Generating economic value from big data is a challenge for many companies these days. On the Internet, a major source of big data is structured and unstructured data generated by users. Companies can use this data to better understand patterns of user behavior and to improve marketing decisions. In this paper, we focus on data generated in real-time advertising where billions of advertising slots are sold by auction. The auctions are triggered by user activity on websites that use this form of advertising to sell their advertising slots. During an auction, so-called bid requests are sent to advertisers who bid for the advertising slots. We develop a model that uses bid requests to predict whether a user will visit a certain website during his or her user journey. These predictions can be used by advertisers to derive user interests early in the sales funnel and, thus, to increase profits from branding campaigns. By iteratively applying a Bayesian multinomial logistic model to data from a case study, we show how to constantly improve the predictive accuracy of the model. We calculate the economic value of our model and show that it can be beneficial for advertisers in the context of cross-channel advertising.
Stange, Martin and Funk, Burkhardt, "PREDICTING ONLINE USER BEHAVIOR BASED ON REAL-TIME ADVERTISING DATA" (2016). Research Papers. 152.