arXiv cs.AI· Renbo Tu, Yang Liu, Shuying Sun·· 11 小时前AI 评分34
重新审视生成式检索与稠密检索：静态价值先验可逆转差距
Revisiting Generative vs. Dense Retrieval: A Static Value Prior Reverses the Gap
AI 导读
在 Amazon Toys 两个数据集上，生成式检索（GR）仅在双方共同覆盖用户的最佳相关商品价值上领先，而稠密检索（DR）在精确目标召回和整条排序列表的相关价值上更优；分析显示 GR 的价值领先主要来自流行度。在 DR 重排阶段加入相同的静态商品价值即可逆转这一领先并保持召回高于 GR。在工业级内部电商数据集上，经 LLM 判定相关性，14B 的 SID-LLM 仍不及小 47 倍的稠密编码器。
正文 · 原文
Abstract:Generative retrieval (GR) has emerged as a strong candidate generator for sequential recommendation as language models and semantic-item identifiers have improved. However, GR is not a uniformly stronger replacement of dense retrieval (DR) in online retail candidate generation. We examine their differences on two Amazon Toys datasets under different metrics. GR leads on best relevant item value among users both methods cover, while DR leads on exact-target recall and relevant value across the ranked list. With deep analysis, we find that popularity contributes the most to GR's measured value lead. Adding the same static item value to DR in one reranking step reverses this lead and keeps recall above GR. Finally, on an industrial-scale internal e-commerce dataset with LLM-judged relevance, we show that a 14B SID-LLM still underperforms a $47\times$ smaller dense encoder.
| Comments: | ACM CIKM EARL 2026 |
| Subjects: | Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.19065 [cs.IR] |
| (or arXiv:2610.19065v2 [cs.IR] for this version) | |
| https://doi.org/10.48550/arXiv.2610.19065 arXiv-issued DOI via DataCite |
Submission history
From: Renbo Tu [view email]
[v1]
Tue, 13 Oct 2026 19:02:30 UTC (227 KB)
[v2]
Wed, 4 Nov 2026 22:24:21 UTC (242 KB)
来源：arXiv cs.AI · arxiv.org