发表机构
Amazon Catalog AI（亚马逊目录人工智能）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对大型产品目录中产品族的不一致问题，本文提出将缺陷检测分解为子任务并引入位置任务条件化（PTC）蒸馏方法，将F1从52%提升至87%，成本降低98%，已部署处理超千万产品族。
AI 中文摘要
大型产品目录中的产品族存在重复项和单位不匹配等不一致问题，这些问题会降低客户体验。检测这些问题需要对冗长的产品列表中的多种错误类型进行推理，而由于长上下文限制，LLM分类质量会下降。我们通过将检测分解为聚焦的子任务来解决这一问题，这些子任务减少了上下文并隔离了错误类型，将F1分数从52%提高到87%。为了可扩展部署，我们引入了位置任务条件化（PTC），该方法通过在结构提示边界强化任务身份，将这种能力蒸馏到单个较小的模型中。PTC在五种模型和两个架构家族中优于基于推理的蒸馏，在成本降低高达98%的情况下，F1分数达到前沿模型的1.79%以内。我们的系统已在多个国家部署，处理超过1000万个产品族。
英文摘要
Product families in large product catalogs suffer from inconsistencies such as duplicates and unit mismatches that degrade customer experience. Detecting these requires reasoning over multiple error types across lengthy product listings, where LLM classification quality degrades due to long-context limitations. We address this by decomposing detection into focused sub-tasks that reduce context and isolate error types, improving F1 from 52% to 87%. For scalable deployment, we introduce Positional Task Conditioning (PTC), which distills this capability into a single smaller model by reinforcing task identity at structural prompt boundaries. PTC outperforms rationale-based distillation across five models and two architecture families, achieving within 1.79% F1 of the frontier at upto 98% lower cost. Our system is deployed across multiple countries processing 10+ million product families.
