跳到正文
arXiv cs.AI· Jianbo Zhu, Xing Fang, Jing Wang, Mingmin Jin, Bokang Wang, Guangxin Song, Zhenyu Xie, Junjie Bai·· 2 小时前AI 评分44

CQ-SID:基于语义聚类 ID 与专家引导 RL 的电商生成式检索

Efficient Generative Retrieval for E-commerce Search with Semantic Cluster IDs and Expert-Guided RL

AI 导读

一种将生成式检索作为召回阶段补充的电商搜索框架 CQ-SID 被提出,配合强化学习方法 EG-GRPO 对齐下游排序目标。离线实验显示 CQ-SID 较 RQ-VAE 基线在语义和个性化点击命中率上分别提升最高 26.76% 和 11.11%,beam search 规模减半;线上 A/B 测试 GMV 提升 1.15%、UCTCVR 提升 0.40%。

正文 · 原文

View PDF HTML (experimental)

Abstract:Generative retrieval offers a promising alternative by unifying the fragmented multi-stage retrieval process into a single end-to-end model. However, its practical adoption in industrial e-commerce search remains challenging, given the massive and dynamic product catalogs, strict latency requirements, and the need to align retrieval with downstream ranking goals. In this work, we propose a retrieval framework tailored for real-world recall scenarios, positioning generative retrieval as a recall-stage supplement rather than an end-to-end replacement. Our method, CQ-SID (Category-and-Query constrained Semantic ID), employs category-aware and query-item contrastive learning along with Residual Quantized VAEs to encode items into hierarchical semantic cluster identifiers, significantly reducing beam search complexity. Additionally, we develop EG-GRPO (Expert-Guided Group Relative Policy Optimization), a reinforcement learning approach that aligns generative recall with downstream ranking under sparse rewards by injecting ground-truth samples to stabilize training. Offline experiments on TmallAPP search logs show that CQ-SID achieves up to 26.76% and 11.11% relative gains in semantic and personalized click hitrate over RQ-VAE baselines, while halving beam search size. EG-GRPO further improves multi-objective performance. Online A/B tests confirm gains in GMV (+1.15%) and UCTCVR (+0.40%). The generative recall channel now contributes substantially in production, accounting for over 50.25% of exposures, 58.96% of clicks, and 72.63% of purchases, demonstrating a viable path for deploying generative retrieval in real-world e-commerce systems.
Subjects: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
Cite as: arXiv:2605.14434 [cs.IR]
  (or arXiv:2605.14434v2 [cs.IR] for this version)
  https://doi.org/10.48550/arXiv.2605.14434

arXiv-issued DOI via DataCite

Journal reference: Proc. 35th ACM Int. Conf. on Information and Knowledge Management (CIKM 2026), pp. 8801-8808
Related DOI: https://doi.org/10.1145/3799682.3840110

DOI(s) linking to related resources

Submission history

From: Jianbo Zhu [view email]
[v1] Thu, 14 May 2026 06:27:46 UTC (392 KB)
[v2] Tue, 17 Nov 2026 02:49:22 UTC (373 KB)

来源:arXiv cs.AI · arxiv.org