discov user topic interest twitter leverag contextu inform explor post link behavior blogger effici sequenti cover algorithm explain subset data rank techniqu cluster base search result textual knowledg dbase paper present framework methodolog improv search experi digit librari system approach cluster textual knowledgebas multipl relat return search result form small focus cluster specif gener multipl relationship network relationship type cluster network search time present rank set cluster rank relationship type intuit approach return cluster contextu relat inform user situat contextu awar search result return rank list document match queri address implicit textual content explicit citat author relat document primari question focu rank cluster search queri explor approach text base rank text similar user queri social network base rank inform central comparison rank method suggest inform central rank rank cluster document document character cluster highest rank workshop summari fourth workshop evalu method machin learn workshop summari fourth workshop evalu method machin learn graph dbase metric empir risk minim speed activ learn network data activ semi supervis learn import techniqu label data scarc recent method suggest combin activ learn semi supervis learn algorithm gaussian field harmon function classifi relat natur reli data present partial label graph network learn problem work show empir risk minim erm method find instanc label provid effici comput erm semisupervis classifi comput problem erm reli comput risk instanc limit candid investig speed activ learn consider case data graphic natur leverag graph structur rapidli identifi instanc good candid label paper describ hybrid approach commun find social network analyt central measur identifi good candid label erm find instanc candid set show real world data limit erm comput fraction instanc compar perform graph mine graph pattern profil improv learn network data combin explicit mine link aaai workshop report roc confid band empir evalu paper construct confid band roc curv introduc machin learn commun band gener method medic field evalu perform confid band repres region true roc curv expect resid design confid level assess contain band begin synthet world true roc curv specif class condit model score distribut method attain reason contain box produc parametr fix width band fwb move context machin learn evalu band confid level bound perform model futur data introduc correct account larger uncertainti widen fwb continu reason contain final assess band larg benchmark data set conclud recommend fwb note parametr attract machin learn studi score distribut normal data set vari substanti learn method learn method confid band roc curv method empir studi ad number text classif convert numer classif text classif supervis learn problem exampl numer text valu featur common approach problem treat presenc absenc word boolean featur combin numer featur enabl applic rang tradit featur vector base learn method paper present altern approach numer featur convert bag word featur enabl rang exist text classif method approach creat set bin featur observ valu fall token defin bin endpoint repres side bin endpoint featur li numer featur assign bag token approach make appli text classif method problem involv numer text valu featur problem sole numer featur convert represent text classif method appli evalu approach rang real world dataset uci repositori sole involv numer featur addit dataset numer text valu featur result show perform text classif method bin represent meet exce tradit supervis learn method nn nbc ripper exist numericalfeatur dataset uci repositori suggest text classif method coupl bin serv credibl learn approach tradit supervis learn problem predict citat rate physic paper construct featur order probit model text classifi numer classif supervis learn problem exampl numer text valu featur tradit featurevector base learn method treat presenc absenc word boolean featur binari valu featur numer featur text classif system bit problemat straight forward approach number consid distinct token treat word paper present altern approach text classif method supervis learn problem numer valu featur numer featur convert bag word featur make directli usabl text classif method show pure numer valu data result textclassif deriv text represent outperform naiv number token represent importantli competit matur numer classif method ripper intellig inform triag applic larg volum time sensit textual inform requir triag rapid approxim priorit subsequ action paper explor prospect indic import time sensit document purpos produc document filter rank prospect import assess action occur futur new stori assess retrospect import base event occur stori appear stock price plummet issuanc follow stori system anticip prospect occurr provid time indic import perfect prescienc imposs su cient correl content inform item event occur subsequ describ process creat evalu approxim inform triag procedur base prospect indic unlik informationretriev applic document label labori manual process prospect criteria build larg label train corpora automat corpora train text classif procedur predict prospect import document paper illustr process case studi demonstr abil predict new stori similar new stori stock price compani new stori move significantli appear stori conclud discuss comprehens learn classifi ca human perform cluster web page preliminari studi increas inform world wide web difficult quickli find desir inform multipl queri topic specif search engin search group html page relat order understand task perform initi studi human cluster web page hope provid insight difficulti autom task result show subject cluster ident fact averag subject similar web page cluster found subject gener creat small cluster access url creat fewer cluster access full text web page gener overlap document cluster subject increas full text percentag document cluster analyz individu subject found behavior queri term overlap size cluster number cluster result provid sober note quest singl correct cluster method web page maintain inform resourc prolifer world wide web import provid advanc tool maintain referenti integr inform resourc grow tendenc build increasingli complex web site make maintain physic file logic resourc view compos refer resourc present program solut problem design infrastructur resourc mainten agent includ data agent track file support parti request notifi occur file compon infrastructur repositori agent support chang notif request logic resourc prototyp implement infrastructur discuss paper 