搜索
InformationContagion:AnEmpiricalStudyoftheSpreadofNewsonDiggandTwitterSocialNetworksKris%na Lerman Rumi Ghosh USC Informa,on Sciences Ins,tute USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 2Online Social Networks Online social networks have become important channels for the spread of ,mely and relevant informa,on powered by TouchGraph USCInformationSciencesInstituteInforma%on flow on networks hGp://@N03/4639307923 3Network 2 Network 1 @N03/4639307471/USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 4Dynamics of Social Informa%on USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 5Social news: Digg Users submit and vote for (digg) news stories Users join networks to see • Stories friends submit • Stories friends vote for Digg features stories with most votes on its front page USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 6Social news: TwiDer + Tweetmeme Users tweet and retweet* URLs to news stories *‘Retweet’ = tweet someone else’s post “RT @x failed bomb plot hGp://bit.ly/xmas09” Users join networks to see • Tweets by users they follow • Retweets by users they follow Tweetmeme aggregates all tweets and features most retweeted URLs on its front page USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 7Social news: Data sets Digg • Stories • 3,553 stories, promoted in June, 2009 • Time submiGed, promoted • Votes for each story • Time of the vote • Name of voter • Ac,ve users • 139,409 who voted for at least one story • 71,834 of them following at least one user • 258,220 links • fan network TwiGer • Stories • 398 most retweeted stories 6/11/09—7/3/09 • extracted from Tweetmeme • Retweets of each story • 1000 most recent retweets • Time of retweet & user name • Ac,ve users • 137,582 who retweeted at least one story • Following/follower rela,ons through TwiGer API • 6,200,051 links • follower network USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 8Ques%ons USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 9Basic terms SubmiGer • user who submiGed link to a story • user who first tweeted link to a story Vote • digg • retweet Fan of user A • user watching A’s ac,vity on Digg • user following A on TwiGer USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 10User ac%vity: distribu%on of fans TwiGer Digg • Typical number of followers on TwiGer ~10, but can be millions • No typical number of fans on Digg – “long tail” USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 11User ac%vity: distribu%on of vo%ng TwiGer • “Long tail” distribu,on of user ac,vity • Difference in slope related to effort of ac,vity [cf Wilkinson 2008] Digg USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 12Dynamics of stories Digg TwiGer 1: U.S. Government Asks TwiGer to Stay Up for #IranElec,on 2: Western Corpora,ons Helped Censor Iranian Internet 3: Iranian clerics defy ayatollah, join protests 1: US gov asks twiGer to stay up 2: Iran Has Built a Censorship Monster with help of west tech 3: Clerics join Iran’s an,‐government protests ‐ CNN.com USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 13Dynamics of stories Digg TwiGer • Two dis,nct phases for Digg stories: upcoming and promoted • Number of votes on both sites saturates aver one day • Saturated value reflects story popularity USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 14Popularity of stories Digg TwiGer lognormal fit lognormal fit • Aggregate over all stories to factor out influence of submiGer and story quality • “Inequality of popularity” – some stories much more popular than others cf social influence study of [Salganik, Dodds & WaGs, 2006] USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 15Informa%on flow on networks Informa,on spreads on a network as fans (followers) vote for (retweet) stories their friends submit or vote for. … fan votes – i.e., votes from fans USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 16Dynamics of informa%on spread on networks Digg TwiGer fanvotesfollowerretweets• Evolu,on of fan votes qualita,vely similar to that of all votes USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 17How far does informa%on spread on networks? Digg TwiGer normal fit normal fit USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 18How far does informa%on spread among submiDer’s fans? number of retweets from submiDer’s followers Digg TwiGer USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 19How fast does informa%on spread on networks? • Two dis,nct phases on Digg: stories spread faster through the network before promo,on than averwards • TwiGer stories spread at a uniform rate, but with greater variability USCInformationSciencesInstituteKris,na Lerman (CC) ‐ ICWSM 2010 20How fast does informa%on spread on networks? Probability next vote is from a fan