Research

Paper

TESTING February 27, 2026

RAD-DPO: Robust Adaptive Denoising Direct Preference Optimization for Generative Retrieval in E-commerce

Authors

Zhiguo Chen, Guohao Sun, Yiming Qiu, Xingzhi Yao, Mingming Li, Huimu Wang, Yangqi Zhang, Songlin Wang, Sulong Xu

Abstract

Generative Retrieval (GR) has emerged as a powerful paradigm in e-commerce search, retrieving items via autoregressive decoding of Semantic IDs (SIDs). However, aligning GR with complex user preferences remains challenging. While Direct Preference Optimization (DPO) offers an efficient alignment solution, its direct application to structured SIDs suffers from three limitations: (i) it penalizes shared hierarchical prefixes, causing gradient conflicts; (ii) it is vulnerable to noisy pseudo-negatives from implicit feedback; and (iii) in multi-label queries with multiple relevant items, it exacerbates a probability "squeezing effect" among valid candidates. To address these issues, we propose RAD-DPO, which introduces token-level gradient detachment to protect prefix structures, similarity-based dynamic reward weighting to mitigate label noise, and a multi-label global contrastive objective integrated with global SFT loss to explicitly expand positive coverage. Extensive offline experiments and online A/B testing on a large-scale e-commerce platform demonstrate significant improvements in ranking quality and training efficiency.

Metadata

arXiv ID: 2602.23964
Provider: ARXIV
Primary Category: cs.IR
Published: 2026-02-27
Fetched: 2026-03-02 06:04

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.23964v1</id>\n    <title>RAD-DPO: Robust Adaptive Denoising Direct Preference Optimization for Generative Retrieval in E-commerce</title>\n    <updated>2026-02-27T12:17:06Z</updated>\n    <link href='https://arxiv.org/abs/2602.23964v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.23964v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Generative Retrieval (GR) has emerged as a powerful paradigm in e-commerce search, retrieving items via autoregressive decoding of Semantic IDs (SIDs). However, aligning GR with complex user preferences remains challenging. While Direct Preference Optimization (DPO) offers an efficient alignment solution, its direct application to structured SIDs suffers from three limitations: (i) it penalizes shared hierarchical prefixes, causing gradient conflicts; (ii) it is vulnerable to noisy pseudo-negatives from implicit feedback; and (iii) in multi-label queries with multiple relevant items, it exacerbates a probability \"squeezing effect\" among valid candidates. To address these issues, we propose RAD-DPO, which introduces token-level gradient detachment to protect prefix structures, similarity-based dynamic reward weighting to mitigate label noise, and a multi-label global contrastive objective integrated with global SFT loss to explicitly expand positive coverage. Extensive offline experiments and online A/B testing on a large-scale e-commerce platform demonstrate significant improvements in ranking quality and training efficiency.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.IR'/>\n    <published>2026-02-27T12:17:06Z</published>\n    <arxiv:primary_category term='cs.IR'/>\n    <author>\n      <name>Zhiguo Chen</name>\n    </author>\n    <author>\n      <name>Guohao Sun</name>\n    </author>\n    <author>\n      <name>Yiming Qiu</name>\n    </author>\n    <author>\n      <name>Xingzhi Yao</name>\n    </author>\n    <author>\n      <name>Mingming Li</name>\n    </author>\n    <author>\n      <name>Huimu Wang</name>\n    </author>\n    <author>\n      <name>Yangqi Zhang</name>\n    </author>\n    <author>\n      <name>Songlin Wang</name>\n    </author>\n    <author>\n      <name>Sulong Xu</name>\n    </author>\n  </entry>"
}