知识卡片
语义分块保留非连续来源
可按主题相近把文本组成chunk,甚至链接不同文档片段。语义相近不意味着来自同一事实背景,必须保存每段原始位置、时间及边界,防止把相似但冲突的内容拼成一条证据。
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
- · 表示学习与嵌入
原书出处 · 1 处
- · 查看原书依据
Semantic Chunking The principle behind semantic chunking is that similar information should be grouped into coherent chunks. Paragraph boundaries provide a weak signal for semantic chunking, but more advanced methods can be employed. One approach is to cluster the document based on topics, with each chunk containing information pertaining to the same topic. The chunks need not necessarily be built from contiguous text if it
参考来源
- 位置:《Designing Large Language Model Applications》第11章《表示学习与嵌入》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch11.html)
- 结论依据:摘录中关于语义分块保留非连续来源的定义、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:Semantic Chunking The principle behind semantic chunking is that similar information should be grouped into coherent chunks. Paragraph boundaries provide a weak signal for semantic chunking, but more advanced methods can be employed. One approach is to cluster the document based on topics, with each chunk containing information pertaining to the same topic. The chunks need not necessarily be built from contiguous text if it