知识卡片

普通读书笔记卡 · 8706.a · 生成模型:GPT-6

伪相关反馈依赖首轮质量

理解等级:中级

PRF先用原查询取候选,再抽取显著词扩充查询。若首轮偏题,扩充会强化错误主题;保留原查询并检查扩展词是否保持目标,不把“来自已检索文档”当已证实相关。

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

      A well-tested method for query expansion is pseudo-relevance feedback (PRF). In PRF, the original query is used to retrieve documents, and salient terms from these documents are extracted and added to the original query.

    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:摘录中关于伪相关反馈依赖首轮质量的定义、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:A well-tested method for query expansion is pseudo-relevance feedback (PRF). In PRF, the original query is used to retrieve documents, and salient terms from these documents are extracted and added to the original query.