In this work we discuss how to optimize mobile offer targeting Campaigns using results of predictive modeling. Mobile providers use their access to users to serve them offers of available services ("add new line", "buy new device", etc).
How to choose which offer should be serve to given user in given context (e.g. time and location)? Methods of supervised learning give us estimated probabilities of clicks on each candidate offer, but if we always serve users with an offer with the highest expected click through rate, then as result we can serve only one offer to all customers.
To maximize expected number of clicks with warranted minimal required number of impressions of each offer we use methods of linear programming, and to make these methods feasible we use clustering of users in space of expected probabilities.
We describe how choice of methods and parameters of predictive modeling, clustering and optimization affect performance of the offer recommendation system.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.