Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EGEAN: An Exposure-Guided Embedding Alignment Network for Post-Click Conversion Estimation

Huajian Feng, Guoxiao Zhang, Yadong Zhang, Yi We, Qiang Liu

TL;DR

This work addresses post-click CVR estimation under covariate shift between click and non-click spaces, where non-click conversions are MNAR. It introduces EGEAN, an exposure-guided embedding alignment network with three pillars: exposure probability prediction, a task-personalized network using LoRA and gate-based personalization, and metric learning via Maximum Mean Discrepancy to align shared and CVR-specific embeddings. To further reduce bias and variance in small-propensity regimes, it proposes PVDR, a steady-state robust estimator that combines imputation and reweighting under a derived condition, generalizing IPS and SDR. Online A/B tests on Meituan show sizeable CVR and GMV improvements, while ablations confirm the contribution of each component, making EGEAN a practical approach for industrial CVR estimation.

Abstract

Accurate post-click conversion rate (CVR) estimation is crucial for online advertising systems. Despite significant advances in causal approaches designed to address the Sample Selection Bias problem, CVR estimation still faces challenges due to Covariate Shift. Given the intrinsic connection between the distribution of covariates in the click and non-click spaces, this study proposes an Exposure-Guided Embedding Alignment Network (EGEAN) to address estimation bias caused by covariate shift. Additionally, we propose a Parameter Varying Doubly Robust Estimator with steady-state control to handle small propensities better. Online A/B tests conducted on the Meituan advertising system demonstrate that our method significantly outperforms baseline models with respect to CVR and GMV, validating its effectiveness. Code is available: https://github.com/hydrogen-maker/EGEAN.

EGEAN: An Exposure-Guided Embedding Alignment Network for Post-Click Conversion Estimation

TL;DR

This work addresses post-click CVR estimation under covariate shift between click and non-click spaces, where non-click conversions are MNAR. It introduces EGEAN, an exposure-guided embedding alignment network with three pillars: exposure probability prediction, a task-personalized network using LoRA and gate-based personalization, and metric learning via Maximum Mean Discrepancy to align shared and CVR-specific embeddings. To further reduce bias and variance in small-propensity regimes, it proposes PVDR, a steady-state robust estimator that combines imputation and reweighting under a derived condition, generalizing IPS and SDR. Online A/B tests on Meituan show sizeable CVR and GMV improvements, while ablations confirm the contribution of each component, making EGEAN a practical approach for industrial CVR estimation.

Abstract

Accurate post-click conversion rate (CVR) estimation is crucial for online advertising systems. Despite significant advances in causal approaches designed to address the Sample Selection Bias problem, CVR estimation still faces challenges due to Covariate Shift. Given the intrinsic connection between the distribution of covariates in the click and non-click spaces, this study proposes an Exposure-Guided Embedding Alignment Network (EGEAN) to address estimation bias caused by covariate shift. Additionally, we propose a Parameter Varying Doubly Robust Estimator with steady-state control to handle small propensities better. Online A/B tests conducted on the Meituan advertising system demonstrate that our method significantly outperforms baseline models with respect to CVR and GMV, validating its effectiveness. Code is available: https://github.com/hydrogen-maker/EGEAN.

Paper Structure

This paper contains 18 sections, 13 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. PROPOSED METHOD
  3. Exposure- Guided Embedding Alignment Network
  4. Exposure Probability Prediction Task.
  5. Task Personalized Network.
  6. Metric learning.
  7. Parameter Varying DR Estimator
  8. Problem Definition
  9. PVDR Estimator.
  10. Experiments
  11. Evaluation Metric
  12. Baseline Methods
  13. Implementation Details
  14. Overall Performance
  15. Ablation Study
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Illustration of user behaviors in Meituan online advertising.
  • Figure 2: The overall framework of EGEAN.
  • Figure 3: The t-SNE visualization of feature distributions for TSN modules on the Industrial dataset Meituan. Blue and cherry blossom representing the unclicked and clicked events, respectively.