Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AlignPxtr: Aligning Predicted Behavior Distributions for Bias-Free Video Recommendations

Chengzhi Lin, Chuyuan Wang, Annan Xie, Wuhong Wang, Ziye Zhang, Canguang Ruan, Yuancai Huang, Yongqi Liu

TL;DR

AlignPxtr tackles bias in video recommendations by treating user interest $Z$ as independent of bias factors $Y$ and decoupling them through distribution alignment. The method aligns the latent behavior distribution $X$ across bias conditions using conditional quantile mapping ($Z=F_{X|Y=y}(X)$) to achieve $Z\perp Y$ and $Z$ distributed as $U(0,1)$, with a practical mean-alignment alternative for real-time systems. It extends to multiple bias dimensions and both continuous and discrete signals, providing training steps for behavior modeling and bias modeling, plus inference procedures that produce a bias-free score and can combine multiple behavior signals. Online experiments on Kuaishou Lite and Kuaishou show consistent, statistically significant gains in active days, app usage time, and watchtime, validating the approach's real-world impact on long-term retention and engagement.

Abstract

In video recommendation systems, user behaviors such as watch time, likes, and follows are commonly used to infer user interest. However, these behaviors are influenced by various biases, including duration bias, demographic biases, and content category biases, which obscure true user preferences. In this paper, we hypothesize that biases and user interest are independent of each other. Based on this assumption, we propose a novel method that aligns predicted behavior distributions across different bias conditions using quantile mapping, theoretically guaranteeing zero mutual information between bias variables and the true user interest. By explicitly modeling the conditional distributions of user behaviors under different biases and mapping these behaviors to quantiles, we effectively decouple user interest from the confounding effects of various biases. Our approach uniquely handles both continuous signals (e.g., watch time) and discrete signals (e.g., likes, comments), while simultaneously addressing multiple bias dimensions. Additionally, we introduce a computationally efficient mean alignment alternative technique for practical real-time inference in large-scale systems. We validate our method through online A/B testing on two major video platforms: Kuaishou Lite and Kuaishou. The results demonstrate significant improvements in user engagement and retention, with \textbf{cumulative lifts of 0.267\% and 0.115\% in active days, and 1.102\% and 0.131\% in average app usage time}, respectively. The results demonstrate that our approach consistently achieves significant improvements in long-term user retention and substantial gains in average app usage time across different platforms. Our core code will be publised at https://github.com/justopit/CQE.

AlignPxtr: Aligning Predicted Behavior Distributions for Bias-Free Video Recommendations

TL;DR

AlignPxtr tackles bias in video recommendations by treating user interest as independent of bias factors and decoupling them through distribution alignment. The method aligns the latent behavior distribution across bias conditions using conditional quantile mapping () to achieve and distributed as , with a practical mean-alignment alternative for real-time systems. It extends to multiple bias dimensions and both continuous and discrete signals, providing training steps for behavior modeling and bias modeling, plus inference procedures that produce a bias-free score and can combine multiple behavior signals. Online experiments on Kuaishou Lite and Kuaishou show consistent, statistically significant gains in active days, app usage time, and watchtime, validating the approach's real-world impact on long-term retention and engagement.

Abstract

In video recommendation systems, user behaviors such as watch time, likes, and follows are commonly used to infer user interest. However, these behaviors are influenced by various biases, including duration bias, demographic biases, and content category biases, which obscure true user preferences. In this paper, we hypothesize that biases and user interest are independent of each other. Based on this assumption, we propose a novel method that aligns predicted behavior distributions across different bias conditions using quantile mapping, theoretically guaranteeing zero mutual information between bias variables and the true user interest. By explicitly modeling the conditional distributions of user behaviors under different biases and mapping these behaviors to quantiles, we effectively decouple user interest from the confounding effects of various biases. Our approach uniquely handles both continuous signals (e.g., watch time) and discrete signals (e.g., likes, comments), while simultaneously addressing multiple bias dimensions. Additionally, we introduce a computationally efficient mean alignment alternative technique for practical real-time inference in large-scale systems. We validate our method through online A/B testing on two major video platforms: Kuaishou Lite and Kuaishou. The results demonstrate significant improvements in user engagement and retention, with \textbf{cumulative lifts of 0.267\% and 0.115\% in active days, and 1.102\% and 0.131\% in average app usage time}, respectively. The results demonstrate that our approach consistently achieves significant improvements in long-term user retention and substantial gains in average app usage time across different platforms. Our core code will be publised at https://github.com/justopit/CQE.

Paper Structure

This paper contains 25 sections, 1 theorem, 7 equations, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Problem Formulation
  4. Causal Framework
  5. Bias Decoupling via Distribution Alignment
  6. Theoretical Foundation
  7. Conditional Quantile Mapping
  8. From Uniform Distribution to Domain-Specific Distribution
  9. Mean Alignment as a Practical Alternative
  10. Model Training and Inference
  11. Training Procedure
  12. Step 1: Behavior Modeling
  13. Step 2: Bias Modeling
  14. Inference
  15. Experiments
  16. ...and 10 more sections

Key Result

proposition 1

Let $X$ be a continuous random variable with conditional CDF $F_{X|Y=y}$ given $Y=y$. Define $Z = F_{X|Y=y}(X)$. Then $Z \perp Y$ (i.e., $Z$ is independent of $Y$).

Figures (1)

  • Figure 1: Causal graph illustrating the relationships between user+video+context, bias factors ($Y$), user interest ($Z$), latent behavior distribution ($X$), and observed user behavior ($S$). The bias factors ($Y$) independently influence user interest ($Z$). User interest ($Z$) and bias factors ($Y$) jointly affect the latent behavior distribution ($X$), which in turn determines the observed user behavior ($S$).

Theorems & Definitions (1)

  • proposition 1