epluribu ethnic social network find improv geograph predict social spatial proxim meme dtrack dynam new cycl track topic idea meme web issu consider interest recent work develop method track topic shift long time scale abrupt spike appear name entiti approach suit identif content spread wide fade time scale order dai time scale perceiv new event develop framework track short distinct phrase travel intact line text develop scalabl algorithm cluster textual variant phrase identifi broad class meme exhibit wide spread rich variat daili basi princip domain studi show meme track approach provid coher represent new cycl daili rhythm new media long subject qualit interpret captur accur permit actual quantit analysi track million mainstream media site blog period month total million articl find set persist tempor pattern new cycl observ typic lag hour peak attent phrase new media blog diverg behavior peak heartbeat pattern handoff new blog develop analyz mathemat model kind tempor variat system exhibit supervis random walk predict recommend link social network meme dtrack dynam new cycl track topic idea meme web issu consider interest recent work develop method track topic shift long time scale abrupt spike appear name entiti approach suit identif content spread wide fade time scale order dai time scale perceiv new event develop framework track short distinct phrase travel intact line text develop scalabl algorithm cluster textual variant phrase identifi broad class meme exhibit wide spread rich variat daili basi princip domain studi show meme track approach provid coher represent new cycl daili rhythm new media long subject qualit interpret captur accur permit actual quantit analysi track million mainstream media site blog period month total million articl find set persist tempor pattern new cycl observ typic lag hour peak attent phrase new media blog diverg behavior peak heartbeat pattern handoff new blog develop analyz mathemat model kind tempor variat system exhibit map world photo microscop evolut social network present detail studi network evolut analyz larg onlin social network full tempor inform node edg arriv time larg scale studi individu node arriv edg creation process collect lead macroscop properti network methodolog base maximum likelihood principl investig wide varieti network format strategi show edg local plai critic role evolut network find supplement earlier network model base inher local preferenti attach base observ develop complet model network evolut node arriv prespecifi rate select lifetim node independ initi edg gap process select destin edg simpl triangl close model free paramet show analyt combin gap distribut node lifetim lead power law degre distribut accur reflect true network case final give model paramet set automat evolut gener realist synthet network arbitrari scale preferenti behavior onlin group abstract paper present horizont view social influenc specif quantit studi influenc neighbour probabl node join group popular onlin social network osn orkut youtub livejourn flickr neighbour osn mutual acknowledg relat defin friendship directli connect graph social network user osn join group user group repres common area interest present simpl social influenc model describ explain group join process user osn end extract social influenc data set osn million sampl node find set neighbour osn time power influenc user join group number stranger keyword social network social influenc spatial variat search engin queri local aspect web search associ web content queri geographi topic grow interest underli question spatial variat manifest search queri understood develop probabilist framework quantifi spatial variat complet yahoo queri log find model local larg class queri mile natur center base distribut activ queri model estim queri geograph center measur spatial dispers indic highli local interest broader region nation appeal show variat model track geograph shift topic time annot map locat distinct queri delin sphere influenc compet queri gener domain wherefor art thou anonym social network fs algorithm featur select cascad neural network group format larg social network membership growth evolut process commun attract member develop time central research issu social scienc polit movement profession organ religi denomin provid fundament exampl commun digit domain line group increasingli promin due growth commun social network site myspac livejourn challeng collect analyz larg scale timeresolv data social group commun left basic question evolut group larg unresolv structur featur influenc individu join commun commun grow rapidli overlap pair commun chang time address question larg sourc data friendship link commun membership livejourn authorship confer public dblp dataset provid explicit user defin commun confer serv proxi commun dblp studi evolut commun relat properti structur underli social network find propens individu join commun commun grow rapidli depend subtl wai underli network structur tendenc individu join commun influenc number friend commun crucial friend kdd dcup result analysi paper summar analyz result kdd cup competit consist task area particl physic protein homolog detect focus problem optim supervis learn perform measur accuraci cross entropi roc area slac squar error averag precis top rank total group particip competit receiv award honor mention approach paper issu sigkdd explor paper analyz approach give insight perform field competitor studi fraction particip found good solut particip optim perform measur homogen submit predict submiss repres maxim perform achiev keep kdd cup www site open ad automat score system submiss order encourag research area 