In performance based display advertising, campaign effectiveness is oftenmeasured in terms of conversions that represent some desired user actionslike purchases and product information requests on advertisers' web site.Hence, identifying and targeting potential converters is of vitalimportance to boost campaign performance. This is often accomplished bymarketers who define the user base of campaigns based on behavioral,demographic, contextual, geographic, search, social, purchase, and othercharacteristics. Such a process is manual and subjective, it often fails toutilize the full potential of targeting. In this paper we show that byusing past converted users of campaigns and campaign meta-data (e.g., adcreatives, landing pages), we can combine disparate user information in aprincipled way to effectively and automatically target converters fornew/existing campaigns. At the heart of our approach is a factor model thatestimates the affinity of each user feature to a campaign using historicalconversion data. In fact, our approach allows building a conversion modelfor a brand new campaign through campaign meta-data alone, and hencetargets potential converters even before the campaign is run. Throughextensive experiments, we show the superiority of our factor model approachrelative to several other baselines. Moreover, we show that the performanceof our approach at the beginning of a campaign's life is typically betterthan the other models even when they are trained using all conversion dataafter campaign completion. This clearly shows the importance and value ofusing historical campaign data in constructing an effective audienceselection strategy for display advertising.
Monday, November 14, 2011
Free and open to the public