LlamaIndex高级RAG实战：从检索增强到知识图谱问答
<h2>一、RAG的局限与高级RAG</h2><p>基础RAG（检索增强生成）存在明显短板：检索精度低、缺乏多跳推理、无法处理复杂查询。高级RAG通过查询改写、重排序、知识图谱增强等技术，将RAG从简单检索提升到深度问答。LlamaIndex是构建高级RAG系统的首选框架，提供丰富的索引结构和检索策略。</p>
<h2>二、LlamaIndex核心架构</h2>
<pre><code>核心组件：
- Document/Node：文档与分片
- Index：索引（向量/关键词/知识图谱）
- Retriever：检索器
- ResponseSynthesizer：响应合成器
- QueryEngine：查询引擎
- Tool/Agent：工具与智能体</code></pre>
<h2>三、环境搭建</h2>
<pre><code>pip install llama-index llama-index-llms-openai
pip install llama-index-embeddings-openai
pip install llama-index-graph-stores-nebula
import os
os.environ["OPENAI_API_KEY"] = "your-api-key"</code></pre>
<h2>四、基础RAG vs 高级RAG对比</h2>
<pre><code>from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
# 基础RAG
documents = SimpleDirectoryReader("data").load_data()
index = VectorStoreIndex.from_documents(documents)
query_engine = index.as_query_engine()
response = query_engine.query("什么是微服务架构？")
print(response)
# 高级RAG：带检索后处理
from llama_index.core.postprocessor import SentenceTransformerRerank
rerank = SentenceTransformerRerank(top_n=3, model="cross-encoder/ms-marco-MiniLM-L-2-v2")
query_engine = index.as_query_engine(
similarity_top_k=10, # 先检索10个
node_postprocessors= # 再重排取前3
)
response = query_engine.query("微服务和单体架构的核心区别是什么？")
print(response)</code></pre>
<h2>五、查询改写：HyDE技术</h2>
<pre><code>from llama_index.core.indices.query.query_transform import HyDEQueryTransform
from llama_index.core.query_engine import TransformQueryEngine
# HyDE：先让LLM生成假设性文档，再用假设文档做检索
hyde = HyDEQueryTransform(include_original=True)
query_engine = index.as_query_engine()
hyde_query_engine = TransformQueryEngine(query_engine, hyde)
# 对比效果
question = "如何设计高并发系统？"
normal = query_engine.query(question)
hyde_result = hyde_query_engine.query(question)
print("普通RAG:", normal)
print("HyDE RAG:", hyde_result)</code></pre>
<h2>六、多跳推理：子问题分解</h2>
<pre><code>from llama_index.core.query_engine import SubQuestionQueryEngine
from llama_index.core.tools import QueryEngineTool, ToolMetadata
# 为不同文档集创建独立索引
sql_index = VectorStoreIndex.from_documents(sql_docs)
java_index = VectorStoreIndex.from_documents(java_docs)
sql_tool = QueryEngineTool(
query_engine=sql_index.as_query_engine(),
metadata=ToolMetadata(name="sql_docs", description="SQL优化相关文档")
)
java_tool = QueryEngineTool(
query_engine=java_index.as_query_engine(),
metadata=ToolMetadata(name="java_docs", description="Java性能优化文档")
)
# 子问题分解引擎
sub_engine = SubQuestionQueryEngine.from_defaults(query_engine_tools=)
# 复杂查询会自动分解为子查询
response = sub_engine.query(
"如何优化Java应用中的数据库查询性能？需要同时考虑Java层面和SQL层面"
)
print(response)</code></pre>
<h2>七、知识图谱RAG</h2>
<pre><code>from llama_index.core import KnowledgeGraphIndex
from llama_index.core.graph_stores import SimpleGraphStore
# 构建知识图谱索引
graph_store = SimpleGraphStore()
kg_index = KnowledgeGraphIndex.from_documents(
documents,
max_triplets_per_chunk=5,
graph_store=graph_store,
include_embeddings=True
)
# 知识图谱查询（支持多跳关系推理）
kg_query_engine = kg_index.as_query_engine(
include_text=True,
response_mode="tree_summarize",
embedding_mode="hybrid",
similarity_top_k=5
)
response = kg_query_engine.query(
"Spring Boot自动配置的完整流程是什么？涉及哪些核心注解？"
)
print(response)</code></pre>
<h2>八、混合检索：向量+关键词+知识图谱</h2>
<pre><code>from llama_index.core.retrievers import QueryFusionRetriever
# 向量检索器
vector_retriever = index.as_retriever(similarity_top_k=5)
# 关键词检索器
keyword_retriever = index.as_retriever(
similarity_top_k=5,
retriever_mode="keyword"
)
# 融合检索器（Reciprocal Rank Fusion）
fusion_retriever = QueryFusionRetriever(
retrievers=,
num_queries=3, # 生成3个改写查询
similarity_top_k=10,
mode="reciprocal_rerank"
)
nodes = fusion_retriever.retrieve("分布式事务如何保证一致性？")
for node in nodes:
print(f"Score: {node.score:.4f} | {node.text[:80]}")</code></pre>
<h2>九、与Spring Boot集成</h2>
<pre><code>@Service
public class AdvancedRAGService {
private final RestTemplate restTemplate = new RestTemplate();
@Value("${llama-index.service.url}")
private String llamaServiceUrl;
public String query(String question, String mode) {
Map<String, Object> body = Map.of(
"question", question,
"mode", mode, // basic / hyde / sub_question / kg
"top_k", 5
);
HttpHeaders headers = new HttpHeaders();
headers.setContentType(MediaType.APPLICATION_JSON);
HttpEntity<Map<String, Object>> entity = new HttpEntity<>(body, headers);
ResponseEntity<map> resp = restTemplate.exchange(
llamaServiceUrl + "/query", HttpMethod.POST, entity, Map.class);
return (String) resp.getBody().get("answer");
}
}</map></code></pre>
<h2>十、评估与优化</h2>
<pre><code>from llama_index.core.evaluation import FaithfulnessEvaluator, RelevancyEvaluator
# 评估忠实度（回答是否基于检索内容）
faith_eval = FaithfulnessEvaluator(llm=llm)
# 评估相关性（回答是否切题）
rel_eval = RelevancyEvaluator(llm=llm)
# 批量评估
questions = ["什么是RAG?", "向量数据库如何选择?", "知识图谱如何构建?"]
for q in questions:
response = query_engine.query(q)
faith_result = faith_eval.evaluate_response(query=q, response=response)
rel_result = rel_eval.evaluate_response(query=q, response=response)
print(f"Q: {q}")
print(f"忠实度: {faith_result.passing}")
print(f"相关性: {rel_result.passing}")</code></pre>
<h2>十一、最佳实践</h2>
<ol>
<li><strong>分块策略</strong>：根据文档类型选择合适的chunk_size（256-1024）</li>
<li><strong>混合检索</strong>：向量+关键词融合效果优于单一检索</li>
<li><strong>重排序</strong>：检索top_k大，rerank后取小，精度更高</li>
<li><strong>知识图谱</strong>：多跳推理场景必须用知识图谱增强</li>
<li><strong>持续评估</strong>：用Faithfulness和Relevancy指标持续监控质量</li>
</ol>
<h2>十二、总结</h2>
<p>高级RAG是基础RAG的全面升级。通过查询改写（HyDE）、重排序、子问题分解、知识图谱增强、混合检索等技术，可以构建真正可用的企业级问答系统。LlamaIndex提供了完整的工具链，配合Spring Boot可快速落地。</p>
</div>
<div id="MySignature" role="contentinfo">
---
📌 **如果觉得文章对你有帮助，欢迎点赞👍收藏⭐！**
💬 有问题或建议？欢迎在评论区留言讨论~
🔗 更多技术干货请关注作者：弥烟袅绕
📚 本文地址：https://www.cnblogs.com/czlws/p/19863106/llamaindex-advanced-rag-knowledge-graph-tutorial<br><br>
来源：https://www.cnblogs.com/czlws/p/19863106/llamaindex-advanced-rag-knowledge-graph-tutorial
頁:
[1]