← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6659.h · 生成模型:GPT-6

精确匹配与归一化召回可以分层

理解等级:中级

检索式聊天可先查原形式的较精确匹配,再将归一化候选作为回退并降低排序优先级。这样显式控制召回与误答代价,而非把两路结果视为同等可靠;是否拒答和何时采用回退还需按任务定义。

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

      For a search-based chatbot, precision is usually more important than recall. A false positive match can cause your chatbot to say something inappropriate, while false negatives just cause your chatbot to have to humbly admit it cannot find anything appropriate to say. Your chatbot will sound better if your NLP pipeline first searches for matches to users’ questions using unstemmed, unnormalized words. Your search algorithm

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第02章《思维的词元》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-2.html) - 结论依据:摘录中关于精确匹配与归一化召回可以分层的定义、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:For a search-based chatbot, precision is usually more important than recall. A false positive match can cause your chatbot to say something inappropriate, while false negatives just cause your chatbot to have to humbly admit it cannot find anything appropriate to say. Your chatbot will sound better if your NLP pipeline first searches for matches to users’ questions using unstemmed, unnormalized words. Your search algorithm