知识卡片
同义增强样本须核对标签稳定
替换词语可形成训练变体或鲁棒性测试,但替换可能改变情感、事实或含义,不能自动复制原标签。测试集经激进替换后的成绩不是天然准确率下界,也不保证增加十倍变体就等于十倍独立信息。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 思维的词元
原书出处 · 1 处
- · 查看原书依据
Search engines can improve their recall for rare terms by using synonym substitution. When you have limited labeled data, you can often expand your dataset tenfold with synonym substitution alone. If you want to find a lower bound on the accuracy of your model, you can aggressively substitute synonyms in your test set to see how robust your model is to these changes. And if
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:摘录中关于同义增强样本须核对标签稳定的定义、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:Search engines can improve their recall for rare terms by using synonym substitution. When you have limited labeled data, you can often expand your dataset tenfold with synonym substitution alone. If you want to find a lower bound on the accuracy of your model, you can aggressively substitute synonyms in your test set to see how robust your model is to these changes. And if