio signal proceo drawback of xist mete instel unfold seold structur withinth videi low le ve unit addrei onlyth de import unitsetshhxn strate gybas low le ve lfe ture cannotb appli togehww videw providingusen ov ofth vide contex variousl ve summar eskl wst formor ficiep databas reaba val brows thispap prewpw ahiexwst xk vide contew detewst summarizationstr gysupport nove joint sent visual similaritystr gy todep wk vide contew ficiewk accuratet avid contep de scription ontolog isadoptel variousvid procehxlw teoc niqu util construct asewx st hpkw vide annotationfram work byint grate acquir conte de scription data hienlsthxhw vide contel structur construct groupm ging clustewsth final lay vide summari diffesh granulariti isassewpk assistus unfoldingth vide contei aprogr sive expesthw onrehwl ordvidek valid effect propos approach associ analysi scan databas mine frequent pattern fp tree avoid costli candid gener repeatedli occurr frequenc check support threshold achiev perform effici apriorilik algorithm databas scan fp tree time consum data ad exist databas scan need data exist data paper present data structur tree pattern tree techniqu tree scan databas obtain fp tree support threshold updat tree data scan data exist data scan mine posit neg associ rule paper describ techniqu call apac mine interest quantit associ rule larg databas effect mine rule apac emploi adjust differ analysi techniqu advantag requir user suppli threshold hard determin apac advantag user discov posit neg associ rule posit associ rule tell record attribut attribut neg associ rule tell record attribut attribut fact apac mine posit neg associ rule object meaning measur determin interesting rule make effect data mine task dynam refin classif rule knowledg discoveri larg databas call naiv bayesian classif make unrealist assumpt valu attribut independ class learn method remark success practic uniformli learn method boost gener method combin multipl classifi due yoav freund rob schapir paper show boost appli naiv bayesian classifi yield combin classifi representation equival standard feedforward multilay perceptron ancillari result naiv bayesian classif nonparametr nonlinear gener logist regress train algorithm boost naiv bayesian learn backpropag definit advantag boost requir linear time constant space hidden node learn increment start import realworld dataset method gener perform good publish result learn method unlik standard learn algorithm naiv bayesian learn boost logarithm time linear number parallel comput unit learn method highli plausibl computation model anim learn argument suggest plausibl behavior messag tcci chair web log data emb web user brows behavior web log discov pattern predict user futur request base current behavior web data complex due larg size sequenti natur past research propos method predict page visit base present visit pattern paper extend work discov pattern predict web page access occur method base extens associ rule classif method extend tradit associ rule includ tempor inform explicitli rule reason confid predict term tempor region compar method tempor event predict demonstr effect method empir realist web log explor tradeoff predict accuraci data mine time model knowledg anchor reduc cognit overhead proceed ieee intern confer data mine text cluster typic involv cluster high dimension space appear difficult regard virtual practic set addit effici runtim gener associ rule user dcenter agent structur inform locat larg scale data mine base data partit build intellig learn databas system induct deduct opposit oper data mine applic induct extract knowledg form rule decis tree exist data deduct appli induct result interpret data intellig learn databas ildb system integr machin learn techniqu databas knowledg base technolog start exist databas technolog perform induct deduct integr databas technolog induct machin learn deduct knowledgebas system plai kei role construct ildb system design effici induct deduct algorithm paper present system structur ildb system discuss practic issu ildb applic instanc select structur induct introduct past thirti year databas research evolv technolog wide build intellig learn databas system induct deduct opposit oper data mine applic induct extract knowledg form rule decis tree exist data deduct appli induct result interpret data intellig learn databas ildb system integr machin learn techniqu databas knowledg base technolog start exist databas technolog perform induct deduct integr databas technolog induct machin learn deduct knowledgebas system plai kei role construct ildb system design effici induct deduct algorithm paper present system structur ildb system discuss practic issu ildb applic instanc select structur induct introduct past thirti year databas research evolv technolog wide knowledg object model class defin gener frame special follow includ hole definit fill special wai special fill hole made incompat rule state precis explicit represent opposit question answer object model support type opposit windowtool previou section opposit base notion abstract loom partit class mainwindow subwindow class xwindow abstract class gener window kind opposit exist clear sibl object oppos opposit relat util section studi context make explicit represent opposit relat induct pre dprocess induct attribut elimin editori call amp calendareditori editori call amp calendareditori research develop knowledg discoveri data mine multi dlayer increment induct paper describ multi layer increment induct algorithm mlii link exist nonincrement induct algorithm learn increment noisi data mlii make oper data partit gener reduct gener learn set rule set exampl refin previou set rule achiev redescript oper call reduct set exampl set rule deriv set exampl describ behaviour rule set rule extract behavior exampl rule meta rule control previou rule order improv predict accuraci experiment result show mlii achiev signific improv exist nonincrement algorithm hcv experi paper term rule accuraci introduct exist machin learn algorithm gener distin interest discoveri increment web explor object dorient model multimedia document multimedia data readili varieti resourc web user rang naiv sophist select present data meaning applic delaunai mm framework queri present multimedia data store distribut data repositori includ web uniqu combin user defin layout ad hoc queri capabl enabl user tailor simpl layout virtual document compos retriev multimedia object paper focu object orient data model declar queri languag result queri dispar resourc integr form coher user defin document introduct databas manag system dbmss tradition enabl storag retriev present tabular dataset primarili profession applic recent phenomenon wide area select hyperdocu search interfac learn design knowledg gener rievl recurs induct learn hand gestur recognit lfa fast chain algorithm rule dbase system schedular schedul knowledg object studi causal discoveri weak link small sampl weak causal relationship small sampl size pose signific difficulti automat discoveri causal model observ data paper examin influenc weak causal link vari sampl size discoveri causal model experiment result illustr effect larger sampl size discov causal model reliabl relev strength causal link complex origin causal model present indic evid superior robust mml minimum messag length method standard signific test recoveri causal link compar result show mml ci mml causal induc causal discoveri system find model tetrad ii small sampl linear causal model experiment result reveal mml ci find weak link smaller sampl size tetrad ii hybrid interpret match multipl match induct sitehelp local agent help increment explor world wide web disorient cognit overhead hypertext system disorient cognit overhead interrel problem fact ultim limit us hypertext cognit overhead author aris viewpoint see author person sole respons insert link work includ link anchor work extern frame cross link anchor work author awar anchor work anchor make insert link problem number anchor hypertext univers reduc problem paper propos bi direct intellig anchor anchor knowledg concept potenti reader seek arriv anchor knowledg concept reader seek jump rule induct decis tree construct ritio rule induct bayesian discret real dvalu attribut discret real valu attribut nomin interv import area symbol induct system real world classif task involv symbol numer attribut supervis unsupervis discret method inform gain base method wide cite paper design discret method call bayesian discret compar perform method includ inform gain method implement hcv version test dataset bayesian discret result term predict accuraci introduct context rule induct decis tree construct deal continu attribut mean discret numer attribut number interv discret interv treat similar nomin valu induct deduct essenti aspect fuzzi interpret induct result appli rule induc train exampl test case demand action match singl match multipl match exist techniqu deal case exclus base probabl estim continu attribut space attribut discretis interv induct fuzzi interpret discretis interv deduct time valuabl paper introduc idea fuzzi border interpret discretis interv deduct time outlin result obtain hcv version softwar introduct knowledg discoveri databas kdd research frontier wu databas technolog machin learn techniqu sustain research recent year act link field offer dual knowledg object ion encapsul abstract principl captur inform ignor detail featur entiti relev understand knowledg object informatica comprehend entiti select part abstract consist data abstract procedur abstract procedur abstract found imper program languag form function procedur reduc complex program code ood data abstract carri definit abstract data type adt commonli call class type atkinson al adt defin term data item oper call method oop appli data item data adt modifi manipul method result notion encapsul lead lectur note machin learn hcv induct algorithm paper present heurist attribut base nois toler data mine program hcv version base newli develop extens matrix approach divid posit exampl pe specif class set intersect group adopt set strategi find heurist conjunct formula group cover group posit exampl neg exampl ne hcv induct algorithm adopt hcv version softwar find descript formula form variabl valu logic pe ne low order polynomi time induct time addit hcv induct algorithm paper outlin techniqu nois handl discret numer domain develop implement hcv version softwar perform comparison hcv version data mine algorithm id rule newid noisi continuo induct learn algorithm frontier prolog dbase present integr knowledg data 