Abstract:
|
Mobile providers use their access to users to serve them promotions (offers) for additional services ("add new line", "buy new device", etc) and 3rd party applications and services. Modeling of mobile offer campaigns have a number of features separating the task from standard classification tasks on one side and recommender systems on the other side. In general the model has form
f(Agg(y)) ~ g(Agg(Event, Entity, TimeWindow)) (1)
Where
y ? {Click, Conversion},
Event ? {Impression, Click, Conversion},
Entity ? {UserId, Offer,Channel, Hour, DayOfWeek, Zip,.},
Offer ? {OfferId, OfferTitle, OfferCategory},
TimeWindow ? {3 days, 5days, 7 days, 14 days, 28 days, 30 days, +∞ },
Agg ? {MovingAverage, ExponentiallySmoothedMovingAverage },
f is linking function and g is a model.
For example, one of arguments of function g in model (1) could be "7days-MovingAverage_ofImpressions_ofOffer37_forUser234_byFridays" We describe variations of predictive modeling methods (classification, regression, clustering, etc.) as well as text mining methods to analyze mobile providers log files, including impressions, clicks and conversio
|
Copyright © American Statistical Association.