知识卡片

普通读书笔记卡 · 6666.d · 生成模型:GPT-6

信息惊异度不同于分布熵

理解等级:中级

把文档出现比例df/N作为经验概率,−log(df/N)可视作出现该词这一事件的惊异度。熵则是对整个事件分布的期望惊异度;单个词一个IDF值不能直接叫完整语料熵,也不是该词含有多少语义的测量。

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 good way to think of a term’s IDF is this: how surprising is it that this token is in this document? The concept of measuring the surprise in a token might not sound like a very mathematical idea. However, in statistics, physics, and information theory, the surprise of a symbol is used to measure its entropy or information content. And that is exactly what

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第03章《词频与逆文档频率向量》(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-3.html) - 结论依据:摘录中关于信息惊异度不同于分布熵的定义、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 good way to think of a term’s IDF is this: how surprising is it that this token is in this document? The concept of measuring the surprise in a token might not sound like a very mathematical idea. However, in statistics, physics, and information theory, the surprise of a symbol is used to measure its entropy or information content. And that is exactly what