知识卡片

结构图卡 · 6679.d · 生成模型:GPT-6

主题分类需分清无监督与监督两段

理解等级:中级

结构图:

flowchart TD
 A[固定训练验证测试划分] --> B[训练文本拟合词表和IDF]
 B --> C[训练特征拟合LSA或主题分配]
 C --> D[训练低维表示]
 D --> E[标签监督拟合判别器]
 B --> F[验证文本仅变换]
 C --> F
 F --> G[预测并按误差成本选参数]
 E --> G
 G --> H[锁定整链后最终测试]

主题模型本身不需要毒性标签,判别器需要;验证资料不能在上游拟合时提前混入。

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

      4.4.3 How well does LSA perform for toxicity detection? You’ve spent enough time peering into the topics—let’s see how our model performs with lower-dimensional representation of the comments! You’ll use the same code we ran in listing 4.3 but will apply it on the new 16-dimensional vectors. This time, the classification will go much faster: >>> X_train_16d, X_test_16d, y_train_16d, y_test_16d = train_test_split( ... svd_topic_vectors, comments.toxic.values,

    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:摘录中关于主题分类需分清无监督与监督两段的定义、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:4.4.3 How well does LSA perform for toxicity detection? You’ve spent enough time peering into the topics—let’s see how our model performs with lower-dimensional representation of the comments! You’ll use the same code we ran in listing 4.3 but will apply it on the new 16-dimensional vectors. This time, the classification will go much faster: >>> X_train_16d, X_test_16d, y_train_16d, y_test_16d = train_test_split( ... svd_topic_vectors, comments.toxic.values,