← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6666.i · 生成模型:GPT-6

全量取对数不同于次线性词频

理解等级:中级

log(TF×IDF)=log(TF)+log(IDF)只在两因子为正时成立,零值需额外约定。对整个TF–IDF再取对数不同于仅对词频做次线性缩放,也会改变相似度几何;一般稀疏检索并不必须为避免下溢而如此处理。

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

      In some implementations, all the calculations can be done in log space so that multiplications become additions and division becomes subtraction: >>> log_tf = log(term_occurences_in_doc) -\ ... log(num_terms_in_doc) #1 >>> log_log_idf = log(log(total_num_docs) -\ ... log(num_docs_containing_term)) #2 >>> log_tf_idf = log_tf + log_log_idf #3 #1 The log probability of a particular term in a particular document #2 Log of the log probability of a particular

    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:摘录中关于全量取对数不同于次线性词频的定义、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:In some implementations, all the calculations can be done in log space so that multiplications become additions and division becomes subtraction: >>> log_tf = log(term_occurences_in_doc) -\ ... log(num_terms_in_doc) #1 >>> log_log_idf = log(log(total_num_docs) -\ ... log(num_docs_containing_term)) #2 >>> log_tf_idf = log_tf + log_log_idf #3 #1 The log probability of a particular term in a particular document #2 Log of the log probability of a particular