{"id":9450,"date":"2025-10-30T14:40:50","date_gmt":"2025-10-30T06:40:50","guid":{"rendered":"\/?p=9450"},"modified":"2025-10-30T14:40:50","modified_gmt":"2025-10-30T06:40:50","slug":"rag%e7%9a%84%e7%9f%a5%e8%af%86%e6%98%af%e6%80%8e%e4%b9%88%e6%9e%84%e5%bb%ba%e7%9a%84","status":"publish","type":"post","link":"\/?p=9450","title":{"rendered":"RAG\u7684\u77e5\u8bc6\u662f\u600e\u4e48\u6784\u5efa\u7684"},"content":{"rendered":"<h2>**\u4e00\u3001\u4e3a\u4ec0\u4e48\u77e5\u8bc6\u5e93\u662fRAG\u7684\u7075\u9b42\uff1f<\/h2>\n<p>RAG \u7684\u6838\u5fc3\u601d\u60f3\u662f\u300c\u8ba9\u6a21\u578b\u5177\u5907\u5916\u90e8\u8bb0\u5fc6\u300d\u3002<\/p>\n<p>\u4f60\u95ee\u5b83\u95ee\u9898\uff0c\u5b83\u4e0d\u662f\u5168\u9760\u8111\u5b50\uff08\u53c2\u6570\uff09\u56de\u5fc6\uff0c\u800c\u662f\u5148\u67e5\u8d44\u6599\uff0c\u518d\u751f\u6210\u7b54\u6848\u3002<\/p>\n<p>\u800c\u8fd9\u4e9b\u201c\u8d44\u6599\u201d\u5c31\u6765\u81ea\u77e5\u8bc6\u5e93\u3002<\/p>\n<p>\u6362\u53e5\u8bdd\u8bf4\uff1a\u77e5\u8bc6\u5e93\u7684\u8d28\u91cf\uff0c\u51b3\u5b9a\u4e86\u6574\u4e2a\u7cfb\u7edf\u7684\u201c\u667a\u5546\u4e0a\u9650\u201d\u3002<\/p>\n<p>\u6240\u4ee5\u5728\u9762\u8bd5\u91cc\uff0c\u5982\u679c\u4f60\u80fd\u628a\u77e5\u8bc6\u5e93\u6784\u5efa\u8bb2\u5f97\u6e05\u695a\uff0c\u9762\u8bd5\u5b98\u57fa\u672c\u5c31\u4f1a\u5224\u65ad\uff1a\u4f60\u662f\u771f\u7684\u7406\u89e3 RAG \u7684\u5e95\u5c42\u903b\u8f91\u3002<\/p>\n<h2>\u4e8c\u3001\u77e5\u8bc6\u5e93\u6784\u5efa\u7684\u5b8c\u6574\u6d41\u7a0b<\/h2>\n<p>\u6784\u5efa\u77e5\u8bc6\u5e93\uff0c\u4e0d\u662f\u201c\u4e0a\u4f20\u4e00\u5806\u6587\u6863\u201d\u90a3\u4e48\u7b80\u5355\u3002 \u5b83\u5176\u5b9e\u662f\u4e00\u5957\u6807\u51c6\u7684\u79bb\u7ebf\u89e3\u6790\u6d41\u6c34\u7ebf\uff0c\u5927\u81f4\u5305\u62ec\u4e94\u6b65\uff1a<\/p>\n<ol>\n<li>\u6587\u6863\u89e3\u6790\uff08Document Parsing\uff09<\/li>\n<li>\u5185\u5bb9\u6e05\u6d17\uff08Data Cleaning\uff09<\/li>\n<li>\u6587\u672c\u5206\u5757\uff08Chunking\uff09<\/li>\n<li>\u5411\u91cf\u5316\uff08Embedding\uff09<\/li>\n<li>\u7d22\u5f15\u6784\u5efa\uff08Indexing\uff09<\/li>\n<\/ol>\n<p>\u8fd9\u4e94\u6b65\uff0c\u662f\u4e00\u4e2a\u4ece\u201c\u975e\u7ed3\u6784\u5316\u201d\u5230\u201c\u53ef\u68c0\u7d22\u201d\u7684\u8f6c\u5316\u8fc7\u7a0b\u3002<\/p>\n<h3><strong>1. \u6587\u6863\u89e3\u6790\uff1a\u8ba9\u673a\u5668\u770b\u61c2\u5404\u79cd\u683c\u5f0f<\/strong><\/h3>\n<p>\u77e5\u8bc6\u6e90\u5f80\u5f80\u662f\u6df7\u4e71\u7684\uff1aPDF\u3001Word\u3001PPT\u3001\u7f51\u9875\u3001\u751a\u81f3\u626b\u63cf\u4ef6\u3002 \u7b2c\u4e00\u6b65\u5c31\u662f<strong>\u7edf\u4e00\u683c\u5f0f<\/strong>\u3002<\/p>\n<ul>\n<li>\u5bf9\u7ed3\u6784\u5316\u6587\u6863\uff08\u5982 Word\u3001Markdown\uff09\uff0c\u76f4\u63a5\u62bd\u53d6\u6587\u5b57\uff1b<\/li>\n<li>\u5bf9\u626b\u63cf\u4ef6\u6216\u56fe\u7247\uff0c\u4f7f\u7528 OCR\uff08\u5982 PaddleOCR\uff09\uff1b<\/li>\n<li>\u4fdd\u7559\u6bb5\u843d\u3001\u6807\u9898\u3001\u8868\u683c\u7ed3\u6784\uff0c\u907f\u514d\u201c\u788e\u6210\u4e00\u9505\u7ca5\u201d\u3002<\/li>\n<\/ul>\n<p>\u5f88\u591a\u4eba\u5230\u8fd9\u4e00\u6b65\u5c31\u4e22\u4e86\u4fe1\u606f\u5c42\u7ea7\uff0c\u6bd4\u5982\u8868\u683c\u91cc\u7684\u201c\u53c2\u6570\u5bf9\u6bd4\u201d\u88ab\u76f4\u63a5\u62c9\u5e73\u6210\u7eaf\u6587\u672c\uff0c\u540e\u7eed\u68c0\u7d22\u5b8c\u5168\u627e\u4e0d\u5230\u5173\u952e\u5b57\u6bb5\u3002 \u6240\u4ee5\u89e3\u6790\u9636\u6bb5\uff0c\u4fdd\u7559\u7ed3\u6784\u4fe1\u606f\u975e\u5e38\u5173\u952e\u3002<\/p>\n<h3><strong>2. \u5185\u5bb9\u6e05\u6d17\uff1a\u628a\u5783\u573e\u4fe1\u606f\u5f7b\u5e95\u6e05\u6389<\/strong><\/h3>\n<p>\u8fd9\u4e00\u6b65\u5f80\u5f80\u6700\u67af\u71e5\uff0c\u4f46\u6700\u5f71\u54cd\u8d28\u91cf\u3002<\/p>\n<p>\u5e38\u89c1\u505a\u6cd5\u5305\u62ec\uff1a<\/p>\n<ul>\n<li>\u53bb\u6389\u63a7\u5236\u7b26\u3001\u9875\u7709\u9875\u811a\u3001\u6c34\u5370\u5e7f\u544a\uff1b<\/li>\n<li>\u7edf\u4e00\u7f16\u7801\u683c\u5f0f\uff1b<\/li>\n<li>\u53bb\u91cd\u3001\u53bb\u566a\uff1b<\/li>\n<li>\u4fdd\u7559\u81ea\u7136\u6bb5\u843d\u8fb9\u754c\u3002<\/li>\n<\/ul>\n<p>\u4e00\u4e2a\u597d\u7684\u7ecf\u9a8c\u6cd5\u5219\u662f\uff1a \u201c\u5982\u679c\u4f60\u89c9\u5f97\u8089\u773c\u770b\u90fd\u4e71\uff0c\u90a3 Embedding \u6a21\u578b\u4e00\u5b9a\u5b66\u4e0d\u51fa\u597d\u7279\u5f81\u3002\u201d<\/p>\n<h3><strong>3. \u6587\u672c\u5206\u5757\uff1a\u51b3\u5b9a\u68c0\u7d22\u7cbe\u5ea6\u7684\u5173\u952e<\/strong><\/h3>\n<p>\u8fd9\u4e00\u6b65\u5f88\u591a\u4eba\u8bb2\u4e0d\u6e05\u695a\u3002 \u5927\u591a\u6570\u521d\u5b66\u8005\u53ea\u4f1a\u8bf4\u201c\u628a\u6587\u6863\u5207\u6210500\u5b57\u4e00\u5757\u201d\uff0c\u4f46\u90a3\u5176\u5b9e\u662f\u6700\u4f4e\u914d\u7248\u672c\u3002<\/p>\n<p>\u66f4\u5408\u7406\u7684\u5206\u5757\u65b9\u5f0f\u662f\u2014\u2014<strong>\u8bed\u4e49\u5206\u5757\uff08Semantic Chunking\uff09<\/strong>\uff1a<\/p>\n<ul>\n<li>\u5148\u6309\u81ea\u7136\u6bb5\u6216\u6807\u9898\u5207\uff1b<\/li>\n<li>\u518d\u7528\u53e5\u5b50\u8fb9\u754c\u9012\u5f52\u7ec6\u5206\uff1b<\/li>\n<li>\u8bbe\u7f6e\u91cd\u53e0\u7a97\u53e3\uff08Overlap\uff09\uff0c\u6bd4\u5982\u6bcf\u5757\u91cd\u53e050~100\u5b57\uff0c\u9632\u6b62\u4e0a\u4e0b\u6587\u65ad\u88c2\u3002<\/li>\n<\/ul>\n<p>\u6bd4\u5982\uff1a<\/p>\n<blockquote>\n<p>\u201cRAG\u4f18\u5316\u7684\u5173\u952e\u4e4b\u4e00\u662f\u68c0\u7d22\u6a21\u5757\u2026\u2026\uff08Overlap\uff09\u2026\u2026\u68c0\u7d22\u6a21\u5757\u901a\u5e38\u91c7\u7528\u5411\u91cf\u53ec\u56de\u4e0eBM25\u6df7\u5408\u7b56\u7565\u3002\u201d<\/p>\n<\/blockquote>\n<p>\u8fd9\u6837\u65e2\u4fdd\u8bc1\u8bed\u4e49\u8fde\u8d2f\uff0c\u53c8\u80fd\u88ab\u9ad8\u6548\u68c0\u7d22\u3002<\/p>\n<p>\u5757\u592a\u5927\uff0c\u6a21\u578b\u53ec\u56de\u4e0d\u51c6\uff1b<\/p>\n<p>\u5757\u592a\u5c0f\uff0c\u8bed\u4e49\u788e\u88c2\u3002 \u8c03\u4f18\u65f6\u8981\u7ed3\u5408\u4e0b\u6e38LLM\u4e0a\u4e0b\u6587\u7a97\u53e3\uff0c\u627e\u5230\u6700\u5408\u9002\u7684\u5757\u957f\uff08\u901a\u5e38\u5728200~800\u5b57\u4e4b\u95f4\uff09\u3002<\/p>\n<h3><strong>4. \u5411\u91cf\u5316\uff1a\u628a\u6587\u672c\u8f6c\u6210\u53ef\u8ba1\u7b97\u7684\u77e5\u8bc6<\/strong><\/h3>\n<p>\u5206\u5757\u540e\uff0c\u8981\u4e3a\u6bcf\u4e2a\u5757\u751f\u6210\u8bed\u4e49\u5411\u91cf\uff08Embedding\uff09\u3002<\/p>\n<ul>\n<li>\u6a21\u578b\u9009\u578b\uff1a\u901a\u7528\uff08\u5982 bge-large\u3001E5-base\uff09\u6216\u9886\u57df\u5fae\u8c03\u6a21\u578b\uff1b<\/li>\n<li>\u5411\u91cf\u5f52\u4e00\u5316\uff1a\u4fdd\u6301\u8ddd\u79bb\u8ba1\u7b97\u7a33\u5b9a\uff1b<\/li>\n<li>\u964d\u7ef4\u6216\u91cf\u5316\uff1a\u964d\u4f4e\u5b58\u50a8\u548c\u8ba1\u7b97\u5f00\u9500\u3002<\/li>\n<\/ul>\n<p>Embedding \u8d28\u91cf\u76f4\u63a5\u5f71\u54cd\u68c0\u7d22\u76f8\u5173\u5ea6\u3002 \u5982\u679cEmbedding\u4e0d\u7a33\u5b9a\uff0c\u518d\u597d\u7684\u53ec\u56de\u6a21\u578b\u4e5f\u627e\u4e0d\u5230\u5bf9\u7684\u5757\u3002<\/p>\n<h3><strong>5. \u7d22\u5f15\u6784\u5efa\uff1a\u8ba9\u77e5\u8bc6\u5e93\u80fd\u201c\u88ab\u627e\u5230\u201d<\/strong><\/h3>\n<p>\u6700\u540e\u4e00\u6b65\u662f\u628a\u6240\u6709Embedding\u548c\u5bf9\u5e94\u6587\u672c\u5757\u5efa\u7acb\u7d22\u5f15\u3002 \u5e38\u7528\u65b9\u6848\u5305\u62ec\uff1a<\/p>\n<ul>\n<li>HNSW\u3001IVF\u7b49\u8fd1\u4f3c\u6700\u8fd1\u90bb\uff08ANN\uff09\u7ed3\u6784\uff1b<\/li>\n<li>\u5411\u91cf\u6570\u636e\u5e93\uff08\u5982 FAISS\u3001Milvus\u3001Elasticsearch\uff09\u3002<\/li>\n<\/ul>\n<p>\u540c\u65f6\u53ef\u4ee5\u4e3a\u6bcf\u4e2a\u5757\u9644\u5e26\u5143\u6570\u636e\uff0c\u6bd4\u5982\uff1a<\/p>\n<ul>\n<li>\u6587\u6863ID<\/li>\n<li>\u6807\u9898<\/li>\n<li>\u65f6\u95f4\u6233<\/li>\n<li>\u6765\u6e90\u7c7b\u522b<\/li>\n<\/ul>\n<p>\u8fd9\u6837\u540e\u7eed\u5c31\u80fd\u57fa\u4e8e\u5143\u6570\u636e\u8fc7\u6ee4\uff0c\u6bd4\u5982\uff1a<\/p>\n<blockquote>\n<p>\u201c\u53ea\u68c0\u7d22\u6700\u65b030\u5929\u7684\u516c\u544a\u6587\u6863\u3002\u201d<\/p>\n<\/blockquote>\n<p>\u8fd9\u4e00\u6b65\u4e5f\u8981\u8003\u8651\u66f4\u65b0\u673a\u5236\uff0c\u6bd4\u5982\u6bcf\u5468\u81ea\u52a8\u91cd\u5efa\u7d22\u5f15\u6216\u505a\u589e\u91cf\u66f4\u65b0\uff0c\u786e\u4fdd\u77e5\u8bc6\u5e93\u6301\u7eed\u65b0\u9c9c\u3002<\/p>\n<h2>\u4e09\u3001\u4f18\u5316\u4e0e\u5de5\u7a0b\u7ec6\u8282\uff1a\u62c9\u5f00\u5019\u9009\u4eba\u5dee\u8ddd\u7684\u5730\u65b9<\/h2>\n<p>\u80fd\u628a\u4e0a\u9762\u4e94\u6b65\u8bb2\u5b8c\u6574\u7684\u4eba\uff0c\u5df2\u7ecf\u80fd\u8fc7\u9762\u8bd5\u57fa\u7840\u5173\u3002 \u4f46\u60f3\u62ff\u9ad8\u5206\uff0c\u5c31\u8981\u8865\u4e00\u53e5\uff1a\u201c\u6211\u4eec\u5728\u9879\u76ee\u91cc\u662f\u600e\u4e48\u4f18\u5316\u7684\u201d\u3002<\/p>\n<h3><strong>1. \u9488\u5bf9\u4e0d\u540c\u6587\u6863\u7c7b\u578b\u5b9a\u5236\u89e3\u6790<\/strong><\/h3>\n<p>\u4f8b\u5982\uff1a<\/p>\n<ul>\n<li>PDF\u5f15\u5165\u7248\u9762\u5206\u6790\uff1b<\/li>\n<li>\u7f51\u9875\u8fc7\u6ee4\u5e7f\u544a\u533a\uff1b<\/li>\n<li>\u626b\u63cf\u4ef6\u542f\u7528\u8868\u683c\u8bc6\u522b\u6a21\u578b\u3002<\/li>\n<\/ul>\n<h3><strong>2. \u8bed\u4e49\u5207\u5206+\u91cd\u53e0\u7a97\u53e3\u7b56\u7565<\/strong><\/h3>\n<p>\u8ba9\u5757\u65e2\u4e0d\u4e22\u4fe1\u606f\uff0c\u4e5f\u4e0d\u4e71\u5957\u903b\u8f91\uff1b \u53ef\u4ee5\u8c03\u8282\u91cd\u53e0\u957f\u5ea6\u3001\u5e73\u5747\u5757\u957f\uff0c\u627e\u5230\u53ec\u56de\u51c6\u786e\u7387\u548c\u751f\u6210\u7a97\u53e3\u4e4b\u95f4\u7684\u5e73\u8861\u70b9\u3002<\/p>\n<h3><strong>3. \u540c\u4e49\u8bcd\u5f52\u4e00\u5316\u4e0e\u6570\u636e\u589e\u5e7f<\/strong><\/h3>\n<p>\u5728\u79bb\u7ebf\u9636\u6bb5\u7edf\u4e00\u66ff\u6362\u201cLLM \u2194 \u5927\u8bed\u8a00\u6a21\u578b\u201d\u201cRAG \u2194 \u68c0\u7d22\u589e\u5f3a\u751f\u6210\u201d\uff0c\u51cf\u5c11\u68c0\u7d22\u504f\u5dee\u3002<\/p>\n<h3><strong>4. \u6307\u6807\u4e0e\u76d1\u63a7<\/strong><\/h3>\n<p>\u8bb0\u5f55\uff1a<\/p>\n<ul>\n<li>\u6587\u6863\u89e3\u6790\u6210\u529f\u7387\uff1b<\/li>\n<li>\u5e73\u5747\u5757\u957f\u5ea6\uff1b<\/li>\n<li>\u5411\u91cf\u751f\u6210\u8017\u65f6\uff1b<\/li>\n<li>\u68c0\u7d22\u53ec\u56de\u7387\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u80fd\u5e2e\u4f60\u6301\u7eed\u8bc4\u4f30\u77e5\u8bc6\u5e93\u7684\u5065\u5eb7\u5ea6\u3002<\/p>\n<h2>\u56db\u3001\u7b54\u9898\u6846\u67b6\u603b\u7ed3\uff08\u9762\u8bd51\u5206\u949f\u7248\u672c\uff09<\/h2>\n<p>\u5982\u679c\u4f60\u5728\u9762\u8bd5\u73b0\u573a\uff0c\u65f6\u95f4\u6709\u9650\uff0c\u53ef\u4ee5\u8fd9\u6837\u7b54\uff1a<\/p>\n<blockquote>\n<p>\u201cRAG \u7684\u77e5\u8bc6\u5e93\u6784\u5efa\u662f\u4e00\u4e2a\u6807\u51c6\u7684\u79bb\u7ebf\u89e3\u6790\u6d41\u7a0b\uff0c\u5206\u4e3a\u4e94\u6b65\uff1a<\/p>\n<ol>\n<li>\u6587\u6863\u89e3\u6790\uff1a\u7edf\u4e00\u683c\u5f0f\u3001OCR\u56fe\u7247\uff1b<\/li>\n<li>\u5185\u5bb9\u6e05\u6d17\uff1a\u53bb\u566a\u3001\u53bb\u91cd\u3001\u89c4\u8303\u5316\uff1b<\/li>\n<li>\u6587\u672c\u5206\u5757\uff1a\u6309\u8bed\u4e49\u5207\u5206\u5e76\u8bbe\u7f6eOverlap\uff1b<\/li>\n<li>\u5411\u91cf\u5316\uff1a\u7528Embedding\u6a21\u578b\u751f\u6210\u5411\u91cf\uff1b<\/li>\n<li>\u7d22\u5f15\u6784\u5efa\uff1a\u5b58\u5165\u5411\u91cf\u5e93\u5e76\u9644\u52a0\u5143\u6570\u636e\u3002<\/li>\n<\/ol>\n<p>\u5b9e\u8df5\u4e2d\u6211\u4eec\u8fd8\u4f1a\u9488\u5bf9\u4e0d\u540c\u683c\u5f0f\u5b9a\u5236\u89e3\u6790\u65b9\u6848\uff0c\u5e76\u76d1\u63a7\u89e3\u6790\u6210\u529f\u7387\u548c\u7d22\u5f15\u66f4\u65b0\u8d28\u91cf\uff0c\u4fdd\u8bc1\u77e5\u8bc6\u5e93\u53ef\u6301\u7eed\u66f4\u65b0\u3002\u201d<\/p>\n<\/blockquote>\n<p>\u8fd9\u79cd\u7b54\u6cd5\u65e2\u7cfb\u7edf\u3001\u53c8\u6709\u5b9e\u64cd\u5473\u9053\u3002<\/p>\n<h3><strong>\u4e94\u3001\u7ed3\u8bed\uff1a\u4ece\u201c\u4e0a\u4f20\u6587\u6863\u201d\u5230\u201c\u6784\u5efa\u77e5\u8bc6\u201d<\/strong><\/h3>\n<p>RAG \u77e5\u8bc6\u5e93\u7684\u6784\u5efa\uff0c\u770b\u4f3c\u662f\u5de5\u7a0b\u7ec6\u8282\uff0c\u5176\u5b9e\u8003\u7684\u662f\u8ba4\u77e5\u6df1\u5ea6\u3002<\/p>\n<p>\u5f88\u591a\u4eba\u4ee5\u4e3a\u8fd9\u90e8\u5206\u201c\u504f\u5e95\u5c42\u201d\uff0c\u9762\u8bd5\u91cc\u4e0d\u91cd\u8981\uff1b<\/p>\n<p>\u4f46\u771f\u6b63\u505a\u8fc7\u9879\u76ee\u7684\u4eba\u90fd\u77e5\u9053\u2014\u2014\u79bb\u7ebf\u89e3\u6790\uff0c\u662fRAG\u7cfb\u7edf\u80fd\u4e0d\u80fd\u8dd1\u8d77\u6765\u7684\u5e95\u5ea7\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>**\u4e00\u3001\u4e3a\u4ec0\u4e48\u77e5\u8bc6\u5e93\u662fRAG\u7684\u7075\u9b42\uff1f RAG \u7684\u6838\u5fc3\u601d\u60f3\u662f\u300c\u8ba9\u6a21\u578b\u5177\u5907\u5916\u90e8\u8bb0\u5fc6\u300d\u3002 \u4f60\u95ee\u5b83\u95ee\u9898\uff0c\u5b83\u4e0d\u662f\u5168\u9760\u8111\u5b50\uff08\u53c2\u6570\uff09\u56de\u5fc6\uff0c\u800c\u662f\u5148\u67e5\u8d44\u6599   \u2026 &#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ngg_post_thumbnail":0},"categories":[287],"tags":[],"_links":{"self":[{"href":"\/index.php?rest_route=\/wp\/v2\/posts\/9450"}],"collection":[{"href":"\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=9450"}],"version-history":[{"count":1,"href":"\/index.php?rest_route=\/wp\/v2\/posts\/9450\/revisions"}],"predecessor-version":[{"id":9451,"href":"\/index.php?rest_route=\/wp\/v2\/posts\/9450\/revisions\/9451"}],"wp:attachment":[{"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=9450"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=9450"},{"taxonomy":"post_tag","embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=9450"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}