Follow the Prophet: Accurate Online Conversion Rate Prediction in the Face of Delayed Feedback

TL;DR

This paper introduces FTP, a novel online learning method for conversion rate prediction that effectively addresses the delayed feedback problem in online advertising by predicting an ideal 'prophet' model through multi-task aggregation.

Contribution

The paper proposes a new approach called 'Following the Prophet' that predicts the ideal feedback model during online learning, improving accuracy despite feedback delays.

Findings

Outperforms state-of-the-art baselines on real-world datasets

Effectively models diverse feedback delays across ad types and users

Provides a practical solution for online advertising feedback challenges

Abstract

The delayed feedback problem is one of the imperative challenges in online advertising, which is caused by the highly diversified feedback delay of a conversion varying from a few minutes to several days. It is hard to design an appropriate online learning system under these non-identical delay for different types of ads and users. In this paper, we propose to tackle the delayed feedback problem in online advertising by "Following the Prophet" (FTP for short). The key insight is that, if the feedback came instantly for all the logged samples, we could get a model without delayed feedback, namely the "prophet". Although the prophet cannot be obtained during online learning, we show that we could predict the prophet's predictions by an aggregation policy on top of a set of multi-task predictions, where each task captures the feedback patterns of different periods. We propose the objective and optimization approach for the policy, and use the logged data to imitate the prophet. Extensive experiments on three real-world advertising datasets show that our method outperforms the previous state-of-the-art baselines.