Handling many conversions per click in modeling delayed feedback

TL;DR

This paper presents a novel approach for modeling delayed conversions in digital advertising, addressing challenges of varying delay distributions and distribution drift with an unbiased estimation model.

Contribution

It introduces a new unbiased estimation method that splits labels by delay buckets, employs thermometer encoding, and utilizes auxiliary information for stability.

Findings

Improved accuracy in predicting delayed conversions.

Reduced inference cost through thermometer encoding.

Enhanced model stability with auxiliary data.

Abstract

Predicting the expected value or number of post-click conversions (purchases or other events) is a key task in performance-based digital advertising. In training a conversion optimizer model, one of the most crucial aspects is handling delayed feedback with respect to conversions, which can happen multiple times with varying delay. This task is difficult, as the delay distribution is different for each advertiser, is long-tailed, often does not follow any particular class of parametric distributions, and can change over time. We tackle these challenges using an unbiased estimation model based on three core ideas. The first idea is to split the label as a sum of labels with different delay buckets, each of which trains only on mature label, the second is to use thermometer encoding to increase accuracy and reduce inference cost, and the third is to use auxiliary information to increase the stability of the model and to handle drift in the distribution.