基于细粒度标签的在线视频广告投放机制研究.pdfVIP

计算机研究与发展 7544／issn1000—1239．2014 JournalofComputerResearchandDevelopment 51(12)：2733—2745。2014 基于细粒度标签的在线视频广告投放机制研究 陆 枫 王子锐 廖小飞 金 海 (服务计算技术与系统教育部重 点实验室(华 中科技大学) 武汉 430074) (集群与 网格计算湖北省重 点实验室(华 中科技大学) 武汉 430074) (1ufengl@hust．edu．cn) OnlineVideoAdvertisingBasedonFine—GrainedVideoTags LuFeng，WangZirui，LiaoXiaofei，andJinHai (KeyLaboratory ofServicesComputingTechn0logyandSytern (HuazhongUniversity 0fScienceand Technol0g )， MinistryofEducation，Wuhan430074) (KeyLaboratoryofClusterandGridComputingofHubeiProvince (HuazhongUniversityofScienceandTechnoZ0g )， Wuhan430074) Abstract W ith thedevelopmentoftheInternet，ithasbeenatrendofmanualtagging，labelingand sharingvideos．Rationaluseoftheseswarm intelligencewillhelptoimprovetheeffectivenessofvideo advertising．Themethod presented in thispaperfirstcollectsthefine—grained uservide0 tags，and generatesthevideohotspotsbythevideotimeline—weightedmethod ． Then，basedon theideaofthe classificationmatching，thedescriptionofthevideohotspotscanbeusedtoselecttheadvertising At ． last，thetimepointsthatthepopularattention hasdroppedby thebiggestlevelare found toput advertising．Experimentsshow that，amongthemega—scalevideoset，thecontentcorrelationbetween thehotspotandtheadvertisementsselectedbythismethodcanreach85％．Theprobabilitvthatthe usersclose adswindows islessthan 10 ． Compared with the ads system thathasbeen wide1v adoptedSOfar，theaveragebroadcasttimeofthenew methodcan beincreasedbv21 ． 5 ，theclick— throughrateisimprovedfrom 0．65 to0．73％． Keywords onlinevideoadvertising；fine—grainedvideotags；videohotspots；machinelearning；target