知识卡片
低置信样本只是错误发现入口之一
按低最高分排序有助发现未知措辞与新意图,但高置信误判不会因此被发现。应同时抽样已接受的回复、相近类别和重要错例;标签新增与流程分支调整是两种修复,需分别回归。
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
原书出处 · 1 处
- · 查看原书依据
As a data-driven conversation designer, you’ll want to prioritize the most-frequent messages from users for labeling and conversation design. One way to do that is to sort your users’ utterances by the maximum predicted label confidence (probability from predict_probas()). You can scan the lowest-confidence label predictions to see if any can be labeled with one of your existing intents. Labeling utterances with existing intents is
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第12章《对话引擎与会话设计》(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-12.html)
- 结论依据:摘录中关于低置信样本只是错误发现入口之一的定义、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:As a data-driven conversation designer, you’ll want to prioritize the most-frequent messages from users for labeling and conversation design. One way to do that is to sort your users’ utterances by the maximum predicted label confidence (probability from predict_probas()). You can scan the lowest-confidence label predictions to see if any can be labeled with one of your existing intents. Labeling utterances with existing intents is