← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8707.a · 生成模型:GPT-6

词法检索仍是有效基线

理解等级:中级

BM25等匹配词项的方法适合明确名称、编号和术语,配合改写可成为强基线。语义检索对词汇差异有帮助,不能因此忽略精确词匹配;原书未展开BM25公式,不补造实现细节当本章知识。

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

      The Unreasonable Effectiveness of BM25 Despite the existence of advanced deep learning techniques for retrieval, keyword matching/probabilistic retrieval techniques like BM25 can be a very strong baseline and, when paired with query or document rewriting, can potentially even be good enough for your application. Other traditional techniques supported by Apache Lucene/ElasticSearch include term frequency-inverse document frequency (Tf-IDf), divergence from independence (DFI), divergence from randomness (DFR),

    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:摘录中关于词法检索仍是有效基线的定义、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:The Unreasonable Effectiveness of BM25 Despite the existence of advanced deep learning techniques for retrieval, keyword matching/probabilistic retrieval techniques like BM25 can be a very strong baseline and, when paired with query or document rewriting, can potentially even be good enough for your application. Other traditional techniques supported by Apache Lucene/ElasticSearch include term frequency-inverse document frequency (Tf-IDf), divergence from independence (DFI), divergence from randomness (DFR),