# 支持更多文檔格式
def extract_text_from_file(file_path):
    """根據(jù)文件類型選擇適當(dāng)?shù)奶崛》椒?""
    ext = os.path.splitext(file_path)[1].lower()
   
    if ext == '.pdf':
        return extract_text_from_pdf(file_path)
    elif ext == '.docx':
        return extract_text_from_docx(file_path)
    elif ext == '.pptx':
        return extract_text_from_pptx(file_path)
    elif ext in ['.txt', '.md', '.py', '.java', '.js', '.html', '.css']:
        with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
            return f.read()
    elif ext in ['.csv', '.xlsx', '.xls']:
        return extract_text_from_spreadsheet(file_path)
    else:
        raise ValueError(f"不支持的文件類型: {ext}")

# 多粒度分塊示例
def create_multi_granularity_chunks(text):
    """創(chuàng)建多粒度分塊"""
    # 大塊(1000字符)：適合概述類問(wèn)題
    large_splitter = RecursiveCharacterTextSplitter(
        chunk_size=1000, chunk_overlap=100
    )
    # 中塊(400字符)：平衡粒度
    medium_splitter = RecursiveCharacterTextSplitter(
        chunk_size=400, chunk_overlap=40
    )
    # 小塊(150字符)：適合精確問(wèn)答
    small_splitter = RecursiveCharacterTextSplitter(
        chunk_size=150, chunk_overlap=20
    )
    
    large_chunks = large_splitter.split_text(text)
    medium_chunks = medium_splitter.split_text(text)
    small_chunks = small_splitter.split_text(text)
    
    return {
        "large": large_chunks,
        "medium": medium_chunks,
        "small": small_chunks
    }

# 多模型嵌入示例
def generate_multi_model_embeddings(text):
    """使用多個(gè)模型生成嵌入并融合"""
    model1 = SentenceTransformer('all-MiniLM-L6-v2')
    model2 = SentenceTransformer('paraphrase-multilingual-mpnet-base-v2')
    
    # 生成嵌入
    emb1 = model1.encode(text)
    emb2 = model2.encode(text)
    
    # 簡(jiǎn)單融合策略：歸一化后拼接
    emb1_norm = emb1 / np.linalg.norm(emb1)
    emb2_norm = emb2 / np.linalg.norm(emb2)
    
    # 返回融合嵌入
    return np.concatenate([emb1_norm, emb2_norm])

# 查詢擴(kuò)展示例
def expand_query(query):
    """使用LLM擴(kuò)展查詢，生成多個(gè)角度的查詢變體"""
    prompt = f"""
    為以下查詢生成3個(gè)不同角度的變體，以提高檢索召回率：
    
    原始查詢: {query}
    
    變體應(yīng)當(dāng)包含同義詞替換、關(guān)鍵詞擴(kuò)展或問(wèn)題重構(gòu)。
    僅返回三個(gè)變體，每行一個(gè)，不要有額外解釋。
    """
    
    response = call_llm_api(prompt)
    variants = [line.strip() for line in response.strip().split('\n')]
    return [query] + variants  # 返回原始查詢和變體

# 與現(xiàn)有系統(tǒng)集成示例
from ragflow import RAGPipeline
import your_company_database


# 從公司數(shù)據(jù)庫(kù)獲取文檔
documents = your_company_database.get_latest_documents()


# 使用RAGFlow處理
pipeline = RAGPipeline()
for doc in documents:
    pipeline.add_document(doc)
    
# 將結(jié)果同步回公司系統(tǒng)
processed_results = pipeline.get_indexed_status()
your_company_database.update_document_status(processed_results)

# 自定義混合檢索策略
from ragflow import RetrievalPipeline
from ragflow.retrievers import VectorRetriever, KeywordRetriever


# 創(chuàng)建自定義的檢索器組合
def custom_hybrid_retriever(query, top_k=10):
    # 使用向量檢索獲取語(yǔ)義相關(guān)結(jié)果
    semantic_results = vector_retriever.retrieve(query, top_k=top_k)
    
    # 使用關(guān)鍵詞檢索獲取精確匹配
    keyword_results = keyword_retriever.retrieve(query, top_k=top_k)
    
    # 自定義融合策略（比如考慮文檔及時(shí)性等）
    results = custom_merge_function(semantic_results, keyword_results, 
                                    recency_weight=0.2)
    return results

# 創(chuàng)建復(fù)雜的數(shù)據(jù)處理和RAG工作流
from ragflow import DocumentProcessor, Indexer, QueryEngine
import schedule
import time


def daily_knowledge_update():
    # 從多個(gè)來(lái)源獲取新文檔
    new_docs = fetch_documents_from_sources()
    
    # 文檔預(yù)處理（去重、格式轉(zhuǎn)換等）
    processed_docs = DocumentProcessor().process_batch(new_docs)
    
    # 增量更新索引
    indexer = Indexer()
    indexer.update_incrementally(processed_docs)
    
    # 生成日?qǐng)?bào)
    report = generate_daily_summary()
    send_email(report)


# 設(shè)置定時(shí)任務(wù)
schedule.every().day.at("01:00").do(daily_knowledge_update)


while True:
    schedule.run_pending()
    time.sleep(60)

# 構(gòu)建RAG系統(tǒng)評(píng)估框架
from ragflow import RAGEvaluator
import matplotlib.pyplot as plt


# 準(zhǔn)備測(cè)試數(shù)據(jù)集
test_queries = load_test_queries()
ground_truth = load_ground_truth_answers()


# 測(cè)試不同的配置
configurations = [
    {"embeddings": "minilm", "chunk_size": 200, "retriever": "hybrid"},
    {"embeddings": "e5-large", "chunk_size": 500, "retriever": "vector"},
    {"embeddings": "bge-large", "chunk_size": 300, "retriever": "hybrid"}
]


results = {}
for config in configurations:
    # 構(gòu)建并測(cè)試每種配置
    rag_system = build_rag_with_config(config)
    eval_results = RAGEvaluator.evaluate(
        rag_system, test_queries, ground_truth
    )
    results[str(config)] = eval_results


# 可視化不同配置的性能比較
plot_evaluation_results(results)

# 醫(yī)療領(lǐng)域RAG增強(qiáng)示例
from ragflow import RAGPipeline
from custom_medical_modules import MedicalTermNormalizer, DiseaseEntityExtractor


# 創(chuàng)建領(lǐng)域特定的文檔處理器
medical_processor = DocumentProcessor()
medical_processor.add_transformer(MedicalTermNormalizer())
medical_processor.add_transformer(DiseaseEntityExtractor())


# 構(gòu)建醫(yī)療特定的RAG系統(tǒng)
medical_rag = RAGPipeline(
    document_processor=medical_processor,
    retrieval_augmentors=[
        CitationRetriever(),  # 添加醫(yī)學(xué)文獻(xiàn)引用
        EvidenceLevelClassifier()  # 分類證據(jù)等級(jí)
    ],
    response_formatters=[
        MedicalDisclaimerFormatter(),  # 添加醫(yī)療免責(zé)聲明
        CitationFormatter()  # 格式化引用
    ]
)