重新审视生成式与稠密检索:静态价值先验逆转差距
原标题:Revisiting Generative vs. Dense Retrieval: A Static Value Prior Reverses the Gap
AI 摘要
该研究重新审视了生成式检索(GR)与稠密检索(DR)在电商候选生成中的表现差异。作者在 Amazon Toys 2014 和 2023 数据集上发现,GR 在双方共同覆盖用户的最佳相关商品价值上领先,而 DR 在精确目标召回和整列表相关价值上领先;分析表明流行度是 GR 价值领先的最大来源(2014 年占 49.8%,2023 年占 65.4%)。作者提出一种无需训练的静态价值先验,对 DR 结果做一次重排即可逆转 GR 的最佳商品价值优势,同时保持召回领先;在工业级内部电商数据集上,14B SID-LLM 仍不及更小的稠密编码器。
正文节选
Research \acmCodeLinkhttps://anonymous.4open.science/r/gr-dr-release-BCE4 Revisiting Generative vs. Dense Retrieval: A Static Value Prior Reverses the Gap 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 dataset