张诗奇—社交传播和影响力算法在腾讯游戏中的应用实践.pdf

1、DataFunSummitDataFunSummit#20242024社交传播和影响力算法在腾讯游戏中的应用实践张诗奇 新加坡国立大学 博士将用户传播能力指标用于熟人推荐场景将影响力最大化算法用于病毒式营销场景提出影响力最大化新变种以更好适配活动推广场景目录 CONTENTINFORMATION DIFFUSION MEETS INVITATION MECHANISMShiqi Zhang,Jiachen Sun,Wenqing Lin,Xiaokui Xiao,Yiqian Huang,Bo TangPyr W i sINVITATION MECHANISMInvitation is alm

2、ost everywhere in Tencent gamesThe invitation behavior can cascadeThe histogram of diffusion depth of an in-game invitation event44%HOW TO MODEL INFORMATION DIFFUSION VIA INVITATION MECHANISM?Traditional diffusion models ignore the conversion funnel in invitation mechanism.Independent Cascade(IC):in

3、dependently influences with probability,social network0.10.10.10.40.60.60.80.30.30.20.50.70.40.30.60.50.4CONVERSION FUNNELA distillation of a users journey Describe how user behavior changes in multiple stagesAwareness:business outreachAdoption:platform activeness Action:e-commerce revenueEVERY STAG

4、E MATTERSINVITATION CONVERSION FUNNELConversion funnel of a userDear user:Your friend is inviting you to win the gifts together Accept?Target UninformedInvitedAcceptedInvitingICI:INDEPENDENT CASCADE WITH INVITATION MODELUser roles:Inactive state:uninformed(grey)Active states:inviter(red),invitee(yel

5、low),acceptor(orange)ICI:INDEPENDENT CASCADE WITH INVITATION MODELGiven the seeds,a diffusion instance unfolds in discrete stepsAt step 0all seeds initial invitersothers uninformedICI:INDEPENDENT CASCADE WITH INVITATION MODELGiven the seeds,a diffusion instance unfolds in discrete stepsAt the subseq

6、uent stepnew inviter has a probability,to invite the uninformed friend if becomes an invitee,it has a probability to be an acceptorif becomes an acceptor,it has a probability to be an inviterICI:INDEPENDENT CASCADE WITH INVITATION MODELGiven the seeds,a diffusion instance unfolds in discrete stepsSt