Query Rewrite – mainri

Query Rewrite / HyDE

What is Query Rewrite

Query Rewrite is a pre-retrieval technique that transforms the user’s original query into one or more alternative query forms before sending them to the retrieval system.
查询重写是一种检索前技术，在将用户原始查询发送到检索系统之前，将其转换为一种或多种替代查询形式。

Imagine you’re asking a librarian for help. Instead of just saying “books about AI,” you might also say “machine learning texts,” “neural network resources,” and “deep learning guides” — because different books use different terminology. Query Rewrite is the LLM doing exactly that: generating multiple ways to ask the same question so the search system has better chances of finding relevant documents.

CN: 想象你在问图书管理员。你不仅说“AI相关的书”，还会说“机器学习书籍”、“神经网络资料”、“深度学习指南”——因为不同的书用不同的术语。查询重写就是让LLM做同样的事：生成多种问法，让检索系统有更大机会找到相关文档。

WHat is HyDE

Hypothetical Document Embeddings，假设性文档嵌入

HyDE is a specific query rewriting technique that generates a hypothetical (fake) answer document to the user’s query first, then embeds that generated document for similarity search — instead of embedding the query directly.

CN: HyDE 是一种特定的查询重写技术，它先生成一个假设性（虚构的）答案文档来回答用户查询，然后对该生成文档做嵌入用于相似度搜索——而不是直接嵌入原始查询。

Instead of asking “where can I find Italian food?” and searching for that question, HyDE first writes a fake Yelp review: “This Italian restaurant has amazing pasta and tiramisu…” — then searches for documents that look like that review. Because the review looks more like actual documents in the database than the question does.

CN: 不是直接问“哪里有好吃的意大利菜？”然后去搜这个问题。HyDE 先写一篇假的Yelp评论：“这家意大利餐厅的意面和提拉米苏超赞…” —— 然后去搜和这篇评论相似的文档。因为评论比问题本身更像数据库里的真实文档。

Query Rewrite 的核心内容

4.1 查询重写的四种主要类型

类型	EN	CN
改写 (Rewriting)	Rephrase the query for better embedding alignment	改写查询以更好地对齐嵌入
扩展 (Expansion)	Add semantically related terms	添加语义相关的术语
分解 (Decomposition)	Split complex queries into sub-queries	将复杂查询拆分为子查询
多路生成 (Multi-Query)	Generate multiple query variants	生成多个查询变体

4.2 多查询检索 (Multi-Query Retrieval) – Fan-Out

EN: Generate multiple query variants from the original, run them in parallel against the vector DB, merge and deduplicate results. This covers more angles of the user’s intent.

CN: 从原始查询生成多个变体，并行发送到向量数据库，合并并去重结果。这覆盖了用户意图的更多角度。

4.3 RRF (Reciprocal Rank Fusion)

EN: When merging results from multiple query variants, RRF scores documents by their rank positions rather than raw similarity scores — documents that appear high in multiple result lists get boosted.

CN: 合并多路查询结果时，RRF 按文档的排名位置而非原始相似度分数来打分——在多路结果中都排名靠前的文档获得加成。

HyDE works step

HyDE works in two steps:

Generate: Given a query, zero-shot prompt an LLM to generate a hypothetical document that answers the query
Embed & Retrieve: Encode that hypothetical document and use it to retrieve real documents via vector similarity

CN: HyDE 分两步工作:

生成： 给定查询，用零样本提示让 LLM 生成一个回答该查询的假设性文档
嵌入与检索： 编码该假设性文档，用其向量通过相似度检索真实文档

Why this works: The hypothetical document is in the same style as real documents (declarative, detailed, expository). So its embedding lands closer to real document embeddings than the query embedding would.

CN – 为什么有效： 假设性文档与真实文档风格相同（陈述句、详细、说明性）。所以它的向量比查询向量更接近真实文档的向量。

Query Rewrite vs HyDE 对比

维度	EN	CN
Query Rewrite	Rewrites the question	改写问题
HyDE	Generates a fake answer then embeds that	生成假答案然后嵌入假答案
Query Rewrite 目标	Make the query easier to retrieve	让查询更容易被检索
HyDE 目标	Make the embedding closer to document space	让向量更接近文档空间
Query Rewrite 开销	1 LLM call + N vector searches	1次LLM调用 + N次向量搜索
HyDE 开销	1 LLM call + 1 vector search	1次LLM调用 + 1次向量搜索

Key Takeaways

要点	EN	CN
Query Rewrite 是检索前技术	Query Rewrite is a pre-retrieval technique	查询重写是检索前技术
HyDE 生成假设文档再嵌入	HyDE generates hypothetical doc then embeds it	HyDE 生成假设文档再嵌入
HyDE 论文推荐生成5个文档取平均	HyDE paper recommends generating 5 docs and averaging	HyDE论文推荐生成5个文档取平均
HyDE 的温度推荐0.7	HyDE recommended temperature is 0.7	HyDE推荐temperature为0.7
Multi-Query 用 asyncio.gather() 并行	Use asyncio.gather() for parallel multi-query	用 asyncio.gather() 实现多查询并行
RRF 按排名位置而非分数融合	RRF fuses by rank position, not raw scores	RRF按排名位置而非原始分数融合
HyDE 弥合”问题空间”与”答案空间”的鸿沟	HyDE bridges the “question space” vs “answer space” gap	HyDE弥合”问题空间”与”答案空间”的鸿沟
Query Rewrite 提升召回率，HyDE 提升精确率	Query Rewrite improves recall, HyDE improves precision	查询重写提升召回率，HyDE提升精确率

Advanced RAG

What is Advanced RAG?

Advanced RAG (Advanced Retrieval-Augmented Generation) is an evolutionary upgrade over the basic “Naive RAG” pipeline. It adds a suite of optimization techniques at every stage of the RAG workflow — pre-retrieval, retrieval, post-retrieval, and evaluation — to systematically improve retrieval precision, recall, and generation quality.

Naive RAG is like sending a junior intern to the library with a single, vague question, grabbing the first 5 books they find, and copying paragraphs directly. Advanced RAG is like sending a team of expert researchers who:

Rephrase the question in 5 different ways (Query Rewrite)
Write a fake “perfect answer” first to know exactly what to look for (HyDE)
Search both the card catalog AND the full-text index (Hybrid Search)
Have a senior editor re-rank the top results (Reranker)
Summarize and compress the findings before presenting them (Context Compression)
And finally, run a quality check on the entire process (Evaluation)

CN: 朴素 RAG 就像派一个初级实习生去图书馆，带着一个模糊的问题，抓起前5本书，直接抄段落。Advanced RAG 就像派一队专家研究员：

把问题用5种不同方式重新表述（查询重写）
先写一篇“完美的假答案”来明确要找什么（HyDE）
同时搜索卡片目录和全文索引（混合检索）
让高级编辑对结果重新排名（重排序器）
在呈现之前总结和压缩发现（上下文压缩）
最后对整个流程做质量检查（评估）

What does Advanced RAG include?

Layer 1: Chunking Strategies

Chunking 策略

EN	CN
Recursive Chunking: Split text by paragraphs/sentences with overlap	递归分块：按段落/句子分割，带重叠
Semantic Chunking: Split by semantic boundaries using embedding similarity	语义分块：用嵌入相似度按语义边界分割
Parent-Child Retrieval: Retrieve child chunks, return parent chunks for context	父子检索：检索子块，返回父块作为上下文

Layer 2:Retrieval Strategies

检索策略

EN	CN
Hybrid Search: Combine dense (vector) + sparse (BM25) retrieval	混合检索：结合稠密（向量）+ 稀疏（BM25）检索
RRF (Reciprocal Rank Fusion): Merge multi-strategy results by rank	RRF（倒数排名融合）：按排名位置合并多路结果
Vector DB Selection: Chroma, Pinecone, Milvus, Azure AI Search	向量数据库选型：Chroma, Pinecone, Milvus, Azure AI Search

Layer 3: Post-Retrieval Optimization

检索后优化

EN	CN
Reranker: Cross-encoder models to re-score retrieved documents	重排序器：用交叉编码器模型重新给检索文档打分
Query Rewrite: Rephrase queries for better embedding alignment	查询重写：改写查询以更好地对齐嵌入
HyDE: Generate hypothetical answer → embed that → retrieve	HyDE：生成假设答案 → 嵌入假设答案 → 检索
Multi-Query Retrieval: Generate multiple query variants, search in parallel	多查询检索：生成多个查询变体，并行搜索
Context Compression: Summarize or filter retrieved docs to fit context window	上下文压缩：总结或过滤检索文档以适应上下文窗口

Layer 4: Evaluation

评估体系

EN	CN
RAG Eval Metrics (ragas): Context Relevancy, Answer Relevancy, Faithfulness	RAG评估指标（ragas）：上下文相关性、答案相关性、忠实度
Golden Dataset: Human-annotated Q&A pairs for regression testing	黄金数据集：人工标注的问答对，用于回归测试

Advanced RAG differences from Naive RAG

维度	EN	CN
检索次数	Naive: 1 single retrieval call. Advanced: Multiple parallel retrievals + reranking.	朴素：单次检索调用。Advanced：多次并行检索 + 重排序。
查询处理	Naive: Embed query directly. Advanced: Query Rewrite / HyDE / Multi-Query.	朴素：直接嵌入查询。Advanced：查询重写 / HyDE / 多查询。
检索类型	Naive: Vector search only. Advanced: Hybrid (Vector + BM25 + RRF).	朴素：仅向量搜索。Advanced：混合（向量 + BM25 + RRF）。
排序	Naive: Raw similarity scores. Advanced: Cross-encoder reranker (more accurate).	朴素：原始相似度分数。Advanced：交叉编码器重排序器（更精确）。
分块	Naive: Fixed-size chunking (e.g., 512 tokens). Advanced: Semantic / Parent-Child.	朴素：固定大小分块（如512 token）。Advanced：语义 / 父子分块。
上下文窗口利用	Naive: Dump all retrieved chunks into prompt. Advanced: Compress / filter / summarize.	朴素：把所有检索块塞进提示词。Advanced：压缩 / 过滤 / 总结。
评估	Naive: None or manual. Advanced: Automated Eval with ragas + Golden Dataset.	朴素：无或手动。Advanced：用 ragas + 黄金数据集自动评估。
失败处理	Naive: If retrieval fails, LLM hallucinates. Advanced: Query rewrite retries, multi-query fallback.	朴素：检索失败则LLM幻觉。Advanced：查询重写重试、多查询降级。

Example: You’re building a customer support RAG for a cloud provider’s documentation.

阶段	Naive RAG	Advanced RAG
用户查询	“how fix 503 error”	“how fix 503 error”
处理	直接嵌入查询	Query Rewrite → “how to troubleshoot HTTP 503 service unavailable error”, “503 error resolution steps”
检索	向量搜索返回5个块，其中3个是关于”503″但不相关的（如负载均衡器通用文档）	混合检索(BM25+向量) + RRF，返回15个候选
重排	无	Cross-encoder reranker 重新打分，选出最相关的3个块
上下文	5个块全部塞入（可能超过窗口）	Context compression 总结每个块，只保留关键步骤
生成	答案可能泛泛而谈	答案精确指出: “检查service mesh sidecar, 重启x, 检查Y”

RAG 2.0

RAG (Retrieval-Augmented Generation) is an AI architecture that retrieves relevant information from external knowledge sources and provides it to an LLM, enabling the model to generate accurate, up-to-date, and context-aware responses.

LLM = Answer from Training
while
RAG = Search First + Then Answer

What is RAG 2.0? There is no official industry standard that defines “RAG 1.0”, “RAG 2.0”, or “RAG 3.0”.

The concept of “RAG 2.0”, known as Agentic RAG, represents a shift from a “simple data retrieval plugin” to a “deeply integrated, agentic system with reasoning capabilities.” The industry has indeed evolved from 1.0 and is now progressing toward 3.0.

Simplest Way to Remember

Version	Meaning
RAG 1.0	Search Once
RAG 2.0	Search + Reason
RAG 3.0 (coming ? on the way?)	Agent + Search + Tools + Workflow

RAG Foundation: Embedding

Embedding is a fundamental technology in machine learning and natural language processing that transforms discrete or complex objects (such as words, sentences, or images) into numerical vector representations of a fixed dimension

more …

RAG Foundation: RAG Pipeline

A simple RAG pipeline has two phases and five stages

Phase 1: Indexing (Offline / 离线阶段)
Build the searchable knowledge base before users ask questions.
离线阶段（索引）：先把所有文档加载进来，切分成小块，调用 Embedding 模型转成向量，最后存进 FAISS 等向量数据库里。
Phase 2: Retrieval & Generation (Online / 在线阶段)
Execute at query time for each user question.
在线阶段（检索+生成）：用户提问时，先把问题也转成向量，去库里找最相似的 Top-K 个块，把这几个块作为“参考资料”连同问题一起扔给 GPT，让它写出最终回答。

more …

In AI, a chunking strategy is the method used to break down large pieces of information—like long documents, complex prompts, or even audio and video—into smaller, more manageable segments called “chunks”. These chunks are the fundamental units that AI systems, especially Large Language Models (LLMs), work with to understand, process, and retrieve information efficiently.

Common Chunking Strategies

The best strategy depends on the type of data and the specific task. Here are the most common approaches:

Strategy	Description	Best For
Fixed-Size Chunking	Splits text into chunks of a predetermined size (e.g., a specific number of words, characters, or tokens). Often uses a “sliding window” with overlap to preserve context across boundaries.	Simple implementation; works well when document structure is uniform and not critical.
Semantic Chunking	Groups text based on meaning, using algorithms to find natural topic boundaries and keep related ideas together.	Maintaining the coherence of ideas within each chunk for better understanding.
Structure-Aware Chunking	Respects the natural format of the document, splitting at points like paragraphs, sentences, or headers.	Documents with clear structure (e.g., articles, reports, code) where breaking mid-section would lose meaning.
Recursive Chunking	Starts with large chunks and recursively splits them into smaller ones until they meet a target size, trying to respect natural boundaries like paragraphs or sentences.	A balanced approach that aims to create the largest possible meaningful chunks.
Multimodal Chunking	Extends the concept to non-text data, such as segmenting audio by silences, video by scene changes, or identifying objects within images.	AI systems that process images, audio, and video, not just text

Chunking Strategy: Recursive Chunking

Recursive Chunking is a hierarchical text splitting strategy that uses a priority list of separators (e.g., ["\n\n", "\n", " ", """]), starting with the highest-level (semantically strongest) separator. If a chunk still exceeds the chunk_size limit after splitting, it recursively applies the next-level separator to that chunk, continuing until all chunks meet the size requirement.

more …

Chunking Strategy: Semantic Chunking

Semantic Chunking is a strategy for splitting documents into smaller pieces (chunks) based on meaning, rather than on fixed character counts or simple separators. It tries to keep sentences or paragraphs that are about the same topic together, and split where the topic changes. Think of it as “a smart editor who knows where one idea ends and the next begins

more …

Chunking Strategy: Parent-Child Retrieval

Parent-Child Retrieval (also known as Small-to-Big Retrieval) is a two-tier hierarchical indexing strategy.

more …

Hybrid Search refers to a search technique that combines multiple search algorithms simultaneously to retrieve the most relevant results. It most commonly merges Lexical (Keyword) Search with Semantic (Vector/Dense) Search.
基于关键词的搜索（词汇搜索）与基于语义的搜索（向量/稠密搜索）

Lexical Search (Sparse Retrieval): Typically powered by algorithms like BM25 or TF-IDF. It relies on exact keyword matching and statistical term frequency. It excels at finding specific proper nouns, IDs, or rare terminology (e.g., “error code 404”).
词汇搜索（稀疏检索）：通常由 BM25 或 TF-IDF 等算法驱动。它依赖于精确的关键词匹配和统计词频。它在查找特定的专有名词、ID 或罕见术语时表现出色（例如：“错误代码 404”）。
Semantic Search (Dense Retrieval): Powered by embedding models (e.g., Sentence-BERT). It converts text into high-dimensional vectors and retrieves documents based on “meaning” rather than exact words. It excels at understanding synonyms, context, and natural language queries (e.g., “How to fix a broken internet connection”).
语义搜索（稠密检索）：由嵌入模型（如 Sentence-BERT）驱动。它将文本转换为高维向量，并根据“含义”而非精确词汇来检索文档。它在理解同义词、上下文和自然语言查询（例如：“如何修复断开的网络连接”）方面表现出色。

BM25 (keyword-based search)

BM25 (Best Matching 25) is a keyword-based ranking algorithm used in information retrieval to score and rank documents based on their relevance to a search query. It’s called “Best Matching 25” because it was the 25th variant in a series of scoring functions proposed by its creators. BM25 is the default ranking algorithm in Elasticsearch and most production search engines

more …

RRF

RRF stands for Reciprocal Rank Fusion. It is an algorithm that merges multiple ranked result lists from different search systems into a single unified ranking.
全称是 Reciprocal Rank Fusion（倒数排名融合）。它是一种将来自不同检索系统的多个排名结果列表合并成一个统一排名的算法.

more …

Vector Database Selection

Compare normal DB, such as SQL DB, MySQL, PostgreSQL, They fundamental difference is what they search for and how they find it. A traditional SQL database (like MySQL, PostgreSQL without pgvector) is built for exact matching and structured queries. It answers questions like: “Find the customer with ID = 12345” or “Give me all orders over $100.” A Vector Database is built for semantic similarity and unstructured data. It answers questions like: “Find all documents that talk about the same topic as this paragraph” or “Show me products that look visually similar to this image.”

Normal DB vs Vector DB

要点	EN	CN
核心：精确匹配 vs. 语义相似度	Core: Exact matching vs. Semantic similarity	核心：精确匹配 vs. 语义相似度
SQL存结构化数据（数字/字符串），Vector存浮点数数组（含义）	SQL stores structured data (numbers/strings); Vector stores float arrays (meaning)	SQL存结构化数据，Vector存浮点数数组（含义）
SQL查询用WHERE精确条件；Vector查询用ORDER BY距离	SQL queries use WHERE exact conditions; Vector queries use ORDER BY distance	SQL查询用WHERE精确条件；Vector查询用ORDER BY距离
SQL用B-Tree/哈希（精确查找）；Vector用HNSW/IVF（近似查找）	SQL uses B-Tree/Hash (exact lookup); Vector uses HNSW/IVF (approximate)	SQL用B-Tree/哈希（精确）；Vector用HNSW/IVF（近似）
SQL结果是二元的（匹配/不匹配）；Vector结果是排序的（相似度分数）	SQL results are binary (match/no match); Vector results are ranked (similarity scores)	SQL结果是二元的；Vector结果是排序的
SQL适合事务、财务报表；Vector适合RAG、推荐、AI	SQL suits transactions, ledgers; Vector suits RAG, recommendations, AI	SQL适合事务、报表；Vector适合RAG、推荐、AI
专用向量库不能做JOIN和ACID；但pgvector可以在PostgreSQL中兼得	Dedicated vector DBs can’t do JOINs/ACID; pgvector lets you have both in PostgreSQL	专用向量库不能做JOIN和ACID；pgvector可以让两者兼得
在实际RAG中，两者是互补的，不是替代关系	In real-world RAG, they are complementary, not replacements	在实际RAG中，两者是互补的，不是替代关系

Vector database selection is the process of choosing the right vector database technology for your AI application from dozens of available options — Milvus, Qdrant, Weaviate, Pinecone, pgvector, Chroma, and more.

Mainstream Vector Database Landscape

Category	Examples	CN
Fully Managed (PaaS)	Pinecone, Zilliz Cloud, Weaviate Cloud	全托管云服务
Self-Hosted Open Source	Qdrant, Milvus, Weaviate, Chroma	自托管开源
Database Extensions	pgvector, MongoDB Vector Search, Elasticsearch	数据库扩展
Cloud Provider Services	Azure AI Search, AWS S3 Vectors, Tencent Cloud VDB	云厂商服务
Embedded/Specialized	SQLite (vector), LanceDB	嵌入式/专用

Detailed Comparison

Database	Avg Query Time	Cost (1M @ 1536-dim)	Best For	CN 最适合
Milvus/Zilliz	50.7ms	$115/mo	Fastest queries + good flexibility	最快查询+灵活性好
Weaviate	51.7ms	$160/mo	Native datetime/geo + hybrid search	原生时间/地理+混合检索
Qdrant	73.1ms	$103/mo	Best balance (speed + flexibility + cost)	最佳平衡（速度+灵活性+成本）
Pinecone	106.3ms	$30/mo	Cheapest (⚠️ poor schema flexibility)	最便宜（⚠️ Schema灵活性差）
Chroma	275.4ms	$139/mo	Easiest setup + prototyping	最简单设置+原型开发

7 Types of Data Stored in VectorDB in AI Projects

类型	EN	CN	更新频率	过滤器	主要用途
RAG 文档块	RAG Document Chunks	RAG 文档块	低	source, page	问答
用户记忆 (Mem0)	User Memory (Mem0)	用户记忆 (Mem0)	高	user_id	个性化
工具 Schema	Tool Schemas	工具 Schema	中	tool_category	工具选择
Agent 轨迹	Agent Trajectories	Agent 轨迹	中	user_id, outcome	经验复用
黄金数据集	Golden Dataset	黄金数据集	低	category	评估
语义缓存	Semantic Cache	语义缓存	中	无	降本提速
代码索引	Code Index	代码索引	低	language, path	代码生成

Reranker is a sophisticated machine learning model designed to refine and reorder a list of candidate items—such as search results, document passages—to maximize their relevance to a specific query or context.

Reranker（重排序器）是一种复杂的机器学习模型，旨在优化并重新排序候选项目列表（如搜索结果、文档片段），以最大程度提高它们与特定查询或上下文的相关性。

Reranker

Reranker实战

Advanced RAG

Query Rewrite

Multi Query Retrieval

Context Compression

RAG Evaluation and ragas

Golden Dataset