LangChain 完整教程

from langchain_community.document_loaders import (
    PyPDFLoader, 
    Docx2txtLoader,
    UnstructuredMarkdownLoader
)
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain_community.vectorstores import Qdrant
from langchain.chains import ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
import os

class KnowledgeBase:
    def __init__(self, collection_name="company_kb"):
        self.embeddings = OpenAIEmbeddings()
        self.vectorstore = Qdrant(
            collection_name=collection_name,
            embeddings=self.embeddings,
            path="./qdrant_db"
        )
        self.llm = ChatOpenAI(model="gpt-4", temperature=0)
        
    def load_documents(self, directory):
        """加载多种格式文档"""
        docs = []
        for filename in os.listdir(directory):
            filepath = os.path.join(directory, filename)
            
            if filename.endswith('.pdf'):
                loader = PyPDFLoader(filepath)
            elif filename.endswith('.docx'):
                loader = Docx2txtLoader(filepath)
            elif filename.endswith('.md'):
                loader = UnstructuredMarkdownLoader(filepath)
            else:
                continue
                
            docs.extend(loader.load())
        
        # 分块
        text_splitter = RecursiveCharacterTextSplitter(
            chunk_size=500,
            chunk_overlap=50,
            separators=["\n\n", "\n", "。", "！", "？", "；", " "]
        )
        
        texts = text_splitter.split_documents(docs)
        
        # 添加到向量库
        self.vectorstore.add_documents(texts)
        print(f"已索引 {len(texts)} 个文档块")
        
    def create_qa_chain(self):
        """创建问答链"""
        memory = ConversationBufferMemory(
            memory_key="chat_history",
            return_messages=True,
            output_key="answer"
        )
        
        qa_chain = ConversationalRetrievalChain.from_llm(
            llm=self.llm,
            retriever=self.vectorstore.as_retriever(
                search_type="mmr",  # 最大边际相关性
                search_kwargs={"k": 5, "fetch_k": 20}
            ),
            memory=memory,
            return_source_documents=True,
            verbose=True
        )
        
        return qa_chain

# 使用示例
kb = KnowledgeBase()
kb.load_documents("./company_docs")
qa_chain = kb.create_qa_chain()

# 提问
result = qa_chain.invoke({"question": "公司的休假政策是什么？"})
print("回答：", result["answer"])
print("来源：", [doc.metadata for doc in result["source_documents"]])