HP研究-社交媒体的热点话题的流行趋势

TrendsinSocialMedia:PersistenceandDecaySitaramAsurSocialComputingLabHPLabsPaloAlto,California,[e-mail]abHPLabsPaloAlto,California,[e-mail]tingLabHPLabsPaloAlto,California,[e-mail]UniversityCalifornia,[e-mail]neratesaprodigiouswealthofreal-timecontentatanincessantrate.Fromallthecontentthatpeoplecreateandshare,onlyafewtopicsmanagetoattractenoughattentiontorisetothetopandbecometemporaltrendswhicharedisplayedtousers.Thequestionofwhatfactorscausetheformationandpersistenceoftrendsisanimportantonethathasnotbeenansweredyet.Inthispaper,weconductanintensivestudyoftrendingtopicsonTwitterandprovideatheoreticalbasisfortheformation,persistenceanddecayoftrends.Wealsodemonstrateempiricallyhowfactorssuchasuseractivityandnumberoffollowersdonotcontributestronglytotrendcreationanditspropagation.Infact,wefindthattheres-onanceofthecontentwiththeusersofthesocialnetworkplaysamajorroleincausingtrends.1.INTRODUCTIONSocialmediaisgrowingatanexplosiverate,withmillionsofpeo-pleallovertheworldgeneratingandsharingcontentonascalebarelyimaginableafewyearsago.Thishasresultedinmassiveparticipationwithcountlessnumberofupdates,opinions,news,commentsandproductreviewsbeingconstantlypostedanddis-cussedinsocialwebsitessuchasFacebook,DiggandTwitter,tonameafew.Thiswidespreadgenerationandconsumptionofcontenthascre-atedanextremelycompetitiveonlineenvironmentwheredifferenttypesofcontentviewitheachotherforthescarceattentionoftheusercommunity.Inspiteoftheseeminglychaoticfashionwithwhichalltheseinteractionstakeplace,certaintopicsmanagetoattractaninordinateamountofattention,thusbubblingtothetopintermsofpopularity.Throughtheirvisibility,thispopulartop-icscontributetothecollectiveawarenessofwhatistrendingandattimescanalsoaffectthepublicagendaofthecommunity.Atpresentthereisnoclearpictureofwhatcausesthesetopicstobecomeextremelypopular,norhowsomepersistinthepubliceyelongerthanothers.Thereisconsiderableevidencethatoneaspectthatcausestopicstodecayovertimeistheirnovelty[11].Anotherfactorresponsiblefortheirdecayisthecompetitivenatureofthemedium.Ascontentstartspropagatingthroughtasocialnetworkitcanusurpthepositionsofearliertopicsofinterest,andduetothelimitedattentionofusersitissoonrenderedinvisiblebynewercontent.Yetanotheraspectresponsibleforthepopularityofcertaintopicsistheinfluenceofmembersofthenetworkonthepropaga-tionofcontent.Someusersgeneratecontentthatresonatesverystronglywiththeirfollowersthuscausingthecontenttopropagateandgainpopularity[9].Thesourceofthatcontentcanoriginateinstandardmediaoutletsorfromuserswhogeneratetopicsthateventuallybecomepartofthetrendsandcapturetheattentionoflargecommunities.Ineithercasethefactthatasmallsetoftopicsbecomepartofthetrendingsetmeansthattheywillcapturetheattentionofalargeaudienceforashorttime,thuscontributinginsomemeasuretothepublicagenda.Whentopicsoriginateinmediaoutlets,thesocialmediumactsasfilterandamplifierofwhatthestandardmediaproducesandthuscontributestotheagendasettingmechanismsthathavebeenthoroughlystudiedformorethanthreedecades[7].Inthispaper,westudytrendingtopicsonTwitter,animmenselypopularmicrobloggingnetworkonwhichmillionsofuserscreateandpropagateenormouscontentviaasteadystreamonadailyba-sis.Thetrendingtopics,whichareshownonthemainwebsite,rep-resentthosepiecesofcontentthatbubbletothesurfaceonTwitterowingtofrequentmentionsbythecommunity.Thustheycanbeequatedtocrowdsourcedpopularity.Wethendeterminethefac-torsthatcontributetothecreationandevolutionofthesetrends,astheyprovideinsightintothecomplexinteractionsthatleadtothepopularityandpersistenceofcertaintopicsonTwitter,whilemostothersfailtocatchonandarelostintheflow.Wefirstanalyzethedistributionofthenumberoftweetsacrosstrendingtopics.Weobservethattheyarecharacterizedbyastronglog-normaldistribution,similartothatfoundinothernetworkssuchasDiggandwhichisgeneratedbyastochasticmultiplicativeprocess[11].Wealsofindthatthedecayfunctionforthetweetsismostlylinear.Subsequentlywestudythepersistenceofthetrendstodeterminewhichtopicslastlongatthetop.Ouranalysisrevealsthattherearefewtopicsthatlastforlongtimes,whilemosttopicsbreakfairlyquickly,intheorderof20-40minutes.Finally,welookattheimpactofusersontrendpersistencetimeswithinTwitter.Wefindthattraditionalnotionsofuserinfluencesuchasthefrequencyofpostingandthenumberoffollowersarenotthemaindriversoftrends,aspreviouslythought.Rather,longtrendsarecharac-terizedbytheresonatingnatureofthecontent,whichisfoundtoarisemainlyfromtraditionalmediasources.Weobservethatsocialmediabehavesasaselectiveamplifierforthecontentgeneratedbytraditionalmedia,withchainsofretweetsbymanyusersleadingtotheobservedtrends.2.RELATEDWORKTherehasbeensomepriorworkonanalyzingconnectionsonTwit-ter.Hubermanetal.[5]studiedsocialinteractionsonTwittertorevealthatthedrivingprocessforusageisasparsehiddennetworkunderlyingthefriendsandfollowers,whilemostofthelinksrep-resentmeaninglessinteractions.Jansenetal.[6]haveexamined