知识卡片
问答字面匹配与语义正确各有边界
EM依规定的规范化后精确比较,token F1按预测与参考的共同词计数计算。任意有一个词重叠并不是通用准确率,字面高分也可能忽略否定或关键数字;生成回答还需独立核实事实支持。
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
To evaluate a retrieval-based chatbot, such as a QA assistant, the metrics will be different, though you still need to have a labeled dataset with questions and matching answers based on your documents. You can generate this dataset with open source tools, like deepset’s annotation tool.58
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:摘录中关于问答字面匹配与语义正确各有边界的定义、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:To evaluate a retrieval-based chatbot, such as a QA assistant, the metrics will be different, though you still need to have a labeled dataset with questions and matching answers based on your documents. You can generate this dataset with open source tools, like deepset’s annotation tool.58