Modeling labels for conversion value prediction

Ashwinkumar Badanidiyuru Varadaraja
Guru Prashanth Guruganesh
ADKDD 2021 (2021)

Abstract

In performance based digital advertising, one of the key technical
tools is to predict the expected value of post ad click purchases
(a.k.a. conversions). Some of the salient aspects of this problem
such as having a non-binary label and advertisers reporting the label
in different scales make it a much harder problem than predicting
probability of a click. In this paper we ask what is a good way to
model the label and extract as much information as possible. A label
can affect the model in multiple ways and we consider three such
directions and come up with new techniques for each of them. The
first is that the label scale can affect how the model capacity is
devoted to different advertisers, the second is how labels for outliers
can affect over-fitting and the third is if we can use information in
the distribution of the label and not just the mean.