increment collabor filter evolutionari dcluster collabor filter euclidean embed detect wikipedia vandal activ learn statist languag model paper propos activ learn approach languag model statist detect wikipedia vandal wikipedia popular influenti collabor inform system collabor natur author high visibl content expos wikipedia articl vandal vandal defin malici edit intend compromis integr content articl extens manual effort made combat vandal autom approach allevi labori process need paper build statist languag model construct distribut word revis histori wikipedia articl vandal involv unexpect word draw attent fit lack thereof edit compar languag model built previou version edit vandal instanc addit paper adopt activ learn model solv problem noisi incomplet label wikipedia vandal wikipedia domain revis histori offer context explor potenti languag model character author intent experiment result present paper demonstr model hold promis vandal detect decis support system cost deffect diagnosi find maxim fulli dcorrel itemset larg databas bag adapt cost abstract ensembl method prove highli effect improv perform base learner circumst paper propos algorithm combin merit exist techniqu bag arc stack basic structur algorithm resembl bag misclassif cost train point repeatedli adjust observ bag vote margin method gain advantag arc build classifi ensembl fixat potenti noisi point comput experi show algorithm perform consist bag arc linear nonlinear base classifi view characterist bace hybrid ensembl learn strategi combin bag version bace propos studi empir index term data mine ensembl method yi zhang nick street member ieee build hospit referr expert system predict optim dbase decis support system algorithm studi present method construct expert system hospit referr problem factor institut characterist patient risk travel distanc chanc surviv complic includ hospit select decis ideal patient treat individu decis process includ con dition belief trade off desir hospit featur expert system complex decis numer factor consid propos method call predict optim base deci sion support system podss algorithm construct expert system explicit knowledg base algorithm obtain knowledg build machin learn classifi collect label case respons queri algorithm custom recommend optim step patient maxim probabl achiev desir case recommend hospit optim solut maxim probabl desir outcom proper formul expert system combin multipl factor give hospit select decis support individu level learn rank maxim auc linear program abstract area roc curv auc evalu rank perform binari classif problem research approach auc optim approxim equival wicoxon mann whitnei wmw statist present linear program approach similar norm support vector machin svm instanc rank approxim wmw statist formul appli nonlinear problem kernel function rank algorithm outperform svm auc classif perform rbf kernel curious polynomi kernel experi variat chunk handl quadrat growth number constraint formul hac heurist algorithm cluster subset learn rank maxim auc linear program abstract area roc curv auc evalu rank perform binari classif problem research approach auc optim approxim equival wicoxon mann whitnei wmw statist present linear program approach similar norm support vector machin svm instanc rank approxim wmw statist formul appli nonlinear problem kernel function rank algorithm outperform svm auc classif perform rbf kernel curious polynomi kernel experi variat chunk handl quadrat growth number constraint formul ensembl prune semi ddefinit program ensembl group learn model jointli solv problem ensembl gener exist techniqu unnecessarili larg lead extra memori usag comput cost occasion decreas effect purpos ensembl prune search good subset ensembl member perform origin ensembl subset select problem combinatori optim problem find exact optim solut computation prohibit heurist method develop obtain approxim solut exist heurist simpl greedi search optim method lack theoret empir qualiti guarante paper ensembl subset select problem formul quadrat integ program problem appli semi definit program sdp solut techniqu approxim solut comput experi show sdp base prune algorithm outperform heurist literatur applic classifi share studi demonstr effect method share classifi ensembl relat problem domain classif ensembl group classifi solv predict problem wai combin predict classifi problem domain techniqu bag boost improv perform research show share classifi close relat problem domain help addit semi definit program base ensembl prune method implement order optim select subset classifi problem domain comput result catalog dataset ensembl result share classifi product categori gener larger auc ensembl train categori prune algorithm prevent occasion decreas effect caus conflict concept problem domain understand problem domain relationship bag adapt cost abstract ensembl method prove highli effect improv perform base learner circumst paper propos algorithm combin merit exist techniqu bag arc stack basic structur algorithm resembl bag misclassif cost train point repeatedli adjust observ bag vote margin method gain advantag arc build classifi ensembl fixat potenti noisi point comput experi show algorithm perform consist bag arc linear nonlinear base classifi view characterist bace hybrid ensembl learn strategi combin bag version bace propos studi empir index term data mine ensembl method yi zhang nick street member ieee obliqu multicategori decis tree nonlinear program inform doi ijoc inform induct decis tree popular effect method solv classif problem data mine applic paper present algorithm multi categori decis tree induct base nonlinear program algorithm term oc sep obliqu categori separ combin advantag method show improv gener perform collect real world data set custom target neural network approach guid genet algorithm inform doi mnsc inform kei problem databas market identif profil household interest product servic princip compon analysi pca custom background inform logist regress analysi respons behavior commonli databas market paper propos approach artifici neural network ann guid genet algorithm ga target household show result select rule accur parsimoni pca logit rule manag clear decis criterion vagu decis criteria procedur lose advantag interpret accur pca logit target household kei word databas market neural network genet algorithm custom relationship manag intellig system custom target data mine approach cluster ddriven refin content dbase digit imag retriev model dbase detect comput aid mammograph prompt system requir reliabl detect varieti sign cancer paper concentr detect spicul lesion mammogram spicul lesion typic characteris abnorm pattern linear structur central mass statist model develop describ detect aspect spicul lesion describ gener method repres pattern linear structur reli factor analysi separ systemat random aspect class pattern model appear central mass local scale orient signatur base recurs median filter approxim princip compon analysi lesion mm larger pattern detect techniqu result sensit fals posit imag whilst mass detect approach result sensit fals posit imag simp intellishopp proactiv person privat shop assist intellishopp shop assist design empow consum person assist observ user shop learn prefer respect featur character shop item proactiv rememb user request autonom monitor vendor site item match user prefer final protect user privaci mean pseudonym ip anonym trust filter pseudonym achiev persona show approach behoov success classif ip anonym perform manner discuss compar context applic trust filter oppos merchant base filter improv privaci allow user select prefer privaci repres paper introduc intellishopp system discuss architectur compon describ prototyp implement outlin preliminari evalu perform adapt assist custom dshop shop onlin confus activ due vendor inconsist inform di erent vendor site organ shop agent base user center busi model unsuccess past shopper perceiv price comparison servic worth pai result shop bot todai bias favor vendor pai fee list make user center busi model viabl shop agent end describ shop assist capabl learn heterogen dynam prefer shopper observ behavior unobtrus present shopper adapt custom inform item discov autonom monitor vendor site evolutionari model select unsupervis learn featur subset select import insight gain determin relev model variabl improv understand scalabl possibl accuraci result model featur select tradition studi supervis learn situat estim accuraci evalu candid subset appli supervis learn lack train signal case propos featur select approach base cluster number heurist criteria estim qualiti cluster built featur subset combin criteria elsa evolutionari local select algorithm maintain divers popul solut approxim pareto front multi dimension object space evolv solut repres featur subset number cluster repres cluster algorithm mean em appli form number cluster base select featur experiment result real synthet data show method consist find approxim pareto optim solut identifi signific featur number cluster result model clearer semant relev increment featur weight learn applic shape dbase queri system similar shape measur comput distanc featur vector featur space captur notion similar human percept current imag retriev system weight measur import featur similar vari equal strength proport direct featur space paper present featur weight base cluster object databas relev feedback show varianc inform shape cluster guid cluster inform initi databas search result standard euclidean distanc automat incorpor user propos shape base queri system increment featur weight learn method ne prototyp contrast exist imag databas system system learn user feedback index retriev result present demonstr ecaci techniqu columbia databas stream ensembl algorithm sea larg dscale classif classif automat featur mine person digit imag retriev intellig object dbase imag retriev cluster ddriven person prefer learn featur select unsupervis learn evolutionari search featur subset select import problem knowl edg discoveri insight gain deter mine relev model variabl improv understand scalabl possibl accuraci result model paper problem featur select unsupervis learn number heurist criteria estim qualiti cluster built featuresubset combin criteria elsa evolutionari lo cal select algorithm maintain divers popul solut approxim pareto front multi dimension objectiv espac evolv solut repr sent featur subset number cluster standard mean algorithm appli form umber cluster base select featur preliminari result real synthet data show promis find pareto optim solut identifi signific featur correct number cluster effici scalabl pareto optim evolutionari local select algorithm local select simpl select scheme evolutionari comput individu fit accumul time compar fix threshold decid reproduc local select coupl fit function stem consumpt finit share environment resourc maintain divers similar fit share effici fit share lend parallel implement distribut task local select prone prematur converg appli minim select pressur popul neural network model prognost predict import difficult predict task domain medic decis make prognosi prognosi present uniqu set problem learn system output unknown paper present approach prognost predict idea nonparametr statist fulli util inform neural architectur techniqu appli breast cancer prognosi result flexibl accur model plai role prevent unnecessari surgeri introduct paper appli artifici neural network classif analysi surviv lifetim data lee object broadli defin predict futur time event work concern specif prognosi predict diseas method appli breast cancer prognosi predict long surgeri expect disea neural network model prognost predict import difficult predict task domain medic decis make prognosi prognosi present uniqu set problem learn system output unknown paper present approach prognost predict idea nonparametr statist fulli util inform neural architectur techniqu appli breast cancer prognosi result flexibl accur model plai role prevent unnecessari surgeri introduct paper appli artifici neural network classif analysi surviv lifetim data lee object broadli defin predict futur time event work concern specif prognosi predict diseas method appli breast cancer prognosi predict long surgeri expect disea featur select mathemat program problem discrimin finit point set dimension featur space separ plane util featur formul mathemat program parametr object function linear constraint step function appear object function approxim sigmoid concav exponenti nonneg real line treat equival linear program equilibrium constraint lpec comput test approach publicli real world databas carri compar adapt optim brain damag obd method reduc neural network complex featur select algorithm concav minim fsv reduc cross valid error cancer prognosi databas reduc problem featur featur select import problem machin learn cluster concav minim problem assign point dimension real space cluster formul determin center sum distanc point nearest center minim polyhedr distanc problem formul minim piecewis linear concav function polyhedr set shown equival bilinear program minim bilinear function polyhedr set fast finit median algorithm consist solv linear program close form lead stationari point bilinear program comput test number realworld databas carri wisconsin diagnost breast cancer wdbc databas median train set correct compar algorithm test set correct addition wisconsin prognost breast cancer wpbc databas distinct clinic import surviv curv induct learn approach prognost predict paper introduc recurr surfac approxim induct learn method base linear program predict recurr time censor train exampl exampl train output lower bound answer approach augment featur select method choos featur set context linear program gener comput result field breast cancer prognosi shown straightforward translat predict method artifici neural network model propos introduct machin learn method successfulli appli analysi complex problem recent year includ biomed applic field benefit type approach analysi surviv lifetim data lee miller jr object broadli defin predict 