知识卡片

普通读书笔记卡 · 8707.c · 生成模型:GPT-6

元数据过滤错误会直接漏召回

理解等级:中级

时间、主题和文档类型可缩小候选域,提高效率与精度。但错误硬过滤会永久排除正确证据,检索模型无法补救;需检查标签质量及查询到条件的转换,不能认为元数据越多就必然越好。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      metadata filters for improving retrieval. The more metadata you gather during the data representation and storage phase, the better the retrieval results. For example, if you have performed topic modeling of your data store in advance, you can restrict your search results to a subset of topics, with the filters being applied either using a hardcoded set of rules or determined by an LLM.

    参考来源

    - 位置:《Designing Large Language Model Applications》第12章《检索增强生成》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch12.html) - 结论依据:摘录中关于元数据过滤错误会直接漏召回的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:metadata filters for improving retrieval. The more metadata you gather during the data representation and storage phase, the better the retrieval results. For example, if you have performed topic modeling of your data store in advance, you can restrict your search results to a subset of topics, with the filters being applied either using a hardcoded set of rules or determined by an LLM.