DADF: A Distribution-Aware Debiasing Framework for Watch-Time Regression in Recommender Systems

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yang, Yiqing, Zhao, Xinlong, Liu, Zhao, Lv, Xiao, Tang, Ruiming, Li, Han, Gai, Kun
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911693727072256
author Yang, Yiqing
Zhao, Xinlong
Liu, Zhao
Lv, Xiao
Tang, Ruiming
Li, Han
Gai, Kun
author_facet Yang, Yiqing
Zhao, Xinlong
Liu, Zhao
Lv, Xiao
Tang, Ruiming
Li, Han
Gai, Kun
contents Watch-time prediction is a central regression task in short-video recommender systems, where labels are highly long-tailed and residual errors vary systematically across observed watch-time regions. In practice, a model may appear globally calibrated while still overestimating short views and underestimating long views, because opposite errors cancel out in aggregate. Existing methods mainly improve the first-stage watch-time predictor, but often leave such residual distributional bias insufficiently corrected. We propose DADF, a distribution-aware debiasing framework for watch-time regression. Instead of replacing a deployed predictor, DADF performs second-stage multiplicative residual correction on top of it. DADF combines three complementary designs: a dynamic distribution-aware transformation for stabilizing long-tailed correction targets, a debias-factor-aware module for modeling heterogeneous residual patterns using inference-time observable factors, especially video duration, and a multi-label-aware module that exploits auxiliary prediction signals from engagement heads. We evaluate DADF on public short-video benchmarks and a large-scale industrial ranking system. DADF consistently improves both pointwise accuracy and ranking quality across datasets and backbones. In the industrial setting, it achieves a 1.88 percentage-point WUAUC gain over the production baseline, reduces MAE by 12.57%, and yields a statistically significant 0.347% lift in average time spent per device in online A/B testing. These results demonstrate that DADF effectively mitigates local calibration bias and provides a practical plug-in solution for debiasing long-tailed continuous targets. The source code is available at https://github.com/liuzhao09/DADF.
format Preprint
id arxiv_https___arxiv_org_abs_2605_17863
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle DADF: A Distribution-Aware Debiasing Framework for Watch-Time Regression in Recommender Systems
Yang, Yiqing
Zhao, Xinlong
Liu, Zhao
Lv, Xiao
Tang, Ruiming
Li, Han
Gai, Kun
Information Retrieval
Watch-time prediction is a central regression task in short-video recommender systems, where labels are highly long-tailed and residual errors vary systematically across observed watch-time regions. In practice, a model may appear globally calibrated while still overestimating short views and underestimating long views, because opposite errors cancel out in aggregate. Existing methods mainly improve the first-stage watch-time predictor, but often leave such residual distributional bias insufficiently corrected. We propose DADF, a distribution-aware debiasing framework for watch-time regression. Instead of replacing a deployed predictor, DADF performs second-stage multiplicative residual correction on top of it. DADF combines three complementary designs: a dynamic distribution-aware transformation for stabilizing long-tailed correction targets, a debias-factor-aware module for modeling heterogeneous residual patterns using inference-time observable factors, especially video duration, and a multi-label-aware module that exploits auxiliary prediction signals from engagement heads. We evaluate DADF on public short-video benchmarks and a large-scale industrial ranking system. DADF consistently improves both pointwise accuracy and ranking quality across datasets and backbones. In the industrial setting, it achieves a 1.88 percentage-point WUAUC gain over the production baseline, reduces MAE by 12.57%, and yields a statistically significant 0.347% lift in average time spent per device in online A/B testing. These results demonstrate that DADF effectively mitigates local calibration bias and provides a practical plug-in solution for debiasing long-tailed continuous targets. The source code is available at https://github.com/liuzhao09/DADF.
title DADF: A Distribution-Aware Debiasing Framework for Watch-Time Regression in Recommender Systems
topic Information Retrieval
url https://arxiv.org/abs/2605.17863