Web-Scale User Modeling for Targeting

Web-Scale User Modeling for Targeting

ID:39776471

大小:448.24 KB

页数:10页

时间:2019-07-11

Web-Scale User Modeling for Targeting_第1页
Web-Scale User Modeling for Targeting_第2页
Web-Scale User Modeling for Targeting_第3页
Web-Scale User Modeling for Targeting_第4页
Web-Scale User Modeling for Targeting_第5页
资源描述:

《Web-Scale User Modeling for Targeting》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库

1、Web-ScaleUserModelingforTargetingMohamedAly,AndrewHatch,VanjaJosifovski,VijayK.NarayananYahoo!ResearchSantaClara,CA95051,USA{aly,aohatch,vanjaj,vnarayan}@yahoo-inc.comABSTRACTpossible.Intheonlineenvironment,adplatformshavetra-ditionallyallowedadvertiserstotargetusersinpopulationsWepresentt

2、heexperiencesfrombuildingaweb-scaleuserthatarelikelytorespondpositivelytotheprovidedadvertis-modelingplatformforoptimizingdisplayadvertisingtar-ing.TraditionallythemethodstospecifyuserpopulationsgetingatYahoo!.Theplatformdescribedinthispaperal-relyondemographicsorlocationinfo,wheretheadver

3、tiserlowsforper-campaignmaximizationofconversionsrepre-choosesasetoflocationsandcertaindemographicattributessentingpurchaseactivitiesortransactions.Conversionsdi-toselectthepopulationsegmenttoseethead.Thedemo-rectlytranslatetoadvertiser'srevenue,andthusprovidethegraphicsandlocationattribut

4、escaneitherbeobtainedfrommostrelevantmetricsofreturnonadvertisinginvestment.registrationdatabasesorinferredfrompastuseractivity.Wefocusontwomajorchallenges:howtoecientlypro-Anothertraditionalonlinetargetingmethodthatusesthecesshistoriesofbillionsofusersonadailybasis,andhowtopastuseractivi

5、tyisbehavioraltargeting.Here,insteadofbuildper-campaignconversionmodelsgiventheextremelybucketingtheusersintodemographics'categories,usersarelowconversionrates(comparedtoclickratesinatraditionalputintoprede nedinterestcategoriesasparenting,auto,setting).We rstpresentmechanismsforbuildingwe

6、b-scalehealth,whicharethenpurchasedbytheadvertisers.userpro lesinadailyincrementalfashion.Second,weshowOneweaknessofthedemographicandbehavioraltarget-howtoreducethelatencythroughin-memoryprocessingofingisthatthesegmentsofusersarenotcreatedusinganybillionsofuserrecords.Finally,wediscussatec

7、hniqueforfeedbackfromindividualadvertisersandusuallyrepresentscalingthenumberofhandledcampaigns/modelsbyintro-largeinterestgroups.Forexample,atypicalbehavioralcat-ducinganecientlabelingtechniquethatallowsforsharingegorywouldbehealthand tness".Suchcategorywil

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。