知识卡片
重排无法挽回首轮漏掉的文档
初始检索覆盖候选,较昂贵重排器再精细打分,通常提高前列相关性。评价时同时看首轮上限和重排增益;若答案不在候选中,单纯换更强重排器不能恢复证据。
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
- · 检索增强生成
原书出处 · 1 处
- · 查看原书依据
Rerank The retrieval process can be broken into a two-stage or multi-stage process, where the initial stage retrieves a list of documents relevant to the query, followed by one or more reranking stages that take the documents and sort them by relevance. The reranker is generally a more complex model than the retriever and thus is run only on the retrieved results (or else we
参考来源
- 位置:《Designing Large Language Model Applications》第12章《检索增强生成》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch12.html)
- 结论依据:摘录中关于重排无法挽回首轮漏掉的文档的定义、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:Rerank The retrieval process can be broken into a two-stage or multi-stage process, where the initial stage retrieves a list of documents relevant to the query, followed by one or more reranking stages that take the documents and sort them by relevance. The reranker is generally a more complex model than the retriever and thus is run only on the retrieved results (or else we