analysis of tweets and tweeters during the 2014 umbrella ... · analysis of tweeters: bots and...
TRANSCRIPT
AnalysisofTweeters:BotsandHumansØ Whoarethetoptweeters?Ø Whatistherela/onshipbetweenthetoptweeters?Ø Whatisthecontribu/onofbotstotheamountoftweets?
Tweet-retweetgraphamongtop3000tweeters(overhalfofthemarebots).Le@:topretweetedusers(majornewsmedia)Right:topretwee/ngusers(bots)
Top3000tweetersgenerate40%oftweets
Graphoffollower-followingrela/onshipscollectedfromtweetsgeneratedat4pm,11-15-2014.AllareElUniversalbotgroupsthatfolloweachother.
Graphoffollower-followingrela/onshipscollectedfromtweetsgeneratedat10am,10-23-2014.AllareFalconsNewsbotgroupsthatfolloweachother.
Tweet-retweetvolumebeforeanda@erbotremoval.Notethesignificantreduc/oninnoisepeaks.
Conclusions ReferencesZiChu,StevenGianvecchio,HainingWang,andSushilJajodia. 2012. Detec/ng Automa/on of TwiYerAccounts: Are You a Human, Bot, or Cyborg?. IEEETrans. Dependable Secur. Comput. 9, 6 (November2012),811-824.Zhao,Siqi,etal."Humanasreal-/mesensorsofsocialand physical events: A case study of twiYer andsportsgames."arXivpreprintarXiv:1106.4300(2011Stockmann, Daniela, Big Data from China and ItsImplica/on for the Study of the Chinese State -- AResearch Report on the 2014Hongkong Protests onWeibo(May26,2015).
� 90%-95%oftweetsareaboutnewseventsorareretweetsofnewsevents.Only5%-10%expressopinionsabouttheevents.
� Top3000tweeters(outof240,000users)generate40%oftweets.Overhalfofthemarebots.Botgroupshaveinteres/nginternalstructures.
� Sen/mentanalysisoftweetsrevealsthelackofsustainedposi/vefeelingsabouttherevolu/on.
Ø Wecollected2,915,037tweetsabouttheHongKongRevolu/onfrom10/05/2014to12/06/2014.
Ø Goals:Understandthedynamicsoftherevolu/onusingtweetsalone.
Introduc;on
Contact
+ComputerScienceDepartment,RiceUniversity,*Poli/calScienceDepartment,RiceUniversity
ZhouhanChen+([email protected]),YunZhou+,RimaTanash+,JieyaWen+,DanWallach+,RichardStoll*,DevikaSubramanian+
AnalysisofTweetsandTweetersduringthe2014UmbrellaRevolu;on
DataCollec;on
Ø AnalyzepeaksinTwiYervolume.Ø UseTwiYerAPItobuildfollower-followingand
tweet-retweetgraphs.Ø Detectbotandhumangroupswithcommunity
detec/onalgorithmsonthosegraphs.Ø Interpretgraphswithvisualiza/ontoolssuchas
Gephi.Ø Applynaturallanguageprocessing(NLP)and
machinelearningforsen/mentanalysisontweets.
Ø Tweetscollectedfrom:TwiYerStreamingAPIØ Numberofprotest-relatedtweets:1,062,606Ø Numberofgeo-enabledtweets:3583
Methods
According to TwiYer, loca%on is “The user-definedlocaCon for thisaccount’sprofile”.Themapshowsthat many people from all over the world werephysicallypresentinHongKongduringtheprotest.
Ø Millionsofrawtweets.Ø Manytoptweetersarebots,introducingnoise
inouranalysis.Ø Rela/onshipbetweentweetersiscomplex.
Challenges
AnalysisofTweets:EventsandSen;ment
Ø Whataretweetsabout?Ø Whatistherela/onshipbetweentweetvolume
andmajorevents?Ø Howdopeoplefeelabouttheevents?
Exampleposi/vetweet:“SomethingaboutthisjustgivesmealiIlemorefaithintheworld.HongKongprotestersaresofreakingnice.”Examplenega/vetweet:“EveryCmeIhearabouttheUmbrellaRevoluConIwanttogroanforafewyears.”