{"id":121349,"date":"2026-11-02T15:18:01","date_gmt":"2026-11-02T07:18:01","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/121349.html"},"modified":"2026-11-02T15:18:01","modified_gmt":"2026-11-02T07:18:01","slug":"%e5%b0%8f%e7%99%bd%e7%a8%8b%e5%ba%8f%e5%91%98%e5%bf%85%e5%a4%87%ef%bc%9a%e6%94%b6%e8%97%8f%e8%bf%99%e4%bb%bdrag%e7%9f%a5%e8%af%86%e9%97%ae%e7%ad%94%e7%b3%bb%e7%bb%9f%e5%bc%80%e5%8f%91%e5%85%a8","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/121349.html","title":{"rendered":"\u5c0f\u767d\u7a0b\u5e8f\u5458\u5fc5\u5907\uff1a\u6536\u85cf\u8fd9\u4efdRAG\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u5f00\u53d1\u5168\u653b\u7565\uff08\u9644\u4ee3\u7801\u793a\u4f8b\uff09"},"content":{"rendered":"

\u672c\u6587\u4ece\u5de5\u7a0b\u5b9e\u8df5\u89d2\u5ea6\u7cfb\u7edf\u68b3\u7406\u4e86RAG（\u68c0\u7d22\u589e\u5f3a\u751f\u6210）\u7cfb\u7edf\u7684\u5168\u6d41\u7a0b\u5f00\u53d1\u6b65\u9aa4\u4e0e\u8c03\u4f18\u7b56\u7565，\u5e2e\u52a9\u8bfb\u8005\u6784\u5efa\u5b9e\u7528\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u3002\u5185\u5bb9\u6db5\u76d6\u6587\u6863\u5904\u7406\u3001\u6587\u672c\u5206\u5757\u3001\u5411\u91cf\u5316\u3001\u5411\u91cf\u6570\u636e\u5e93\u3001\u68c0\u7d22\u7b56\u7565\u3001\u751f\u6210\u4f18\u5316\u3001\u8bc4\u4f30\u4f53\u7cfb\u7b49\u6838\u5fc3\u73af\u8282，\u6bcf\u4e2a\u73af\u8282\u914d\u6709\u4ee3\u7801\u793a\u4f8b\u548c\u8c03\u4f18\u8981\u70b9，\u9002\u5408\u6709\u4e00\u5b9a\u5f00\u53d1\u57fa\u7840\u7684AI\u5de5\u7a0b\u5e08\u548c\u67b6\u6784\u5e08\u5b66\u4e60\u3002<\/p>\n

\u5f53\u6211\u4eec\u8ba9\u5927\u6a21\u578b\u56de\u7b54"\u516c\u53f8\u4e0a\u5b63\u5ea6\u7684\u8425\u6536\u662f\u591a\u5c11"\u6216"\u4ea7\u54c1X\u7684\u4f7f\u7528\u9650\u5236\u662f\u4ec0\u4e48"\u65f6，\u5927\u6a21\u578b\u5f80\u5f80\u65e0\u6cd5\u7ed9\u51fa\u51c6\u786e\u7b54\u6848\u2014\u2014\u56e0\u4e3a\u8fd9\u4e9b\u79c1\u57df\u77e5\u8bc6\u5e76\u4e0d\u5728\u5176\u8bad\u7ec3\u6570\u636e\u4e2d\u3002RAG（Retrieval-Augmented Generation，\u68c0\u7d22\u589e\u5f3a\u751f\u6210）\u6b63\u662f\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\u800c\u8bde\u751f\u7684\u6280\u672f\u8303\u5f0f\u3002<\/p>\n

\u4e00\u3001RAG\u7cfb\u7edf\u6574\u4f53\u67b6\u6784<\/p>\n

\n

\u4e00\u4e2a\u5b8c\u6574\u7684RAG\u7cfb\u7edf\u53ef\u4ee5\u5206\u4e3a\u79bb\u7ebf\u7d22\u5f15\u548c\u5728\u7ebf\u68c0\u7d22\u4e24\u5927\u9636\u6bb5，\u5305\u542b\u4ee5\u4e0b\u6838\u5fc3\u6a21\u5757：<\/p>\n

\"\u56fe\u7247\"<\/p>\n

\u7406\u89e3\u4e86\u6574\u4f53\u67b6\u6784，\u6211\u4eec\u63a5\u4e0b\u6765\u9010\u4e00\u6df1\u5165\u6bcf\u4e2a\u73af\u8282\u7684\u5b9e\u73b0\u7ec6\u8282\u4e0e\u4f18\u5316\u6280\u5de7\u3002<\/p>\n

\u4e8c\u3001\u6587\u6863\u5904\u7406\u4e0e\u89e3\u6790<\/p>\n

\n

\u6587\u6863\u5904\u7406\u662fRAG\u7cfb\u7edf\u7684"\u5730\u57fa"\u3002\u4f4e\u8d28\u91cf\u7684\u6587\u6863\u89e3\u6790\u4f1a\u5bfc\u81f4\u540e\u7eed\u6240\u6709\u73af\u8282\u6548\u679c\u5927\u6253\u6298\u6263\u3002\u8fd9\u4e00\u6b65\u7684\u6838\u5fc3\u76ee\u6807\u662f\u5c06\u5404\u79cd\u683c\u5f0f\u7684\u539f\u59cb\u6587\u6863（PDF\u3001Word\u3001HTML\u3001Markdown\u7b49）\u8f6c\u5316\u4e3a\u5e72\u51c0\u3001\u7ed3\u6784\u5316\u7684\u7eaf\u6587\u672c\u3002<\/p>\n

2.1 \u5e38\u89c1\u6587\u6863\u683c\u5f0f\u5904\u7406\u65b9\u6848<\/h4>\n\n\u6587\u6863\u683c\u5f0f\u63a8\u8350\u5de5\u5177\u6ce8\u610f\u4e8b\u9879<\/tr>\n\n\n\n\n\n\n
PDF<\/td>\nPyMuPDF \/ pdfplumber<\/td>\n\u626b\u63cf\u4ef6\u9700\u8981OCR\u9884\u5904\u7406<\/td>\n<\/tr>\n
Word<\/td>\npython-docx \/ mammoth<\/td>\n\u6ce8\u610f\u4fdd\u7559\u6807\u9898\u5c42\u7ea7<\/td>\n<\/tr>\n
HTML<\/td>\nBeautifulSoup \/ trafilatura<\/td>\n\u53bb\u9664\u5bfc\u822a\u3001\u5e7f\u544a\u7b49\u566a\u97f3<\/td>\n<\/tr>\n
\u8868\u683c\u6570\u636e<\/td>\npandas \/ camelot<\/td>\n\u8868\u683c\u8f6c\u4e3a\u81ea\u7136\u8bed\u8a00\u63cf\u8ff0<\/td>\n<\/tr>\n
\u56fe\u7247\/\u626b\u63cf\u4ef6<\/td>\nPaddleOCR \/ Azure Doc AI<\/td>\n\u7248\u9762\u5206\u6790+OCR\u7ec4\u5408<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

2.2 \u6587\u6863\u6e05\u6d17\u8981\u70b9<\/h4>\n

1\u3001\u53bb\u9664\u566a\u97f3：\u9875\u7709\u9875\u811a\u3001\u9875\u7801\u3001\u6c34\u5370\u3001\u5e7f\u544a\u5185\u5bb9\u3001\u5bfc\u822a\u680f\u7b49\u975e\u6b63\u6587\u4fe1\u606f<\/p>\n

2\u3001\u4fdd\u7559\u7ed3\u6784：\u6807\u9898\u5c42\u7ea7\u3001\u5217\u8868\u3001\u8868\u683c\u7b49\u7ed3\u6784\u5316\u4fe1\u606f\u5bf9\u540e\u7eed\u5206\u5757\u548c\u68c0\u7d22\u81f3\u5173\u91cd\u8981<\/p>\n

3\u3001\u5143\u6570\u636e\u63d0\u53d6：\u6587\u6863\u6807\u9898\u3001\u4f5c\u8005\u3001\u65e5\u671f\u3001\u7ae0\u8282\u4fe1\u606f\u7b49，\u7528\u4e8e\u540e\u7eed\u8fc7\u6ee4\u548c\u5f15\u7528<\/p>\n

4\u3001\u7279\u6b8a\u5185\u5bb9\u5904\u7406：\u4ee3\u7801\u5757\u3001\u516c\u5f0f\u3001\u56fe\u8868\u9700\u8981\u7279\u6b8a\u6807\u8bb0\u6216\u8f6c\u5316\u4e3a\u6587\u5b57\u63cf\u8ff0<\/p>\n

# \u793a\u4f8b：\u4f7f\u7528 PyMuPDF \u89e3\u6790PDF\u5e76\u63d0\u53d6\u7ed3\u6784\u5316\u5185\u5bb9import fitz # PyMuPDFdef extract_pdf_with_structure(pdf_path): doc = fitz.open(pdf_path) sections = []
\n for page in doc: blocks = page.get_text("dict")["blocks"] for block in blocks: if block["type"] == 0: # \u6587\u672c\u5757 for line in block["lines"]: text = "".join( span["text"] for span in line["spans"] ) font_size = line["spans"][0]["size"] # \u901a\u8fc7\u5b57\u53f7\u5224\u65ad\u6807\u9898\u5c42\u7ea7 if font_size > 16: sections.append({ "type": "heading", "text": text.strip() }) else: sections.append({ "type": "paragraph", "text": text.strip() }) return sections<\/p>\n

\u4e09\u3001\u6587\u672c\u5206\u5757\u7b56\u7565<\/p>\n

\n

\u5206\u5757（Chunking）\u662fRAG\u4e2d\u6700\u88ab\u4f4e\u4f30\u5374\u5f71\u54cd\u6700\u5927\u7684\u73af\u8282\u3002\u5206\u5757\u7b56\u7565\u76f4\u63a5\u51b3\u5b9a\u4e86\u68c0\u7d22\u7684\u7cbe\u5ea6\u548c\u53ec\u56de\u7387\u3002\u6838\u5fc3\u6311\u6218\u5728\u4e8e：\u5757\u592a\u5927\u5219\u566a\u97f3\u591a\u3001\u76f8\u5173\u6027\u88ab\u7a00\u91ca；\u5757\u592a\u5c0f\u5219\u4e0a\u4e0b\u6587\u4e0d\u5b8c\u6574\u3001\u8bed\u4e49\u788e\u7247\u5316\u3002<\/p>\n

3.1 \u4e3b\u6d41\u5206\u5757\u7b56\u7565\u5bf9\u6bd4<\/h4>\n\n\u7b56\u7565\u9002\u7528\u573a\u666f\u4f18\u70b9\u7f3a\u70b9<\/tr>\n\n\n\n\n\n\n
\u56fa\u5b9a\u5927\u5c0f\u5206\u5757<\/td>\n\u901a\u7528\u6587\u672c<\/td>\n\u5b9e\u73b0\u7b80\u5355<\/td>\n\u53ef\u80fd\u622a\u65ad\u8bed\u4e49<\/td>\n<\/tr>\n
\u9012\u5f52\u5b57\u7b26\u5206\u5272<\/td>\n\u957f\u6587\u6863<\/td>\n\u4fdd\u7559\u6bb5\u843d\u8fb9\u754c<\/td>\n\u9700\u8c03\u53c2<\/td>\n<\/tr>\n
\u8bed\u4e49\u5206\u5757<\/td>\n\u591a\u4e3b\u9898\u6587\u6863<\/td>\n\u8bed\u4e49\u5b8c\u6574<\/td>\n\u8ba1\u7b97\u6210\u672c\u9ad8<\/td>\n<\/tr>\n
\u57fa\u4e8e\u7ed3\u6784\u5206\u5757<\/td>\n\u7ed3\u6784\u5316\u6587\u6863<\/td>\n\u5929\u7136\u5bf9\u9f50\u7ae0\u8282<\/td>\n\u4f9d\u8d56\u6587\u6863\u683c\u5f0f<\/td>\n<\/tr>\n
\u6ed1\u52a8\u7a97\u53e3\u5206\u5757<\/td>\n\u53d9\u4e8b\u578b\u6587\u672c<\/td>\n\u4e0a\u4e0b\u6587\u8fde\u8d2f<\/td>\n\u5197\u4f59\u8f83\u591a<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

3.2 \u63a8\u8350：\u9012\u5f52\u5206\u5757 + \u8bed\u4e49\u611f\u77e5<\/h4>\n

\u5b9e\u8df5\u4e2d，\u6211\u4eec\u63a8\u8350\u5148\u6309\u6587\u6863\u7ed3\u6784（\u6807\u9898\u3001\u6bb5\u843d）\u5206\u5272，\u518d\u5bf9\u8fc7\u957f\u6bb5\u843d\u8fdb\u884c\u9012\u5f52\u5206\u5272\u7684\u6df7\u5408\u7b56\u7565\u3002\u540c\u65f6\u52a0\u5165 overlap（\u91cd\u53e0\u7a97\u53e3）\u6765\u4fdd\u8bc1\u5206\u5757\u8fb9\u754c\u5904\u7684\u4e0a\u4e0b\u6587\u8fde\u8d2f\u6027\u3002<\/p>\n

# \u63a8\u8350\u5206\u5757\u5b9e\u73b0：\u9012\u5f52\u5206\u5757 + overlapfrom langchain.text_splitter import RecursiveCharacterTextSplittersplitter = RecursiveCharacterTextSplitter( chunk_size=512, # \u6bcf\u5757\u7ea6512 tokens chunk_overlap=64, # \u91cd\u53e064 tokens separators=[ # \u5206\u5272\u4f18\u5148\u7ea7 "\/n\/n", # \u6bb5\u843d "\/n", # \u6362\u884c "\u3002", # \u53e5\u53f7 "；", # \u5206\u53f7 "，", # \u9017\u53f7 " ", # \u7a7a\u683c ], length_function=token_counter,)chunks = splitter.split_documents(documents) <\/p>\n

💡 \u8c03\u4f18\u7ecf\u9a8c：<\/p>\n

chunk_size \u5efa\u8bae\u4ece 256~512 tokens \u8d77\u6b65，\u6839\u636e\u8bc4\u6d4b\u7ed3\u679c\u8c03\u6574<\/p>\n

overlap \u4e00\u822c\u8bbe\u4e3a chunk_size \u7684 10%~20%<\/p>\n

\u4e3a\u6bcf\u4e2a chunk \u9644\u52a0\u5143\u6570\u636e（\u6765\u6e90\u6587\u6863\u3001\u7ae0\u8282\u6807\u9898\u3001\u9875\u7801），\u63d0\u5347\u68c0\u7d22\u53ef\u89e3\u91ca\u6027<\/p>\n

\u5bf9\u4e8e\u95ee\u7b54\u573a\u666f，\u8f83\u5c0f\u7684 chunk \u901a\u5e38\u6548\u679c\u66f4\u597d；\u5bf9\u4e8e\u6458\u8981\u573a\u666f，\u8f83\u5927\u7684 chunk \u66f4\u5408\u9002<\/p>\n

\u56db\u3001\u5411\u91cf\u5316（Embedding）<\/p>\n

\n

\u5411\u91cf\u5316\u662f\u5c06\u6587\u672c\u5757\u8f6c\u5316\u4e3a\u9ad8\u7ef4\u5411\u91cf\u8868\u793a\u7684\u8fc7\u7a0b，\u662f\u8bed\u4e49\u68c0\u7d22\u7684\u6838\u5fc3\u3002\u9009\u62e9\u5408\u9002\u7684 Embedding \u6a21\u578b\u76f4\u63a5\u5f71\u54cd\u68c0\u7d22\u8d28\u91cf\u3002<\/p>\n

4.1 Embedding\u6a21\u578b\u9009\u578b<\/h4>\n\n\u6a21\u578b\u7ef4\u5ea6\u4e2d\u6587\u80fd\u529b\u63a8\u8350\u573a\u666f<\/tr>\n\n\n\n\n\n\n
text-embedding-3-large<\/td>\n3072<\/td>\n\u2b50\u2b50\u2b50\u2b50<\/td>\n\u901a\u7528\u9996\u9009<\/td>\n<\/tr>\n
bge-large-zh-v1.5<\/td>\n1024<\/td>\n\u2b50\u2b50\u2b50\u2b50\u2b50<\/td>\n\u4e2d\u6587\u4e13\u9879\u4f18\u5316<\/td>\n<\/tr>\n
GTE-Qwen2<\/td>\n1536<\/td>\n\u2b50\u2b50\u2b50\u2b50\u2b50<\/td>\n\u4e2d\u6587+\u591a\u8bed\u8a00<\/td>\n<\/tr>\n
Cohere embed-v3<\/td>\n1024<\/td>\n\u2b50\u2b50\u2b50<\/td>\n\u82f1\u6587\u4e3a\u4e3b\u573a\u666f<\/td>\n<\/tr>\n
jina-embeddings-v3<\/td>\n1024<\/td>\n\u2b50\u2b50\u2b50\u2b50<\/td>\n\u957f\u6587\u672c\u3001\u591a\u8bed\u8a00<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

4.2 Embedding \u8c03\u4f18\u7b56\u7565<\/h4>\n

1\u3001\u6307\u4ee4\u524d\u7f00（Instruction Prefix）：\u5bf9\u4e8e\u652f\u6301\u6307\u4ee4\u7684\u6a21\u578b（\u5982BGE），\u5728query\u548cdocument\u524d\u52a0\u4e0d\u540c\u524d\u7f00\u53ef\u663e\u8457\u63d0\u5347\u6548\u679c<\/p>\n

2\u3001Fine-tune Embedding：\u4f7f\u7528\u9886\u57df\u6570\u636e\u5bf9embedding\u6a21\u578b\u5fae\u8c03，\u53ef\u5c06\u68c0\u7d22\u51c6\u786e\u7387\u63d0\u534710%~30%<\/p>\n

3\u3001\u591a\u5411\u91cf\u8868\u793a：\u5bf9\u540c\u4e00chunk\u751f\u6210\u591a\u4e2a\u89c6\u89d2\u7684embedding（\u6807\u9898embedding + \u5185\u5bb9embedding + \u5047\u8bbe\u95ee\u9898embedding）<\/p>\n

4\u3001\u964d\u7ef4\u4e0e\u91cf\u5316：\u4f7f\u7528PQ（\u4e58\u79ef\u91cf\u5316）\u6216Matryoshka\u8868\u793a\u964d\u4f4e\u5b58\u50a8\u6210\u672c，\u540c\u65f6\u4fdd\u6301\u68c0\u7d22\u7cbe\u5ea6<\/p>\n

# BGE\u6a21\u578b\u4f7f\u7528\u793a\u4f8b（\u5e26\u6307\u4ee4\u524d\u7f00）from FlagEmbedding import FlagModelmodel = FlagModel( "BAAI\/bge-large-zh-v1.5", use_fp16=True)# \u6587\u6863\u5411\u91cf\u5316（\u4e0d\u52a0\u524d\u7f00）doc_embeddings = model.encode(chunks)# \u67e5\u8be2\u5411\u91cf\u5316（\u52a0\u68c0\u7d22\u6307\u4ee4\u524d\u7f00）query_prefix = "\u4e3a\u8fd9\u4e2a\u53e5\u5b50\u751f\u6210\u8868\u793a\u4ee5\u7528\u4e8e\u68c0\u7d22\u76f8\u5173\u6587\u7ae0："query_embedding = model.encode( query_prefix + user_query)<\/p>\n

\u4e94\u3001\u5411\u91cf\u6570\u636e\u5e93\u4e0e\u7d22\u5f15<\/p>\n

\n

\u5411\u91cf\u6570\u636e\u5e93\u8d1f\u8d23\u9ad8\u6548\u5b58\u50a8\u548c\u68c0\u7d22\u767e\u4e07\u4e43\u81f3\u4ebf\u7ea7\u5411\u91cf\u3002\u9009\u578b\u9700\u8981\u7efc\u5408\u8003\u8651\u89c4\u6a21\u3001\u5ef6\u8fdf\u3001\u6210\u672c\u548c\u8fd0\u7ef4\u590d\u6742\u5ea6\u3002<\/p>\n

\"\u56fe\u7247\"<\/p>\n

5.1 \u5411\u91cf\u6570\u636e\u5e93\u9009\u578b<\/h4>\n\n\u6570\u636e\u5e93\u90e8\u7f72\u65b9\u5f0f\u6838\u5fc3\u4f18\u52bf\u9002\u7528\u89c4\u6a21<\/tr>\n\n\n\n\n\n\n
Milvus<\/td>\n\u81ea\u6258\u7ba1\/\u4e91<\/td>\n\u9ad8\u6027\u80fd\u3001\u5206\u5e03\u5f0f<\/td>\n\u5927\u89c4\u6a21\u751f\u4ea7<\/td>\n<\/tr>\n
Pinecone<\/td>\n\u5168\u6258\u7ba1<\/td>\n\u96f6\u8fd0\u7ef4\u3001\u5feb\u901f\u4e0a\u624b<\/td>\n\u4e2d\u5c0f\u89c4\u6a21<\/td>\n<\/tr>\n
Weaviate<\/td>\n\u81ea\u6258\u7ba1\/\u4e91<\/td>\n\u6df7\u5408\u641c\u7d22\u539f\u751f\u652f\u6301<\/td>\n\u4e2d\u5927\u89c4\u6a21<\/td>\n<\/tr>\n
Qdrant<\/td>\n\u81ea\u6258\u7ba1\/\u4e91<\/td>\nRust\u9ad8\u6027\u80fd\u3001\u8fc7\u6ee4\u5f3a<\/td>\n\u4e2d\u5927\u89c4\u6a21<\/td>\n<\/tr>\n
pgvector<\/td>\nPostgreSQL\u63d2\u4ef6<\/td>\n\u65e0\u9700\u989d\u5916\u7ec4\u4ef6<\/td>\n\u5c0f\u89c4\u6a21\/\u539f\u578b<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

5.2 \u7d22\u5f15\u4f18\u5316\u7b56\u7565<\/h4>\n

HNSW\u53c2\u6570\u8c03\u4f18（\u4ee5Milvus\u4e3a\u4f8b）：<\/p>\n

M (\u8fde\u63a5\u6570)：16~64，\u8d8a\u5927\u7cbe\u5ea6\u8d8a\u9ad8\u4f46\u5185\u5b58\u6d88\u8017\u8d8a\u5927<\/p>\n

ef_construction：128~512，\u6784\u5efa\u65f6\u63a2\u7d22\u8303\u56f4，\u5f71\u54cd\u7d22\u5f15\u8d28\u91cf<\/p>\n

ef_search：64~256，\u68c0\u7d22\u65f6\u63a2\u7d22\u8303\u56f4，\u5f71\u54cd\u67e5\u8be2\u7cbe\u5ea6\u548c\u5ef6\u8fdf<\/p>\n

\u7ecf\u9a8c\u516c\u5f0f：ef_search \u2265 top_k，\u4e14\u5efa\u8bae ef_search = top_k \u00d7 2~4<\/p>\n

\u516d\u3001\u68c0\u7d22\u7b56\u7565\u4e0e\u4f18\u5316<\/p>\n

\n

\u68c0\u7d22\u662fRAG\u7cfb\u7edf\u7684"\u5fc3\u810f"\u3002\u5355\u7eaf\u7684\u5411\u91cf\u76f8\u4f3c\u5ea6\u641c\u7d22\u5f80\u5f80\u4e0d\u591f\u2014\u2014\u6211\u4eec\u9700\u8981\u591a\u79cd\u68c0\u7d22\u7b56\u7565\u534f\u540c\u5de5\u4f5c\u624d\u80fd\u8fbe\u5230\u6700\u4f73\u6548\u679c\u3002<\/p>\n

6.1 \u6df7\u5408\u68c0\u7d22（Hybrid Search）<\/h4>\n

\u6df7\u5408\u68c0\u7d22\u7ed3\u5408\u4e86\u7a20\u5bc6\u5411\u91cf\u68c0\u7d22（\u8bed\u4e49\u5339\u914d）\u548c\u7a00\u758f\u68c0\u7d22（\u5173\u952e\u8bcd\u5339\u914d）\u7684\u4f18\u52bf\u3002\u7a20\u5bc6\u68c0\u7d22\u64c5\u957f\u7406\u89e3\u540c\u4e49\u8bcd\u548c\u8bed\u4e49\u76f8\u8fd1\u7684\u8868\u8fbe，\u7a00\u758f\u68c0\u7d22\u64c5\u957f\u7cbe\u786e\u5339\u914d\u4e13\u6709\u540d\u8bcd\u3001\u578b\u53f7\u3001\u7f16\u53f7\u7b49\u3002<\/p>\n

# \u6df7\u5408\u68c0\u7d22\u5b9e\u73b0\u793a\u4f8bfrom rank_bm25 import BM25Okapiimport numpy as npclass HybridRetriever: def __init__(self, vector_store, bm25_index, alpha=0.7): self.vector_store = vector_store self.bm25_index = bm25_index self.alpha = alpha # \u5411\u91cf\u68c0\u7d22\u6743\u91cd
\n def search(self, query, top_k=10): # \u7a20\u5bc6\u5411\u91cf\u68c0\u7d22 dense_results = self.vector_store.similarity_search( query, k=top_k * 2 ) # \u7a00\u758fBM25\u68c0\u7d22 sparse_results = self.bm25_index.get_top_n( query.split(), top_k * 2 ) # RRF (Reciprocal Rank Fusion) \u878d\u5408 return self._rrf_fusion( dense_results, sparse_results, top_k )
\n def _rrf_fusion(self, list1, list2, top_k, k=60): scores = {} for rank, doc in enumerate(list1): scores[doc.id] = self.alpha \/ (k + rank + 1) for rank, doc in enumerate(list2): scores[doc.id] = scores.get(doc.id, 0) + \/ (1 – self.alpha) \/ (k + rank + 1) return sorted(scores.items(), key=lambda x: x[1], reverse=True )[:top_k]<\/p>\n

6.2 \u67e5\u8be2\u6539\u5199\u4e0e\u6269\u5c55<\/h4>\n

\u7528\u6237\u7684\u539f\u59cb\u67e5\u8be2\u5f80\u5f80\u4e0d\u591f\u7cbe\u786e\u6216\u8fc7\u4e8e\u53e3\u8bed\u5316\u3002\u901a\u8fc7\u67e5\u8be2\u6539\u5199\u53ef\u4ee5\u663e\u8457\u63d0\u5347\u68c0\u7d22\u6548\u679c：<\/p>\n

1\u3001HyDE（\u5047\u8bbe\u6587\u6863\u5d4c\u5165）：\u8ba9LLM\u5148\u751f\u6210\u4e00\u4e2a"\u5047\u8bbe\u7684\u7406\u60f3\u7b54\u6848"，\u7528\u8fd9\u4e2a\u7b54\u6848\u53bb\u68c0\u7d22，\u5f80\u5f80\u6bd4\u539f\u59cbquery\u6548\u679c\u66f4\u597d<\/p>\n

2\u3001Multi-Query：\u5c06\u539f\u59cbquery\u6539\u5199\u4e3a\u591a\u4e2a\u4e0d\u540c\u8868\u8ff0，\u5206\u522b\u68c0\u7d22\u540e\u5408\u5e76\u53bb\u91cd，\u63d0\u5347\u53ec\u56de\u7387<\/p>\n

3\u3001Step-back Prompting：\u5148\u8ba9LLM\u63d0\u51fa\u4e00\u4e2a\u66f4\u62bd\u8c61\u7684\u9ad8\u5c42\u95ee\u9898，\u7528\u9ad8\u5c42\u95ee\u9898\u68c0\u7d22\u83b7\u53d6\u80cc\u666f\u77e5\u8bc6<\/p>\n

4\u3001\u5173\u952e\u8bcd\u63d0\u53d6：\u4ecequery\u4e2d\u63d0\u53d6\u6838\u5fc3\u5b9e\u4f53\u548c\u5173\u952e\u8bcd，\u7528\u4e8e\u7a00\u758f\u68c0\u7d22\u548c\u5143\u6570\u636e\u8fc7\u6ee4<\/p>\n

# HyDE\u5b9e\u73b0\u793a\u4f8bdef hyde_retrieval(query, llm, retriever): # Step 1: \u8ba9LLM\u751f\u6210\u5047\u8bbe\u7b54\u6848 hyde_prompt = f"""\u8bf7\u6839\u636e\u4ee5\u4e0b\u95ee\u9898，\u5199\u4e00\u6bb5\u53ef\u80fd\u7684\u7b54\u6848（\u5373\u4f7f\u4e0d\u786e\u5b9a\u4e5f\u8bf7\u5c1d\u8bd5）：\u95ee\u9898：{query}\u5047\u8bbe\u7b54\u6848："""
\n hypothetical_answer = llm.generate(hyde_prompt)
\n # Step 2: \u7528\u5047\u8bbe\u7b54\u6848\u53bb\u68c0\u7d22 results = retriever.search(hypothetical_answer)
\n return results<\/p>\n

6.3 \u91cd\u6392\u5e8f（Reranking）<\/h4>\n

\u521d\u59cb\u68c0\u7d22\u8fd4\u56de\u7684top-k\u7ed3\u679c\u4e2d，\u6392\u5e8f\u5f80\u5f80\u4e0d\u591f\u7cbe\u51c6\u3002Reranker\u6a21\u578b\u901a\u8fc7\u4ea4\u53c9\u7f16\u7801（Cross-Encoder）\u5bf9query-document\u5bf9\u8fdb\u884c\u7cbe\u7ec6\u6253\u5206，\u663e\u8457\u63d0\u5347\u6700\u7ec8\u6392\u5e8f\u8d28\u91cf\u3002<\/p>\n

\u26a0\ufe0f Reranking\u7684ROI\u6781\u9ad8：\u5728\u6211\u4eec\u7684\u5b9e\u8df5\u4e2d，\u52a0\u5165Reranker\u540e，Top-3\u547d\u4e2d\u7387\u5e73\u5747\u63d0\u534715%~25%\u3002\u63a8\u8350\u6a21\u578b：<\/p>\n

bge-reranker-v2-m3：\u5f00\u6e90\u6700\u4f73，\u4e2d\u82f1\u6587\u4ff1\u4f73<\/p>\n

Cohere Rerank：\u5546\u4e1aAPI，\u6548\u679c\u7a33\u5b9a<\/p>\n

cross-encoder\/ms-marco：\u82f1\u6587\u573a\u666f\u7ecf\u5178\u9009\u62e9<\/p>\n

# Reranker\u4f7f\u7528\u793a\u4f8bfrom FlagEmbedding import FlagRerankerreranker = FlagReranker( "BAAI\/bge-reranker-v2-m3", use_fp16=True)# \u5bf9\u521d\u59cb\u68c0\u7d22\u7ed3\u679c\u91cd\u6392\u5e8fpairs = [[query, doc.text] for doc in initial_results]scores = reranker.compute_score(pairs)# \u6309\u5206\u6570\u91cd\u65b0\u6392\u5e8freranked = sorted( zip(initial_results, scores), key=lambda x: x[1], reverse=True)final_results = [doc for doc, _ in reranked[:5]]<\/p>\n

\u4e03\u3001\u751f\u6210\u4f18\u5316<\/p>\n

\n

\u68c0\u7d22\u5230\u9ad8\u8d28\u91cf\u7684\u4e0a\u4e0b\u6587\u540e，\u5982\u4f55\u8ba9\u5927\u6a21\u578b\u57fa\u4e8e\u8fd9\u4e9b\u4fe1\u606f\u751f\u6210\u51c6\u786e\u3001\u6709\u7528\u7684\u56de\u7b54，\u540c\u6837\u9700\u8981\u7cbe\u5fc3\u8bbe\u8ba1\u3002<\/p>\n

7.1 Prompt\u5de5\u7a0b<\/h4>\n

\u4e00\u4e2a\u597d\u7684RAG Prompt\u9700\u8981\u660e\u786e\u544a\u8bc9\u6a21\u578b：\u4f60\u7684\u4fe1\u606f\u6765\u6e90\u662f\u4ec0\u4e48\u3001\u5982\u4f55\u5904\u7406\u68c0\u7d22\u5185\u5bb9\u4e0e\u81ea\u8eab\u77e5\u8bc6\u7684\u51b2\u7a81\u3001\u4f55\u65f6\u627f\u8ba4"\u4e0d\u77e5\u9053"\u3002<\/p>\n

# \u751f\u4ea7\u7ea7RAG Prompt\u6a21\u677fRAG_PROMPT = """\u4f60\u662f\u4e00\u4e2a\u4e13\u4e1a\u7684\u77e5\u8bc6\u52a9\u624b\u3002\u8bf7\u4e25\u683c\u57fa\u4e8e\u4ee5\u4e0b\u53c2\u8003\u8d44\u6599\u56de\u7b54\u7528\u6237\u95ee\u9898\u3002
\n## \u89c4\u52191. \u53ea\u4f7f\u7528\u53c2\u8003\u8d44\u6599\u4e2d\u7684\u4fe1\u606f\u56de\u7b54，\u4e0d\u8981\u4f7f\u7528\u4f60\u7684\u5185\u90e8\u77e5\u8bc62. \u5982\u679c\u53c2\u8003\u8d44\u6599\u4e0d\u8db3\u4ee5\u56de\u7b54\u95ee\u9898，\u8bf7\u660e\u786e\u8bf4\u660e"\u6839\u636e\u73b0\u6709\u8d44\u6599\u65e0\u6cd5\u56de\u7b54"3. \u5f15\u7528\u5177\u4f53\u6765\u6e90，\u683c\u5f0f\u4e3a [\u6765\u6e90X]4. \u4fdd\u6301\u56de\u7b54\u7b80\u6d01\u3001\u51c6\u786e\u3001\u6709\u6761\u7406
\n## \u53c2\u8003\u8d44\u6599{context}
\n## \u7528\u6237\u95ee\u9898{question}
\n## \u56de\u7b54"""<\/p>\n

7.2 \u4e0a\u4e0b\u6587\u7a97\u53e3\u7ba1\u7406<\/h4>\n

1\u3001Lost in the Middle\u95ee\u9898：\u5927\u6a21\u578b\u5bf9\u4e0a\u4e0b\u6587\u4e2d\u95f4\u4f4d\u7f6e\u7684\u4fe1\u606f\u5229\u7528\u7387\u8f83\u4f4e，\u5e94\u5c06\u6700\u76f8\u5173\u7684\u5185\u5bb9\u653e\u5728\u5f00\u5934\u548c\u7ed3\u5c3e<\/p>\n

2\u3001\u4e0a\u4e0b\u6587\u538b\u7f29：\u4f7f\u7528LLM\u5bf9\u68c0\u7d22\u7ed3\u679c\u8fdb\u884c\u6458\u8981\u538b\u7f29，\u53bb\u9664\u5197\u4f59\u4fe1\u606f，\u53ea\u4fdd\u7559\u4e0equery\u76f8\u5173\u7684\u90e8\u5206<\/p>\n

3\u3001\u52a8\u6001Top-K：\u6839\u636e\u68c0\u7d22\u5206\u6570\u7684\u5206\u5e03\u52a8\u6001\u51b3\u5b9a\u9001\u5165\u591a\u5c11\u6761\u7ed3\u679c，\u907f\u514d\u5f15\u5165\u4f4e\u76f8\u5173\u6027\u566a\u97f3<\/p>\n

4\u3001\u5f15\u7528\u8ffd\u8e2a：\u4e3a\u6bcf\u6761\u68c0\u7d22\u7ed3\u679c\u6807\u6ce8\u6765\u6e90\u6807\u7b7e，\u8ba9\u6a21\u578b\u5728\u56de\u7b54\u4e2d\u6807\u6ce8\u5f15\u7528，\u63d0\u5347\u53ef\u4fe1\u5ea6<\/p>\n

7.3 \u5e7b\u89c9\u63a7\u5236<\/h4>\n

\u964d\u4f4e\u5e7b\u89c9\u7684\u5b9e\u7528\u6280\u5de7：<\/p>\n

\u5728system prompt\u4e2d\u5f3a\u8c03"\u4e0d\u786e\u5b9a\u65f6\u8bf7\u8bf4\u4e0d\u77e5\u9053"<\/p>\n

\u4f7f\u7528temperature=0\u6216\u8f83\u4f4e\u503c（\u59820.1）\u51cf\u5c11\u968f\u673a\u6027<\/p>\n

\u8981\u6c42\u6a21\u578b\u5148\u5f15\u7528\u539f\u6587\u518d\u7ed9\u51fa\u603b\u7ed3<\/p>\n

\u540e\u5904\u7406：\u5bf9\u751f\u6210\u7ed3\u679c\u505a\u4e8b\u5b9e\u6838\u67e5（NLI\u6a21\u578b\u9a8c\u8bc1）<\/p>\n

\u5f15\u5165Citation\u673a\u5236：\u8981\u6c42\u6a21\u578b\u6807\u6ce8\u6bcf\u53e5\u8bdd\u7684\u4fe1\u606f\u6765\u6e90<\/p>\n

\u516b\u3001\u8bc4\u4f30\u4f53\u7cfb<\/p>\n

\n

\u6ca1\u6709\u8bc4\u4f30\u5c31\u6ca1\u6709\u4f18\u5316\u3002RAG\u7cfb\u7edf\u9700\u8981\u5efa\u7acb\u591a\u7ef4\u5ea6\u7684\u8bc4\u4f30\u4f53\u7cfb，\u8986\u76d6\u68c0\u7d22\u548c\u751f\u6210\u4e24\u4e2a\u9636\u6bb5\u3002<\/p>\n

RAG \u8bc4\u4f30\u6307\u6807\u4f53\u7cfb\u68c0\u7d22\u8d28\u91cf\u6307\u6807Recall@K (\u53ec\u56de\u7387)MRR (\u5e73\u5747\u5012\u6570\u6392\u540d)NDCG@K (\u5f52\u4e00\u5316\u6298\u635f\u589e\u76ca)Hit Rate (\u547d\u4e2d\u7387)\u751f\u6210\u8d28\u91cf\u6307\u6807Faithfulness (\u5fe0\u5b9e\u5ea6)Answer Relevancy (\u76f8\u5173\u6027)Completeness (\u5b8c\u6574\u6027)Hallucination Rate (\u5e7b\u89c9\u7387)<\/p>\n

8.1 \u63a8\u8350\u8bc4\u4f30\u6846\u67b6<\/h4>\n

# \u4f7f\u7528 RAGAS \u6846\u67b6\u8fdb\u884c\u81ea\u52a8\u5316\u8bc4\u4f30from ragas import evaluatefrom ragas.metrics import ( faithfulness, answer_relevancy, context_precision, context_recall,)from datasets import Dataset
\n# \u6784\u9020\u8bc4\u4f30\u6570\u636e\u96c6eval_data = { "question": questions, "answer": generated_answers, "contexts": retrieved_contexts, "ground_truth": reference_answers,}dataset = Dataset.from_dict(eval_data)
\n# \u8fd0\u884c\u8bc4\u4f30results = evaluate( dataset, metrics=[ faithfulness, # \u751f\u6210\u662f\u5426\u5fe0\u4e8e\u68c0\u7d22\u5185\u5bb9 answer_relevancy, # \u56de\u7b54\u4e0e\u95ee\u9898\u7684\u76f8\u5173\u5ea6 context_precision, # \u68c0\u7d22\u7cbe\u5ea6 context_recall, # \u68c0\u7d22\u53ec\u56de ],)print(results) <\/p>\n

\u4e5d\u3001\u8fdb\u9636\u4f18\u5316\u7b56\u7565<\/p>\n

\n

\u5f53\u57fa\u7840RAG\u7cfb\u7edf\u642d\u5efa\u5b8c\u6210\u540e，\u4ee5\u4e0b\u8fdb\u9636\u7b56\u7565\u53ef\u4ee5\u8fdb\u4e00\u6b65\u63d0\u5347\u7cfb\u7edf\u8868\u73b0：<\/p>\n

9.1 \u591a\u8def\u53ec\u56de + \u81ea\u9002\u5e94\u878d\u5408<\/h4>\n

\u9664\u4e86\u5411\u91cf+BM25\u7684\u53cc\u8def\u53ec\u56de，\u8fd8\u53ef\u4ee5\u5f15\u5165\u77e5\u8bc6\u56fe\u8c31\u68c0\u7d22\u3001\u5173\u952e\u8bcd\u7d22\u5f15\u3001\u5143\u6570\u636e\u8fc7\u6ee4\u7b49\u591a\u8def\u53ec\u56de\u901a\u9053\u3002\u901a\u8fc7\u5b66\u4e60\u578b\u878d\u5408\u6a21\u578b（\u800c\u975e\u56fa\u5b9a\u6743\u91cd）\u52a8\u6001\u8c03\u6574\u5404\u8def\u6743\u91cd\u3002<\/p>\n

9.2 Agentic RAG<\/h4>\n

\u8ba9\u5927\u6a21\u578b\u4f5c\u4e3a"\u667a\u80fd\u4ee3\u7406"\u4e3b\u52a8\u51b3\u5b9a\u68c0\u7d22\u7b56\u7565：\u5224\u65ad\u662f\u5426\u9700\u8981\u68c0\u7d22\u3001\u9700\u8981\u68c0\u7d22\u4ec0\u4e48\u3001\u68c0\u7d22\u7ed3\u679c\u662f\u5426\u8db3\u591f\u3001\u662f\u5426\u9700\u8981\u8ffd\u52a0\u68c0\u7d22\u3002\u8fd9\u6bd4\u4f20\u7edf\u7684"\u4e00\u6b21\u68c0\u7d22"\u6a21\u5f0f\u66f4\u52a0\u7075\u6d3b\u548c\u7cbe\u51c6\u3002<\/p>\n

# Agentic RAG \u6838\u5fc3\u5faa\u73afdef agentic_rag(query, max_rounds=3): context = []
\n for round in range(max_rounds): # \u8ba9LLM\u51b3\u5b9a\u4e0b\u4e00\u6b65\u52a8\u4f5c action = llm.decide_action(query, context)
\n if action.type == "search": new_results = retriever.search(action.query) context.extend(new_results) elif action.type == "answer": return llm.generate_answer(query, context) elif action.type == "refine_query": query = action.refined_query
\n return llm.generate_answer(query, context)<\/p>\n

9.3 GraphRAG<\/h4>\n

\u5c06\u6587\u6863\u4e2d\u7684\u5b9e\u4f53\u5173\u7cfb\u62bd\u53d6\u4e3a\u77e5\u8bc6\u56fe\u8c31，\u5728\u68c0\u7d22\u65f6\u4e0d\u4ec5\u627e\u5230\u76f4\u63a5\u5339\u914d\u7684\u6587\u672c\u5757，\u8fd8\u80fd\u901a\u8fc7\u56fe\u5173\u7cfb\u627e\u5230\u95f4\u63a5\u76f8\u5173\u7684\u77e5\u8bc6\u3002\u5c24\u5176\u9002\u5408\u9700\u8981\u591a\u8df3\u63a8\u7406\u7684\u590d\u6742\u95ee\u7b54\u573a\u666f\u3002<\/p>\n

9.4 \u7f13\u5b58\u4e0e\u6027\u80fd\u4f18\u5316<\/h4>\n

\u751f\u4ea7\u73af\u5883\u6027\u80fd\u4f18\u5316\u6e05\u5355：<\/p>\n

\u8bed\u4e49\u7f13\u5b58：\u5bf9\u76f8\u4f3cquery\u7684\u7ed3\u679c\u8fdb\u884c\u7f13\u5b58，\u51cf\u5c11\u91cd\u590d\u8ba1\u7b97<\/p>\n

\u5f02\u6b65Embedding：\u6587\u6863\u5165\u5e93\u65f6\u4f7f\u7528batch\u5904\u7406，\u63d0\u5347\u541e\u5410<\/p>\n

\u5206\u7ea7\u68c0\u7d22：\u5148\u7c97\u7b5b（ANN）\u518d\u7cbe\u6392（Reranker），\u63a7\u5236\u5ef6\u8fdf<\/p>\n

\u9884\u8ba1\u7b97：\u70ed\u95e8\u95ee\u9898\u7684\u68c0\u7d22\u7ed3\u679c\u9884\u8ba1\u7b97\u5e76\u7f13\u5b58<\/p>\n

Streaming\u8f93\u51fa：\u751f\u6210\u9636\u6bb5\u4f7f\u7528\u6d41\u5f0f\u8f93\u51fa，\u4f18\u5316\u7528\u6237\u4f53\u9a8c<\/p>\n

\u5341\u3001\u603b\u7ed3\u4e0e\u6700\u4f73\u5b9e\u8df5<\/p>\n

\n

\"\u56fe\u7247\"<\/p>\n

\u6700\u540e，\u5206\u4eab\u4e00\u4e2a\u5728\u4f17\u591a\u9879\u76ee\u4e2d\u9a8c\u8bc1\u8fc7\u7684RAG\u7cfb\u7edf\u8fed\u4ee3\u8def\u5f84：<\/p>\n

1\u3001MVP\u9636\u6bb5：\u56fa\u5b9a\u5206\u5757 + \u5355\u8def\u5411\u91cf\u68c0\u7d22 + \u57fa\u7840Prompt \u2192 \u5feb\u901f\u9a8c\u8bc1\u53ef\u884c\u6027<\/p>\n

2\u3001\u4f18\u5316\u9636\u6bb5：\u8c03\u4f18\u5206\u5757\u7b56\u7565 + \u6df7\u5408\u68c0\u7d22 + Reranker \u2192 \u63d0\u5347\u68c0\u7d22\u51c6\u786e\u7387<\/p>\n

3\u3001\u8fdb\u9636\u9636\u6bb5：\u67e5\u8be2\u6539\u5199 + Agentic RAG + GraphRAG \u2192 \u5904\u7406\u590d\u6742\u573a\u666f<\/p>\n

4\u3001\u751f\u4ea7\u9636\u6bb5：\u7f13\u5b58\u4f18\u5316 + \u76d1\u63a7\u544a\u8b66 + A\/B\u6d4b\u8bd5 + \u6301\u7eed\u8fed\u4ee3 \u2192 \u7a33\u5b9a\u8fd0\u884c<\/p>\n

RAG\u4e0d\u662f\u4e00\u4e2a\u4e00\u6b21\u6027\u7684\u5de5\u7a0b，\u800c\u662f\u4e00\u4e2a\u9700\u8981\u6301\u7eed\u6253\u78e8\u7684\u7cfb\u7edf\u3002\u5e0c\u671b\u8fd9\u7bc7\u6307\u5357\u80fd\u5e2e\u52a9\u4f60\u5728RAG\u5f00\u53d1\u4e4b\u8def\u4e0a\u5c11\u8d70\u5f2f\u8def，\u6784\u5efa\u51fa\u771f\u6b63\u597d\u7528\u7684\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u3002<\/p>\n

\u5927\u6a21\u578b\u7684\u77e5\u8bc6\u53c8\u591a\u53c8\u6742，\u5230\u5e95\u600e\u4e48\u5b66？<\/h4>\n

\u505a\u8fd9\u884c\u4e45\u4e86，\u88ab\u95ee\u6700\u591a\u7684\u95ee\u9898\u5c31\u4e00\u4e2a：\u5927\u6a21\u578b\u7684\u77e5\u8bc6\u53c8\u591a\u53c8\u6742，\u6211\u5230\u5e95\u5148\u5b66\u5565？\u7ffb\u4e86\u4e00\u5806\u8d44\u6599，\u770b\u5565\u90fd\u50cf\u91cd\u70b9，\u7ed3\u679c\u5565\u90fd\u6ca1\u5b66\u8fdb\u53bb\u3002\u4eca\u5929\u628a\u6211\u81ea\u5df1\u4e00\u8def\u6478\u51fa\u6765\u7684\u987a\u5e8f\u634b\u4e00\u904d，\u96f6\u57fa\u7840\u3001\u5e94\u5c4a\u751f\u3001\u8f6c\u884c\u7684\u90fd\u9002\u7528\u3002\u8ddf\u7740\u8d70，\u81f3\u5c11\u65b9\u5411\u4e0d\u4e71\u3001\u4e0d\u778e\u5fd9，\u5b66\u5b8c\u62ff\u5f97\u51fa\u624b\u3002<\/p>\n

\u7b2c\u4e00\u9636\u6bb5（10\u5929）：\u521d\u9636\u5e94\u7528<\/h6>\n

\u8fd9\u4e00\u9636\u6bb5\u5c31\u4e00\u4e2a\u76ee\u6807：\u628a\u5927\u6a21\u578b\u771f\u6b63"\u73a9\u660e\u767d"，\u4e0d\u662f\u53ea\u4f1a\u804a\u5929\u90a3\u79cd，\u800c\u662f\u80fd\u8c03\u6559\u5b83\u3001\u80fd\u7528\u4ee3\u7801\u628a\u5927\u6a21\u578b\u548c\u4e1a\u52a1\u63a5\u8d77\u6765\u3002<\/p>\n

    \n
  • \u5927\u6a21\u578b AI \u80fd\u5e72\u4ec0\u4e48？<\/li>\n
  • \u5927\u6a21\u578b\u662f\u600e\u6837\u83b7\u5f97\u300c\u667a\u80fd\u300d\u7684？<\/li>\n
  • \u7528\u597d AI \u7684\u6838\u5fc3\u5fc3\u6cd5<\/li>\n
  • \u5927\u6a21\u578b\u5e94\u7528\u4e1a\u52a1\u67b6\u6784<\/li>\n
  • \u5927\u6a21\u578b\u5e94\u7528\u6280\u672f\u67b6\u6784<\/li>\n
  • \u4ee3\u7801\u793a\u4f8b：\u5411 GPT-3.5 \u704c\u5165\u65b0\u77e5\u8bc6<\/li>\n
  • \u63d0\u793a\u5de5\u7a0b\u7684\u610f\u4e49\u548c\u6838\u5fc3\u601d\u60f3<\/li>\n
  • Prompt \u5178\u578b\u6784\u6210<\/li>\n
  • \u6307\u4ee4\u8c03\u4f18\u65b9\u6cd5\u8bba<\/li>\n
  • \u601d\u7ef4\u94fe\u548c\u601d\u7ef4\u6811<\/li>\n
  • Prompt \u653b\u51fb\u548c\u9632\u8303<\/li>\n
  • \u2026<\/li>\n<\/ul>\n
    \u7b2c\u4e8c\u9636\u6bb5（30\u5929）：\u9ad8\u9636\u5e94\u7528<\/h6>\n

    \u8fd9\u9636\u6bb5\u6b63\u5f0f\u8fdb\u5b9e\u6218，\u76ee\u6807\u5f88\u5b9e\u5728：\u505a\u51fa\u4e24\u4e2a\u80fd\u62ff\u51fa\u624b\u7684\u4e1c\u897f\u2014\u2014\u4e00\u4e2a\u79c1\u6709\u77e5\u8bc6\u5e93，\u4e00\u4e2a agent \u5bf9\u8bdd\u673a\u5668\u4eba\u3002<\/p>\n

      \n
    • \n

      \u4e3a\u4ec0\u4e48\u8981\u505a RAG<\/p>\n<\/li>\n

    • \n

      \u642d\u5efa\u4e00\u4e2a\u7b80\u5355\u7684 ChatPDF<\/p>\n<\/li>\n

    • \n

      \u68c0\u7d22\u7684\u57fa\u7840\u6982\u5ff5<\/p>\n<\/li>\n

    • \n

      \u4ec0\u4e48\u662f\u5411\u91cf\u8868\u793a（Embeddings）<\/p>\n<\/li>\n

    • \n

      \u5411\u91cf\u6570\u636e\u5e93\u4e0e\u5411\u91cf\u68c0\u7d22<\/p>\n<\/li>\n

    • \n

      \u57fa\u4e8e\u5411\u91cf\u68c0\u7d22\u7684 RAG<\/p>\n<\/li>\n

    • \n

      \u642d\u5efa RAG \u7cfb\u7edf\u7684\u6269\u5c55\u77e5\u8bc6<\/p>\n<\/li>\n

    • \n

      \u6df7\u5408\u68c0\u7d22\u4e0e RAG-Fusion \u7b80\u4ecb<\/p>\n<\/li>\n

    • \n

      \u5411\u91cf\u6a21\u578b\u672c\u5730\u90e8\u7f72<\/p>\n<\/li>\n

    • \n

      \u2026<\/p>\n<\/li>\n<\/ul>\n

      \u8d70\u5b8c\u8fd9 30 \u5929，Python \u6216 JavaScript \u80cc\u666f\u7684\u57fa\u672c\u80fd\u72ec\u7acb\u505a\u4e00\u5957 RAG \u5e94\u7528，\u7b80\u5386\u4e0a\u4e5f\u6562\u5199"\u5927\u6a21\u578b\u5e94\u7528\u5f00\u53d1"\u4e86\u3002<\/p>\n

      \u7b2c\u4e09\u9636\u6bb5（\u7ea620\u5929）：\u52a8\u624b\u5fae\u8c03<\/h6>\n

      \u5b66\u5230\u8fd9，\u4f60\u5df2\u7ecf\u5f00\u59cb \u201c\u9020\u201d \u6a21\u578b\u4e86 \u2014\u2014 \u901a\u8fc7\u5fae\u8c03，\u641e\u51fa\u5782\u76f4\u9886\u57df\u7684\u4e13\u7528\u5c0f\u6a21\u578b\u3002<\/p>\n

        \n
      • \u4e3a\u4ec0\u4e48\u8981\u505a RAG<\/li>\n
      • \u4ec0\u4e48\u662f\u6a21\u578b<\/li>\n
      • \u4ec0\u4e48\u662f\u6a21\u578b\u8bad\u7ec3<\/li>\n
      • \u6c42\u89e3\u5668 & \u635f\u5931\u51fd\u6570\u7b80\u4ecb<\/li>\n
      • \u5c0f\u5b9e\u9a8c2：\u624b\u5199\u4e00\u4e2a\u7b80\u5355\u7684\u795e\u7ecf\u7f51\u7edc\u5e76\u8bad\u7ec3\u5b83<\/li>\n
      • \u4ec0\u4e48\u662f\u8bad\u7ec3\/\u9884\u8bad\u7ec3\/\u5fae\u8c03\/\u8f7b\u91cf\u5316\u5fae\u8c03<\/li>\n
      • Transformer\u7ed3\u6784\u7b80\u4ecb<\/li>\n
      • \u8f7b\u91cf\u5316\u5fae\u8c03<\/li>\n
      • \u5b9e\u9a8c\u6570\u636e\u96c6\u7684\u6784\u5efa<\/li>\n
      • \u2026<\/li>\n<\/ul>\n

        \u606d\u559c\u4f60，\u5982\u679c\u5b66\u5230\u8fd9\u91cc，\u4f60\u57fa\u672c\u53ef\u4ee5\u627e\u5230\u4e00\u4efd\u5927\u6a21\u578b AI\u76f8\u5173\u7684\u5de5\u4f5c，\u81ea\u5df1\u4e5f\u80fd\u8bad\u7ec3 GPT \u4e86！\u901a\u8fc7\u5fae\u8c03，\u8bad\u7ec3\u81ea\u5df1\u7684\u5782\u76f4\u5927\u6a21\u578b，\u80fd\u72ec\u7acb\u8bad\u7ec3\u5f00\u6e90\u591a\u6a21\u6001\u5927\u6a21\u578b，\u638c\u63e1\u66f4\u591a\u6280\u672f\u65b9\u6848\u3002<\/p>\n

        \u5230\u6b64\u4e3a\u6b62，\u5927\u69822\u4e2a\u6708\u7684\u65f6\u95f4\u3002\u4f60\u4ece \u201c\u53ea\u4f1a\u7528\u201d \u53d8\u6210 \u201c\u80fd\u9020\u201d \u7684 AI \u5c0f\u5b50\u3002\u90a3\u4e48\u4f60\u8fd8\u60f3\u5f80\u4e0b\u63a2\u7d22\u5417？<\/p>\n

        \u60f3\u8981\u5165\u5c40AI\u5927\u6a21\u578b\u8d5b\u9053\u3001\u62a2\u5360\u884c\u4e1a\u7ea2\u5229\u7684\u670b\u53cb，\u5fae\u4fe1\u626b\u63cf\u4e0b\u65b9CSDN\u5b98\u65b9\u8ba4\u8bc1\u4e8c\u7ef4\u7801，\u5373\u53ef100%\u514d\u8d39\u9886\u53d6\u5168\u5957\u5b66\u4e60\u8d44\u6599！\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

        \u7b2c\u56db\u9636\u6bb5（20\u5929）：\u5546\u4e1a\u95ed\u73af，\u90e8\u7f72\u843d\u5730<\/h6>\n

        \u6280\u672f\u5b66\u5b8c，\u843d\u5730\u624d\u7b97\u6570\u3002\u8fd9\u9636\u6bb5\u5e26\u4f60\u641e\u660e\u767d\u5927\u6a21\u578b\u600e\u4e48\u771f\u6b63\u8dd1\u8d77\u6765\u3001\u7528\u8d77\u6765\u3002<\/p>\n

          \n
        • \u786c\u4ef6\u9009\u578b<\/li>\n
        • \u5e26\u4f60\u4e86\u89e3\u5168\u7403\u5927\u6a21\u578b<\/li>\n
        • \u4f7f\u7528\u56fd\u4ea7\u5927\u6a21\u578b\u670d\u52a1<\/li>\n
        • \u642d\u5efa OpenAI \u4ee3\u7406<\/li>\n
        • \u70ed\u8eab：\u57fa\u4e8e\u963f\u91cc\u4e91 PAI \u90e8\u7f72 Stable Diffusion<\/li>\n
        • \u5728\u672c\u5730\u8ba1\u7b97\u673a\u8fd0\u884c\u5927\u6a21\u578b<\/li>\n
        • \u5927\u6a21\u578b\u7684\u79c1\u6709\u5316\u90e8\u7f72<\/li>\n
        • \u57fa\u4e8e vLLM \u90e8\u7f72\u5927\u6a21\u578b<\/li>\n
        • \u6848\u4f8b：\u5982\u4f55\u4f18\u96c5\u5730\u5728\u963f\u91cc\u4e91\u79c1\u6709\u90e8\u7f72\u5f00\u6e90\u5927\u6a21\u578b<\/li>\n
        • \u90e8\u7f72\u4e00\u5957\u5f00\u6e90 LLM \u9879\u76ee<\/li>\n
        • \u5185\u5bb9\u5b89\u5168<\/li>\n
        • \u4e92\u8054\u7f51\u4fe1\u606f\u670d\u52a1\u7b97\u6cd5\u5907\u6848<\/li>\n
        • \u2026<\/li>\n<\/ul>\n

          \u60f3\u8981\u5165\u5751\u5927\u6a21\u578b\u7684\u670b\u53cb，\u7ed9\u5927\u5bb6\u51c6\u5907\u4e86\u4e00\u4efd2026 \u5e74\u6700\u65b0 AI \u5927\u6a21\u578b\u5168\u5957\u5b66\u4e60\u8d44\u6599：<\/p>\n

          1\u3001\u5927\u6a21\u578b\u7cfb\u7edf\u5316\u5b66\u4e60\u8def\u7ebf<\/h6>\n

          \"\u8bf7\u6dfb\u52a0\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

          2\u3001\u5927\u6a21\u578b\u5b66\u4e60\u89c6\u9891\u6559\u7a0b<\/h6>\n

          \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

          3\u3001\u5927\u6a21\u578b\u5b66\u4e60\u4e66\u7c4d&\u7535\u5b50\u6587\u6863<\/h6>\n

          \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

          4\u3001AI\u5927\u6a21\u578b\u6700\u65b0\u884c\u4e1a\u62a5\u544a<\/h6>\n

          \"img\"<\/p>\n

          5\u3001\u5927\u6a21\u578b\u9879\u76ee\u5b9e\u6218&\u914d\u5957\u6e90\u7801<\/h6>\n

          \"img\"<\/p>\n

          6\u30012026\u5927\u6a21\u578b\u5927\u5382\u9762\u8bd5\u771f\u9898<\/h6>\n

          \"img\"<\/p>\n

          7\u3001\u4e3a\u4ec0\u4e48\u8981\u5b66\u4e60\u5927\u6a21\u578b？<\/h6>\n

          \u6211\u56fd\u5728A\u5927\u6a21\u578b\u9886\u57df\u9762\u4e34\u4eba\u624d\u77ed\u7f3a,\u6570\u91cf\u4e0e\u8d28\u91cf\u5747\u843d\u540e\u4e8e\u53d1\u8fbe\u56fd\u5bb6\u30022023\u5e74，\u4eba\u624d\u7f3a\u53e3\u5df2\u8d85\u767e\u4e07，\u51f8\u663e\u57f9\u517b\u4e0d\u8db3\u3002\u968f\u7740AI\u6280\u672f\u98de\u901f\u53d1\u5c55，\u9884\u8ba1\u52302025\u5e74,\u8fd9\u4e00\u7f3a\u53e3\u5c06\u6025\u5267\u6269\u5927\u81f3400\u4e07,\u4e25\u91cd\u5236\u7ea6\u6211\u56fdAI\u4ea7\u4e1a\u7684\u521b\u65b0\u6b65\u4f10\u3002\u52a0\u5f3a\u4eba\u624d\u57f9\u517b,\u4f18\u5316\u6559\u80b2\u4f53\u7cfb,\u56fd\u9645\u5408\u4f5c\u5e76\u8fdb\u662f\u7834\u89e3\u56f0\u5c40\u3001\u63a8\u52a8AI\u53d1\u5c55\u7684\u5173\u952e\u3002<\/p>\n

          \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

          \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

          8\u3001\u9002\u5408\u8c01\u5b66？<\/h6>\n
            \n
          • \u4f20\u7edf\u540e\u7aef \/ Java \/ \u524d\u7aef\u5f00\u53d1，\u60f3\u8f6c\u578b AI \u5e94\u7528<\/li>\n
          • \u5927\u5b66\u751f\u3001\u5e94\u5c4a\u751f，\u60f3\u62ff\u66f4\u597d\u7684 offer<\/li>\n
          • \u4ea7\u54c1\u7ecf\u7406\u3001\u8fd0\u8425，\u60f3\u6b66\u88c5\u804c\u4e1a\u7ade\u4e89\u529b<\/li>\n
          • \u6280\u672f\u8d1f\u8d23\u4eba，\u60f3\u7ed9\u56e2\u961f\u843d\u5730\u63d0\u6548<\/li>\n<\/ul>\n

            \u5b66\u4e60\u662f\u53cd\u4eba\u6027\u7684，\u4f46\u56de\u62a5\u662f\u771f\u91d1\u767d\u94f6\u3002\u6280\u672f\u4f1a\u66f4\u65b0，\u8d5b\u9053\u4f1a\u5207\u6362，\u4f46\u53ea\u8981\u4f60\u5148\u52a8\u624b，\u673a\u4f1a\u5c31\u6c38\u8fdc\u7ad9\u5728\u4f60\u8fd9\u8fb9\u3002<\/p>\n

            9\u3001\u8fd9\u4e9b\u8d44\u6599\u771f\u7684\u6709\u7528\u5417？<\/h6>\n

            \u8fd9\u4efd\u8d44\u6599\u7531\u6211\u548c\u9c81\u4e3a\u6c11\u535a\u58eb(\u5317\u4eac\u6e05\u534e\u5927\u5b66\u5b66\u58eb\u548c\u7f8e\u56fd\u52a0\u5dde\u7406\u5de5\u5b66\u9662\u535a\u58eb)\u5171\u540c\u6574\u7406，\u73b0\u4efb\u4e0a\u6d77\u6bb7\u6cca\u4fe1\u606f\u79d1\u6280CEO，\u5176\u521b\u7acb\u7684MoPaaS\u4e91\u5e73\u53f0\u83b7Forrester\u5168\u7403\u2019\u5f3a\u52b2\u8868\u73b0\u8005\u2019\u8ba4\u8bc1，\u670d\u52a1\u822a\u5929\u79d1\u5de5\u3001\u56fd\u5bb6\u7535\u7f51\u7b491000+\u4f01\u4e1a，\u4ee5\u7b2c\u4e00\u4f5c\u8005\u5728IEEE Transactions\u53d1\u8868\u8bba\u658750+\u7bc7，\u83b7NASA JPL\u706b\u661f\u63a2\u6d4b\u7cfb\u7edf\u5f3a\u5316\u5b66\u4e60\u4e13\u5229\u7b4935\u9879\u4e2d\u7f8e\u4e13\u5229\u3002\u672c\u5957AI\u5927\u6a21\u578b\u8bfe\u7a0b\u7531\u6e05\u534e\u5927\u5b66-\u52a0\u5dde\u7406\u5de5\u53cc\u6599\u535a\u58eb\u3001\u5434\u6587\u4fca\u4eba\u5de5\u667a\u80fd\u5956\u5f97\u4e3b\u9c81\u4e3a\u6c11\u6559\u6388\u9886\u8854\u7814\u53d1\u3002<\/p>\n

            \u8d44\u6599\u5185\u5bb9\u6db5\u76d6\u4e86\u4ece\u5165\u95e8\u5230\u8fdb\u9636\u7684\u5404\u7c7b\u89c6\u9891\u6559\u7a0b\u548c\u5b9e\u6218\u9879\u76ee，\u65e0\u8bba\u4f60\u662f\u5c0f\u767d\u8fd8\u662f\u6709\u4e9b\u6280\u672f\u57fa\u7840\u7684\u6280\u672f\u4eba\u5458，\u8fd9\u4efd\u8d44\u6599\u90fd\u7edd\u5bf9\u80fd\u5e2e\u52a9\u4f60\u63d0\u5347\u85aa\u8d44\u5f85\u9047，\u8f6c\u884c\u5927\u6a21\u578b\u5c97\u4f4d\u3002
            \n\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
            \n\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

            \u60f3\u8981\u5165\u5c40AI\u5927\u6a21\u578b\u8d5b\u9053\u3001\u62a2\u5360\u884c\u4e1a\u7ea2\u5229\u7684\u670b\u53cb，\u5fae\u4fe1\u626b\u63cf\u4e0b\u65b9CSDN\u5b98\u65b9\u8ba4\u8bc1\u4e8c\u7ef4\u7801，\u5373\u53ef100%\u514d\u8d39\u9886\u53d6\u5168\u5957\u5b66\u4e60\u8d44\u6599！\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

            \u597d\u4e86，\u5c31\u5199\u5230\u8fd9\u3002\u8def\u7ebf\u662f\u6b7b\u7684，\u4eba\u662f\u6d3b\u7684，\u5361\u4f4f\u4e86\u968f\u65f6\u6765\u95ee\u6211\u3002\u89c9\u5f97\u6709\u7528\u5c31\u70b9\u4e2a\u6536\u85cf，\u522b\u8ba9\u5b83\u8fdb\u6536\u85cf\u5939\u5403\u7070\u2014\u2014\u52a8\u8d77\u6765\u6bd4\u5565\u90fd\u5f3a\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

            \u672c\u6587\u4ece\u5de5\u7a0b\u5b9e\u8df5\u89d2\u5ea6\u7cfb\u7edf\u68b3\u7406\u4e86RAG（\u68c0\u7d22\u589e\u5f3a\u751f\u6210）\u7cfb\u7edf\u7684\u5168\u6d41\u7a0b\u5f00\u53d1\u6b65\u9aa4\u4e0e\u8c03\u4f18\u7b56\u7565，\u5e2e\u52a9\u8bfb\u8005\u6784\u5efa\u5b9e\u7528\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u3002\u5185\u5bb9\u6db5\u76d6\u6587\u6863\u5904\u7406\u3001\u6587\u672c\u5206\u5757\u3001\u5411\u91cf\u5316\u3001\u5411\u91cf\u6570\u636e\u5e93\u3001\u68c0\u7d22\u7b56\u7565\u3001\u751f\u6210\u4f18\u5316\u3001\u8bc4\u4f30\u4f53\u7cfb\u7b49\u6838\u5fc3\u73af\u8282，\u6bcf\u4e2a\u73af\u8282\u914d\u6709\u4ee3\u7801\u793a\u4f8b\u548c\u8c03\u4f18\u8981\u70b9，\u9002\u5408\u6709\u4e00\u5b9a\u5f00\u53d1\u57fa\u7840\u7684AI\u5de5\u7a0b\u5e08\u548c\u67b6\u6784\u5e08\u5b66\u4e60\u3002
            \n\u5f53\u6211\u4eec\u8ba9\u5927\u6a21\u578b\u56de\u7b54\\”\u516c\u53f8\u4e0a\u5b63\u5ea6\u7684\u8425\u6536\u662f\u591a\u5c11\\”\u6216\\”\u4ea7\u54c1X\u7684\u4f7f\u7528\u9650\u5236\u662f\u4ec0\u4e48\\”\u65f6，\u5927\u6a21\u578b\u5f80\u5f80\u65e0\u6cd5\u7ed9\u51fa\u51c6\u786e\u7b54\u6848\u2014\u2014\u56e0\u4e3a\u8fd9\u4e9b\u79c1\u57df<\/p>\n","protected":false},"author":2,"featured_media":121335,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[50,87,132,614,371],"topic":[],"class_list":["post-121349","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-50","tag-87","tag-132","tag-614","tag-371"],"yoast_head":"\n\u5c0f\u767d\u7a0b\u5e8f\u5458\u5fc5\u5907\uff1a\u6536\u85cf\u8fd9\u4efdRAG\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u5f00\u53d1\u5168\u653b\u7565\uff08\u9644\u4ee3\u7801\u793a\u4f8b\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/121349.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u5c0f\u767d\u7a0b\u5e8f\u5458\u5fc5\u5907\uff1a\u6536\u85cf\u8fd9\u4efdRAG\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u5f00\u53d1\u5168\u653b\u7565\uff08\u9644\u4ee3\u7801\u793a\u4f8b\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u672c\u6587\u4ece\u5de5\u7a0b\u5b9e\u8df5\u89d2\u5ea6\u7cfb\u7edf\u68b3\u7406\u4e86RAG（\u68c0\u7d22\u589e\u5f3a\u751f\u6210）\u7cfb\u7edf\u7684\u5168\u6d41\u7a0b\u5f00\u53d1\u6b65\u9aa4\u4e0e\u8c03\u4f18\u7b56\u7565，\u5e2e\u52a9\u8bfb\u8005\u6784\u5efa\u5b9e\u7528\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u3002\u5185\u5bb9\u6db5\u76d6\u6587\u6863\u5904\u7406\u3001\u6587\u672c\u5206\u5757\u3001\u5411\u91cf\u5316\u3001\u5411\u91cf\u6570\u636e\u5e93\u3001\u68c0\u7d22\u7b56\u7565\u3001\u751f\u6210\u4f18\u5316\u3001\u8bc4\u4f30\u4f53\u7cfb\u7b49\u6838\u5fc3\u73af\u8282，\u6bcf\u4e2a\u73af\u8282\u914d\u6709\u4ee3\u7801\u793a\u4f8b\u548c\u8c03\u4f18\u8981\u70b9，\u9002\u5408\u6709\u4e00\u5b9a\u5f00\u53d1\u57fa\u7840\u7684AI\u5de5\u7a0b\u5e08\u548c\u67b6\u6784\u5e08\u5b66\u4e60\u3002 \u5f53\u6211\u4eec\u8ba9\u5927\u6a21\u578b\u56de\u7b54"\u516c\u53f8\u4e0a\u5b63\u5ea6\u7684\u8425\u6536\u662f\u591a\u5c11"\u6216"\u4ea7\u54c1X\u7684\u4f7f\u7528\u9650\u5236\u662f\u4ec0\u4e48"\u65f6，\u5927\u6a21\u578b\u5f80\u5f80\u65e0\u6cd5\u7ed9\u51fa\u51c6\u786e\u7b54\u6848\u2014\u2014\u56e0\u4e3a\u8fd9\u4e9b\u79c1\u57df\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/121349.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-11-02T07:18:01+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/11\/20261102071757-6ae8392582463.jpg\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/121349.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/121349.html\",\"name\":\"\u5c0f\u767d\u7a0b\u5e8f\u5458\u5fc5\u5907\uff1a\u6536\u85cf\u8fd9\u4efdRAG\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u5f00\u53d1\u5168\u653b\u7565\uff08\u9644\u4ee3\u7801\u793a\u4f8b\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-11-02T07:18:01+00:00\",\"dateModified\":\"2026-11-02T07:18:01+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/121349.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/121349.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/121349.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u5c0f\u767d\u7a0b\u5e8f\u5458\u5fc5\u5907\uff1a\u6536\u85cf\u8fd9\u4efdRAG\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u5f00\u53d1\u5168\u653b\u7565\uff08\u9644\u4ee3\u7801\u793a\u4f8b\uff09\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u5c0f\u767d\u7a0b\u5e8f\u5458\u5fc5\u5907\uff1a\u6536\u85cf\u8fd9\u4efdRAG\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u5f00\u53d1\u5168\u653b\u7565\uff08\u9644\u4ee3\u7801\u793a\u4f8b\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/121349.html","og_locale":"zh_CN","og_type":"article","og_title":"\u5c0f\u767d\u7a0b\u5e8f\u5458\u5fc5\u5907\uff1a\u6536\u85cf\u8fd9\u4efdRAG\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u5f00\u53d1\u5168\u653b\u7565\uff08\u9644\u4ee3\u7801\u793a\u4f8b\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u672c\u6587\u4ece\u5de5\u7a0b\u5b9e\u8df5\u89d2\u5ea6\u7cfb\u7edf\u68b3\u7406\u4e86RAG（\u68c0\u7d22\u589e\u5f3a\u751f\u6210）\u7cfb\u7edf\u7684\u5168\u6d41\u7a0b\u5f00\u53d1\u6b65\u9aa4\u4e0e\u8c03\u4f18\u7b56\u7565，\u5e2e\u52a9\u8bfb\u8005\u6784\u5efa\u5b9e\u7528\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u3002\u5185\u5bb9\u6db5\u76d6\u6587\u6863\u5904\u7406\u3001\u6587\u672c\u5206\u5757\u3001\u5411\u91cf\u5316\u3001\u5411\u91cf\u6570\u636e\u5e93\u3001\u68c0\u7d22\u7b56\u7565\u3001\u751f\u6210\u4f18\u5316\u3001\u8bc4\u4f30\u4f53\u7cfb\u7b49\u6838\u5fc3\u73af\u8282，\u6bcf\u4e2a\u73af\u8282\u914d\u6709\u4ee3\u7801\u793a\u4f8b\u548c\u8c03\u4f18\u8981\u70b9，\u9002\u5408\u6709\u4e00\u5b9a\u5f00\u53d1\u57fa\u7840\u7684AI\u5de5\u7a0b\u5e08\u548c\u67b6\u6784\u5e08\u5b66\u4e60\u3002 \u5f53\u6211\u4eec\u8ba9\u5927\u6a21\u578b\u56de\u7b54\"\u516c\u53f8\u4e0a\u5b63\u5ea6\u7684\u8425\u6536\u662f\u591a\u5c11\"\u6216\"\u4ea7\u54c1X\u7684\u4f7f\u7528\u9650\u5236\u662f\u4ec0\u4e48\"\u65f6，\u5927\u6a21\u578b\u5f80\u5f80\u65e0\u6cd5\u7ed9\u51fa\u51c6\u786e\u7b54\u6848\u2014\u2014\u56e0\u4e3a\u8fd9\u4e9b\u79c1\u57df","og_url":"https:\/\/www.wsisp.com\/helps\/121349.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-11-02T07:18:01+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/11\/20261102071757-6ae8392582463.jpg"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"6 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/121349.html","url":"https:\/\/www.wsisp.com\/helps\/121349.html","name":"\u5c0f\u767d\u7a0b\u5e8f\u5458\u5fc5\u5907\uff1a\u6536\u85cf\u8fd9\u4efdRAG\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u5f00\u53d1\u5168\u653b\u7565\uff08\u9644\u4ee3\u7801\u793a\u4f8b\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-11-02T07:18:01+00:00","dateModified":"2026-11-02T07:18:01+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/121349.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/121349.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/121349.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u5c0f\u767d\u7a0b\u5e8f\u5458\u5fc5\u5907\uff1a\u6536\u85cf\u8fd9\u4efdRAG\u77e5\u8bc6\u95ee\u7b54\u7cfb\u7edf\u5f00\u53d1\u5168\u653b\u7565\uff08\u9644\u4ee3\u7801\u793a\u4f8b\uff09"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/121349","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=121349"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/121349\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/121335"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=121349"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=121349"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=121349"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=121349"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}