Auto-sync: 2026-04-16 17:30
This commit is contained in:
@@ -1,29 +0,0 @@
|
||||
---
|
||||
title: "混合搜索"
|
||||
type: concept
|
||||
tags: [vector-search, information-retrieval, hybrid]
|
||||
date: 2026-04-16
|
||||
---
|
||||
|
||||
## Definition
|
||||
融合多种检索方法的搜索策略,通常结合:
|
||||
1. **Dense Vector**(语义相似度):理解查询意图
|
||||
2. **BM25**(关键词匹配):捕获精确术语
|
||||
3. **RRF**(Reciprocal Rank Fusion):多结果集融合排序
|
||||
|
||||
## Why Hybrid Wins
|
||||
- 纯向量搜索:同义词命中好,但精确术语漏检
|
||||
- 纯 BM25:精确术语好,但无法捕捉语义泛化
|
||||
- 混合:两者互补,RRF 融合排序
|
||||
|
||||
## Formula
|
||||
RRF score for a document d:
|
||||
```
|
||||
RRF(d) = Σ 1/(k + rank_i(d))
|
||||
```
|
||||
其中 k 通常为 60,rank_i 是第 i 种检索方法的排名。
|
||||
|
||||
## Connections
|
||||
- [[memsearch]]:混合搜索的具体实现
|
||||
- [[语义搜索]]:混合搜索的组成部分
|
||||
- [[Personal-Knowledge-Base-RAG]]:RAG 管道中可使用混合搜索
|
||||
Reference in New Issue
Block a user