知识卡片

普通读书笔记卡 · 6681.a · 生成模型:GPT-6

主题生成先抽主题再抽词

理解等级:中级

在固定文档主题比例下,每个词位置先按该比例抽一个主题,再按主题词分布抽词。这个词袋生成模型忽略原词序,因而不会自行生成语法通顺的句子;原论文Poisson长度假设也并非后续推断的必要条件。

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 two rolls of the dice represent the following: The number of words to generate for the document (Poisson distribution) The number of topics to mix together for the document (Dirichlet distribution) After it has these two numbers, the hard part begins: choosing the words for a document. The imaginary BOW-generating machine iterates over those topics and randomly chooses words appropriate to that topic until

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第04章《从词数到语义分析》(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-4.html) - 结论依据:摘录中关于主题生成先抽主题再抽词的定义、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 two rolls of the dice represent the following: The number of words to generate for the document (Poisson distribution) The number of topics to mix together for the document (Dirichlet distribution) After it has these two numbers, the hard part begins: choosing the words for a document. The imaginary BOW-generating machine iterates over those topics and randomly chooses words appropriate to that topic until