USD: A User-Intent-Driven Sampling and Dual-Debiasing Framework for Large-Scale Homepage Recommendations

2025-07-08

AI Summary

In large-scale homepage recommendation, exposure bias introduces spurious negatives (non-clicks ≠ disinterest) and spurious positives (clicks ≠ genuine interest). Existing methods lack systematic analysis of ineffective exposures and typically address sampling or debiasing in isolation. This paper presents the first systematic modeling of ineffective exposure effects, proposing a user-intent-driven unified framework for joint sampling and debiasing. Specifically, it dynamically selects high-quality negative samples via intent-aware identification, and introduces a dual-debiasing module that jointly corrects exposure bias and click bias in an end-to-end manner. Deployed in Taobao’s homepage marketing system, online A/B tests demonstrate significant improvements: UCTR increases notably, with Baiyibutie and Taobaomiaosha conversion rates rising by 35.4% and 14.5%, respectively.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Data Mining & Knowledge Management: Recommender Systems
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Economics and fairness of platforms and recommendation systems

Abstract

Large-scale homepage recommendations face critical challenges from pseudo-negative samples caused by exposure bias, where non-clicks may indicate inattention rather than disinterest. Existing work lacks thorough analysis of invalid exposures and typically addresses isolated aspects (e.g., sampling strategies), overlooking the critical impact of pseudo-positive samples - such as homepage clicks merely to visit marketing portals. We propose a unified framework for large-scale homepage recommendation sampling and debiasing. Our framework consists of two key components: (1) a user intent-aware negative sampling module to filter invalid exposure samples, and (2) an intent-driven dual-debiasing module that jointly corrects exposure bias and click bias. Extensive online experiments on Taobao demonstrate the efficacy of our framework, achieving significant improvements in user click-through rates (UCTR) by 35.4% and 14.5% in two variants of the marketing block on the Taobao homepage, Baiyibutie and Taobaomiaosha.

Problem

• Addresses pseudo-negative samples from exposure bias in recommendations
• Identifies and mitigates invalid exposures and pseudo-positive samples
• Proposes a dual-debiasing framework to correct exposure and click biases

Innovation

• User intent-aware negative sampling module
• Intent-driven dual-debiasing module
• Corrects exposure and click bias jointly