{"id":15594,"date":"2024-12-14T15:10:06","date_gmt":"2024-12-14T07:10:06","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=15594"},"modified":"2024-12-21T20:06:25","modified_gmt":"2024-12-21T12:06:25","slug":"2024nianduragqingdan","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/de\/2024nianduragqingdan\/","title":{"rendered":"2024\u5e74\u5ea6RAG\u6e05\u5355\uff0cRAG\u5e94\u7528\u7b56\u7565100+"},"content":{"rendered":"

<\/p>\n

 <\/p>\n

\u56de\u987e2024\uff0c\u5927\u6a21\u578b\u65e5\u65b0\u6708\u5f02\uff0c\u667a\u80fd\u4f53\u767e\u5bb6\u4e89\u9e23\u3002\u4f5c\u4e3aAI\u5e94\u7528\u7684\u91cd\u8981\u7ec4\u6210\u90e8\u5206\uff0cRAG\u4e5f\u662f\u201c\u7fa4\u96c4\u9010\u9e7f\uff0c\u8bf8\u4faf\u5e76\u8d77\u201d\u3002\u5e74\u521dModularRAG\u6301\u7eed\u5347\u6e29\u3001GraphRAG\u5927\u653e\u5f02\u5f69\uff0c\u5e74\u4e2d\u5f00\u6e90\u5de5\u5177\u5982\u706b\u5982\u837c\u3001\u77e5\u8bc6\u56fe\u8c31\u518d\u521b\u65b0\u673a\uff0c\u5e74\u672b\u56fe\u8868\u7406\u89e3\u3001\u591a\u6a21\u6001RAG\u53c8\u542f\u65b0\u5f81\u7a0b\uff0c\u7b80\u76f4\u201c\u4f60\u65b9\u5531\u7f62\u6211\u767b\u573a\u201d\uff0c\u5947\u6280\u53e0\u51fa\uff0c\u4e0d\u80dc\u679a\u4e3e\uff01<\/p>\n

\u6211\u5728\u8fd9\u91cc\u9074\u9009\u4e862024\u5e74\u5ea6\u5178\u578b\u7684RAG\u7cfb\u7edf\u548c\u8bba\u6587\uff08\u542bAI\u6ce8\u89e3\u3001\u6765\u6e90\u3001\u6458\u8981\u4fe1\u606f\uff09\uff0c\u5e76\u4e8e\u6587\u672b\u9644\u4e0aRAG\u7efc\u8ff0\u548c\u6d4b\u8bd5\u57fa\u51c6\u6750\u6599\uff0c\u671b\u6267\u6b64\u4e00\u6587\uff0c\u8ba1\u4e00\u4e07\u516d\u5343\u8a00\uff0c\u52a9\u5927\u5bb6\u901f\u901aRAG\u3002<\/p>\n

\u5168\u6587\u517172\u7bc7\uff0c\u9010\u6708\u4e3a\u7eb2\uff0c\u5f3a\u8c13\u4e4b\u201cRAG\u4e03\u5341\u4e8c\u5f0f\u201d\uff0c\u4ee5\u732e\u8bf8\u4f4d\u3002<\/strong><\/p>\n

\u5907\u6ce8\uff1a\u00a0<\/strong><\/p>\n

\u6587\u4e2d\u6240\u6709\u5185\u5bb9\u5df2\u6258\u7ba1\u5230\u5f00\u6e90\u4ed3\u5e93Awesome-RAG\uff0c\u6b22\u8fce\u63d0\u4ea4PR\u67e5\u7f3a\u8865\u6f0f\u3002<\/p>\n

GitHub\u5730\u5740\uff1ahttps:\/\/github.com\/awesome-rag\/awesome-rag<\/p><\/blockquote>\n

 <\/p>\n

(01) GraphReader\u3010\u56fe\u89e3\u4e13\u5bb6\u3011<\/h2>\n

\u56fe\u89e3\u4e13\u5bb6<\/strong>\uff1a\u50cf\u4e2a\u5584\u4e8e\u5236\u4f5c\u601d\u7ef4\u5bfc\u56fe\u7684\u5bfc\u5e08\uff0c\u5c06\u5197\u957f\u7684\u6587\u672c\u8f6c\u5316\u4e3a\u6e05\u6670\u7684\u77e5\u8bc6\u7f51\u7edc\uff0c\u8ba9AI\u80fd\u591f\u50cf\u6cbf\u7740\u5730\u56fe\u63a2\u7d22\u4e00\u6837\uff0c\u8f7b\u677e\u627e\u5230\u7b54\u6848\u9700\u8981\u7684\u5404\u4e2a\u5173\u952e\u70b9\uff0c\u6709\u6548\u514b\u670d\u4e86\u5904\u7406\u957f\u6587\u672c\u65f6\u7684”\u8ff7\u8def”\u95ee\u9898\u3002<\/p><\/blockquote>\n

\u2022 \u65f6\u95f4\uff1a01.20<\/p>\n

\u2022 \u8bba\u6587\uff1aGraphReader: Building Graph-based Agent to Enhance Long-Context Abilities of Large Language Models<\/p>\n

GraphReader\u662f\u4e00\u79cd\u57fa\u4e8e\u56fe\u7684\u667a\u80fd\u4f53\u7cfb\u7edf\uff0c\u65e8\u5728\u901a\u8fc7\u5c06\u957f\u6587\u672c\u6784\u5efa\u6210\u56fe\u5e76\u4f7f\u7528\u667a\u80fd\u4f53\u81ea\u4e3b\u63a2\u7d22\u8be5\u56fe\u6765\u5904\u7406\u957f\u6587\u672c\u3002\u5728\u63a5\u6536\u5230\u95ee\u9898\u540e\uff0c\u667a\u80fd\u4f53\u9996\u5148\u8fdb\u884c\u9010\u6b65\u5206\u6790\u5e76\u5236\u5b9a\u5408\u7406\u7684\u8ba1\u5212\u3002\u7136\u540e\uff0c\u5b83\u8c03\u7528\u4e00\u7ec4\u9884\u5b9a\u4e49\u7684\u51fd\u6570\u6765\u8bfb\u53d6\u8282\u70b9\u5185\u5bb9\u548c\u90bb\u5c45\uff0c\u4fc3\u8fdb\u5bf9\u56fe\u8fdb\u884c\u4ece\u7c97\u5230\u7ec6\u7684\u63a2\u7d22\u3002\u5728\u6574\u4e2a\u63a2\u7d22\u8fc7\u7a0b\u4e2d\uff0c\u667a\u80fd\u4f53\u4e0d\u65ad\u8bb0\u5f55\u65b0\u7684\u89c1\u89e3\u5e76\u53cd\u601d\u5f53\u524d\u60c5\u51b5\u4ee5\u4f18\u5316\u8fc7\u7a0b\uff0c\u76f4\u5230\u5b83\u6536\u96c6\u5230\u8db3\u591f\u7684\u4fe1\u606f\u6765\u751f\u6210\u7b54\u6848\u3002<\/p>\n

<\/p>\n

(02) MM-RAG\u3010\u591a\u9762\u624b\u3011<\/h2>\n

\u591a\u9762\u624b<\/strong>\uff1a\u5c31\u50cf\u4e00\u4e2a\u80fd\u540c\u65f6\u7cbe\u901a\u89c6\u89c9\u3001\u542c\u89c9\u548c\u8bed\u8a00\u7684\u5168\u80fd\u9009\u624b\uff0c\u4e0d\u4ec5\u80fd\u7406\u89e3\u4e0d\u540c\u5f62\u5f0f\u7684\u4fe1\u606f\uff0c\u8fd8\u80fd\u5728\u5b83\u4eec\u4e4b\u95f4\u81ea\u5982\u5207\u6362\u548c\u5173\u8054\u3002\u901a\u8fc7\u5bf9\u5404\u79cd\u4fe1\u606f\u7684\u7efc\u5408\u7406\u89e3\uff0c\u5b83\u80fd\u5728\u63a8\u8350\u3001\u52a9\u624b\u3001\u5a92\u4f53\u7b49\u591a\u4e2a\u9886\u57df\u63d0\u4f9b\u66f4\u667a\u80fd\u3001\u66f4\u81ea\u7136\u7684\u670d\u52a1\u3002<\/p><\/blockquote>\n

\u4ecb\u7ecd\u4e86\u591a\u6a21\u6001\u673a\u5668\u5b66\u4e60\u7684\u53d1\u5c55\uff0c\u5305\u62ec\u5bf9\u6bd4\u5b66\u4e60\u3001\u591a\u6a21\u6001\u5d4c\u5165\u5b9e\u73b0\u7684\u4efb\u610f\u6a21\u6001\u641c\u7d22\u3001\u591a\u6a21\u6001\u68c0\u7d22\u589e\u5f3a\u751f\u6210\uff08MM-RAG\uff09\u4ee5\u53ca\u5982\u4f55\u4f7f\u7528\u5411\u91cf\u6570\u636e\u5e93\u6784\u5efa\u591a\u6a21\u6001\u751f\u4ea7\u7cfb\u7edf\u7b49\u3002\u540c\u65f6\u8fd8\u63a2\u8ba8\u4e86\u591a\u6a21\u6001\u4eba\u5de5\u667a\u80fd\u7684\u672a\u6765\u53d1\u5c55\u8d8b\u52bf\uff0c\u5f3a\u8c03\u4e86\u5176\u5728\u63a8\u8350\u7cfb\u7edf\u3001\u865a\u62df\u52a9\u624b\u3001\u5a92\u4f53\u548c\u7535\u5b50\u5546\u52a1\u7b49\u9886\u57df\u7684\u5e94\u7528\u524d\u666f\u3002<\/p>\n

<\/p>\n

(03) CRAG\u3010\u81ea\u6211\u6821\u6b63\u3011<\/h2>\n

\u81ea\u6211\u6821\u6b63<\/strong>\uff1a\u50cf\u4e2a\u7ecf\u9a8c\u4e30\u5bcc\u7684\u7f16\u8f91\uff0c\u5148\u7528\u7b80\u5355\u5feb\u901f\u7684\u65b9\u5f0f\u7b5b\u9009\u521d\u6b65\u8d44\u6599\uff0c\u518d\u901a\u8fc7\u7f51\u7edc\u641c\u7d22\u6269\u5145\u4fe1\u606f\uff0c\u6700\u540e\u901a\u8fc7\u62c6\u89e3\u91cd\u7ec4\u7684\u65b9\u5f0f\uff0c\u786e\u4fdd\u6700\u7ec8\u5448\u73b0\u7684\u5185\u5bb9\u65e2\u51c6\u786e\u53c8\u53ef\u9760\u3002\u5c31\u50cf\u662f\u7ed9RAG\u88c5\u4e0a\u4e86\u4e00\u4e2a\u8d28\u91cf\u63a7\u5236\u7cfb\u7edf\uff0c\u8ba9\u5b83\u4ea7\u51fa\u7684\u5185\u5bb9\u66f4\u503c\u5f97\u4fe1\u8d56\u3002<\/p><\/blockquote>\n

\u2022 \u65f6\u95f4\uff1a01.29<\/p>\n

\u2022 \u8bba\u6587\uff1aCorrective Retrieval<\/a> Augmented Generation<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/HuskyInSalt\/CRAG<\/p>\n

CRAG\u901a\u8fc7\u8bbe\u8ba1\u8f7b\u91cf\u7ea7\u7684\u68c0\u7d22\u8bc4\u4f30\u5668\u548c\u5f15\u5165\u5927\u89c4\u6a21\u7f51\u7edc\u641c\u7d22\uff0c\u6765\u6539\u8fdb\u68c0\u7d22\u6587\u6863\u7684\u8d28\u91cf\uff0c\u5e76\u901a\u8fc7\u5206\u89e3\u518d\u91cd\u7ec4\u7b97\u6cd5\u8fdb\u4e00\u6b65\u63d0\u70bc\u68c0\u7d22\u5230\u7684\u4fe1\u606f\uff0c\u4ece\u800c\u63d0\u5347\u751f\u6210\u6587\u672c\u7684\u51c6\u786e\u6027\u548c\u53ef\u9760\u6027\u3002CRAG\u662f\u5bf9\u73b0\u6709RAG\u6280\u672f\u7684\u6709\u76ca\u8865\u5145\u548c\u6539\u8fdb\uff0c\u5b83\u901a\u8fc7\u81ea\u6211\u6821\u6b63\u68c0\u7d22\u7ed3\u679c\uff0c\u589e\u5f3a\u4e86\u751f\u6210\u6587\u672c\u7684\u9c81\u68d2\u6027\u3002<\/p>\n

<\/p>\n

(04) RAPTOR\u3010\u5206\u5c42\u5f52\u7eb3\u3011<\/h2>\n

\u5206\u5c42\u5f52\u7eb3<\/strong>\uff1a\u50cf\u4e2a\u5584\u4e8e\u7ec4\u7ec7\u7684\u56fe\u4e66\u7ba1\u7406\u5458\uff0c\u5c06\u6587\u6863\u5185\u5bb9\u81ea\u4e0b\u800c\u4e0a\u5730\u6574\u7406\u6210\u6811\u72b6\u7ed3\u6784\uff0c\u8ba9\u4fe1\u606f\u68c0\u7d22\u80fd\u5728\u4e0d\u540c\u5c42\u7ea7\u95f4\u7075\u6d3b\u7a7f\u68ad\uff0c\u65e2\u80fd\u770b\u5230\u6574\u4f53\u6982\u8981\uff0c\u53c8\u80fd\u6df1\u5165\u7ec6\u8282\u3002<\/p><\/blockquote>\n

\u2022 \u65f6\u95f4\uff1a01.31<\/p>\n

\u2022 \u8bba\u6587\uff1aRAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/parthsarthi03\/raptor<\/p>\n

RAPTOR\uff08Recursive Abstractive Processing for Tree-Organized Retrieval\uff09\u5f15\u5165\u4e86\u4e00\u79cd\u65b0\u65b9\u6cd5\uff0c\u5373\u9012\u5f52\u5d4c\u5165\u3001\u805a\u7c7b\u548c\u603b\u7ed3\u6587\u672c\u5757\uff0c\u4ece\u4e0b\u5f80\u4e0a\u6784\u5efa\u5177\u6709\u4e0d\u540c\u603b\u7ed3\u7ea7\u522b\u7684\u6811\u3002\u5728\u63a8\u7406\u65f6\uff0cRAPTOR \u6a21\u578b\u4ece\u8fd9\u68f5\u6811\u4e2d\u68c0\u7d22\uff0c\u6574\u5408\u4e0d\u540c\u62bd\u8c61\u7ea7\u522b\u7684\u957f\u6587\u6863\u4e2d\u7684\u4fe1\u606f\u3002<\/p>\n

<\/p>\n

(05) T-RAG\u3010\u79c1\u4eba\u987e\u95ee\u3011<\/h2>\n

\u79c1\u4eba\u987e\u95ee<\/strong>\uff1a\u50cf\u4e2a\u719f\u6089\u7ec4\u7ec7\u67b6\u6784\u7684\u5185\u90e8\u987e\u95ee\uff0c\u5584\u4e8e\u5229\u7528\u6811\u72b6\u7ed3\u6784\u7ec4\u7ec7\u4fe1\u606f\uff0c\u5728\u4fdd\u62a4\u9690\u79c1\u7684\u540c\u65f6\uff0c\u9ad8\u6548\u4e14\u7ecf\u6d4e\u5730\u63d0\u4f9b\u672c\u5730\u5316\u670d\u52a1\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aT-RAG: Lessons from the LLM Trenches<\/p>\n

T-RAG\uff08\u6811\u72b6\u68c0\u7d22\u589e\u5f3a\u751f\u6210\uff09\u7ed3\u5408RAG\u4e0e\u5fae\u8c03\u7684\u5f00\u6e90LLM\uff0c\u4f7f\u7528\u6811\u7ed3\u6784\u6765\u8868\u793a\u7ec4\u7ec7\u5185\u7684\u5b9e\u4f53\u5c42\u6b21\u7ed3\u6784\u589e\u5f3a\u4e0a\u4e0b\u6587\uff0c\u5229\u7528\u672c\u5730\u6258\u7ba1\u7684\u5f00\u6e90\u6a21\u578b\u6765\u89e3\u51b3\u6570\u636e\u9690\u79c1\u95ee\u9898\uff0c\u540c\u65f6\u89e3\u51b3\u63a8\u7406\u5ef6\u8fdf\u3001\u4ee4\u724c\u4f7f\u7528\u6210\u672c\u4ee5\u53ca\u533a\u57df\u548c\u5730\u7406\u53ef\u7528\u6027\u95ee\u9898\u3002<\/p>\n

<\/p>\n

(06) RAT\u3010\u601d\u8003\u8005\u3011<\/h2>\n

\u601d\u8003\u8005<\/strong>\uff1a\u50cf\u4e2a\u5584\u4e8e\u53cd\u601d\u7684\u5bfc\u5e08\uff0c\u4e0d\u662f\u4e00\u6b21\u6027\u5f97\u51fa\u7ed3\u8bba\uff0c\u800c\u662f\u5148\u6709\u521d\u6b65\u60f3\u6cd5\uff0c\u7136\u540e\u5229\u7528\u68c0\u7d22\u5230\u7684\u76f8\u5173\u4fe1\u606f\uff0c\u4e0d\u65ad\u5ba1\u89c6\u548c\u5b8c\u5584\u6bcf\u4e00\u6b65\u63a8\u7406\u8fc7\u7a0b\uff0c\u8ba9\u601d\u7ef4\u94fe\u6761\u66f4\u52a0\u4e25\u5bc6\u53ef\u9760\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aRAT: Retrieval Augmented Thoughts Elicit<\/a> Context-Aware Reasoning in Long-Horizon Generation<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/CraftJarvis\/RAT<\/p>\n

RAT\uff08\u68c0\u7d22\u589e\u5f3a\u601d\u7ef4\uff09\u5728\u751f\u6210\u521d\u59cb\u96f6\u6837\u672c\u601d\u7ef4\u94fe\uff08CoT\uff09\u540e\uff0c\u5229\u7528\u4e0e\u4efb\u52a1\u67e5\u8be2\u3001\u5f53\u524d\u548c\u8fc7\u53bb\u601d\u7ef4\u6b65\u9aa4\u76f8\u5173\u7684\u68c0\u7d22\u4fe1\u606f\u9010\u4e2a\u4fee\u8ba2\u6bcf\u4e2a\u601d\u7ef4\u6b65\u9aa4\uff0cRAT\u53ef\u663e\u8457\u63d0\u9ad8\u5404\u79cd\u957f\u65f6\u751f\u6210\u4efb\u52a1\u4e0a\u7684\u6027\u80fd\u3002<\/p>\n

<\/p>\n

(07) RAFT\u3010\u5f00\u5377\u9ad8\u624b\u3011<\/h2>\n

\u5f00\u5377\u9ad8\u624b<\/strong>\uff1a\u50cf\u4e2a\u4f18\u79c0\u7684\u8003\u751f\uff0c\u4e0d\u4ec5\u4f1a\u627e\u5bf9\u53c2\u8003\u8d44\u6599\uff0c\u8fd8\u80fd\u51c6\u786e\u5f15\u7528\u5173\u952e\u5185\u5bb9\uff0c\u5e76\u6e05\u6670\u5730\u89e3\u91ca\u63a8\u7406\u8fc7\u7a0b\uff0c\u8ba9\u7b54\u6848\u65e2\u6709\u636e\u53ef\u5faa\u53c8\u5408\u60c5\u5408\u7406\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aRAFT: Adapting Language Model to Domain Specific RAG<\/a><\/p>\n

RAFT\u65e8\u5728\u63d0\u9ad8\u6a21\u578b\u5728\u7279\u5b9a\u9886\u57df\u5185\u7684\u201c\u5f00\u5377\u201d\u73af\u5883\u4e2d\u56de\u7b54\u95ee\u9898\u7684\u80fd\u529b\uff0c\u901a\u8fc7\u8bad\u7ec3\u6a21\u578b\u5ffd\u7565\u65e0\u5173\u6587\u6863\uff0c\u5e76\u9010\u5b57\u5f15\u7528\u76f8\u5173\u6587\u6863\u4e2d\u7684\u6b63\u786e\u5e8f\u5217\u6765\u56de\u7b54\u95ee\u9898\uff0c\u7ed3\u5408\u601d\u7ef4\u94fe\u5f0f\u54cd\u5e94\uff0c\u663e\u8457\u63d0\u5347\u4e86\u6a21\u578b\u7684\u63a8\u7406\u80fd\u529b\u3002<\/p>\n

<\/p>\n

(08) Adaptive-RAG\u3010\u56e0\u6750\u65bd\u6559\u3011<\/h2>\n

\u56e0\u6750\u65bd\u6559<\/strong>\uff1a\u9762\u5bf9\u4e0d\u540c\u96be\u5ea6\u7684\u95ee\u9898\uff0c\u5b83\u4f1a\u667a\u80fd\u5730\u9009\u62e9\u6700\u5408\u9002\u7684\u89e3\u7b54\u65b9\u5f0f\u3002\u7b80\u5355\u95ee\u9898\u76f4\u63a5\u56de\u7b54\uff0c\u590d\u6742\u95ee\u9898\u5219\u4f1a\u67e5\u9605\u66f4\u591a\u8d44\u6599\u6216\u5206\u6b65\u9aa4\u63a8\u7406\uff0c\u5c31\u50cf\u4e00\u4e2a\u7ecf\u9a8c\u4e30\u5bcc\u7684\u8001\u5e08\uff0c\u61c2\u5f97\u6839\u636e\u5b66\u751f\u7684\u5177\u4f53\u95ee\u9898\u8c03\u6574\u6559\u5b66\u65b9\u6cd5\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aAdaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/starsuzi\/Adaptive-RAG<\/p>\n

Adaptive-RAG\u6839\u636e\u67e5\u8be2\u7684\u590d\u6742\u7a0b\u5ea6\u52a8\u6001\u9009\u62e9\u6700\u9002\u5408\u7684\u68c0\u7d22\u589e\u5f3a\u7b56\u7565\uff0c\u4ece\u6700\u7b80\u5355\u5230\u6700\u590d\u6742\u7684\u7b56\u7565\u4e2d\u52a8\u6001\u5730\u4e3aLLM\u9009\u62e9\u6700\u5408\u9002\u7684\u7b56\u7565\u3002\u8fd9\u4e2a\u9009\u62e9\u8fc7\u7a0b\u901a\u8fc7\u4e00\u4e2a\u5c0f\u8bed\u8a00\u6a21\u578b\u5206\u7c7b\u5668\u6765\u5b9e\u73b0\uff0c\u9884\u6d4b\u67e5\u8be2\u7684\u590d\u6742\u6027\u5e76\u81ea\u52a8\u6536\u96c6\u6807\u7b7e\u4ee5\u4f18\u5316\u9009\u62e9\u8fc7\u7a0b\u3002\u8fd9\u79cd\u65b9\u6cd5\u63d0\u4f9b\u4e86\u4e00\u79cd\u5e73\u8861\u7684\u7b56\u7565\uff0c\u80fd\u591f\u5728\u8fed\u4ee3\u5f0f\u548c\u5355\u6b65\u68c0\u7d22\u589e\u5f3a\u578b LLMs \u4ee5\u53ca\u65e0\u68c0\u7d22\u65b9\u6cd5\u4e4b\u95f4\u65e0\u7f1d\u9002\u5e94\uff0c\u4ee5\u5e94\u5bf9\u4e00\u7cfb\u5217\u67e5\u8be2\u590d\u6742\u5ea6\u3002<\/p>\n

<\/p>\n

(09) HippoRAG\u3010\u6d77\u9a6c\u4f53\u3011<\/h2>\n

\u6d77\u9a6c\u4f53<\/strong>\uff1a\u50cf\u4eba\u8111\u6d77\u9a6c\u4f53\u4e00\u6837\uff0c\u628a\u65b0\u65e7\u77e5\u8bc6\u5de7\u5999\u7f16\u7ec7\u6210\u7f51\u3002\u4e0d\u662f\u7b80\u5355\u5730\u5806\u79ef\u4fe1\u606f\uff0c\u800c\u662f\u8ba9\u6bcf\u6761\u65b0\u77e5\u8bc6\u90fd\u627e\u5230\u6700\u6070\u5f53\u7684\u5f52\u5c5e\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aHippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/OSU-NLP-Group\/HippoRAG<\/p>\n

HippoRAG\u662f\u4e00\u79cd\u65b0\u9896\u7684\u68c0\u7d22\u6846\u67b6\uff0c\u5176\u7075\u611f\u6765\u6e90\u4e8e\u4eba\u7c7b\u957f\u671f\u8bb0\u5fc6\u7684\u6d77\u9a6c\u4f53\u7d22\u5f15\u7406\u8bba\uff0c\u65e8\u5728\u5b9e\u73b0\u5bf9\u65b0\u7ecf\u9a8c\u66f4\u6df1\u5165\u3001\u66f4\u9ad8\u6548\u7684\u77e5\u8bc6\u6574\u5408\u3002HippoRAG\u534f\u540c\u7f16\u6392 LLMs\u3001\u77e5\u8bc6\u56fe\u8c31\u548c\u4e2a\u6027\u5316PageRank\u7b97\u6cd5\uff0c\u4ee5\u6a21\u62df\u65b0\u76ae\u5c42\u548c\u6d77\u9a6c\u4f53\u5728\u4eba\u7c7b\u8bb0\u5fc6\u4e2d\u7684\u4e0d\u540c\u89d2\u8272\u3002<\/p>\n

<\/p>\n

(10) RAE\u3010\u667a\u80fd\u7f16\u8f91\u3011<\/h2>\n

\u667a\u80fd\u7f16\u8f91<\/strong>\uff1a\u50cf\u4e2a\u7ec6\u5fc3\u7684\u65b0\u95fb\u7f16\u8f91\uff0c\u4e0d\u4ec5\u4f1a\u6df1\u5165\u6316\u6398\u76f8\u5173\u4e8b\u5b9e\uff0c\u8fd8\u80fd\u901a\u8fc7\u8fde\u73af\u63a8\u7406\u627e\u51fa\u5bb9\u6613\u88ab\u5ffd\u7565\u7684\u5173\u952e\u4fe1\u606f\uff0c\u540c\u65f6\u61c2\u5f97\u5220\u51cf\u5197\u4f59\u5185\u5bb9\uff0c\u786e\u4fdd\u6700\u7ec8\u5448\u73b0\u7684\u4fe1\u606f\u65e2\u51c6\u786e\u53c8\u7cbe\u70bc\uff0c\u907f\u514d”\u8bf4\u5f97\u5929\u82b1\u4e71\u5760\u5374\u4e0d\u9760\u8c31”\u7684\u95ee\u9898\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aRetrieval-enhanced Knowledge Editing in Language Models for Multi-Hop Question Answering<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/sycny\/RAE<\/p>\n

RAE\uff08\u591a\u8df3\u95ee\u7b54\u68c0\u7d22\u589e\u5f3a\u6a21\u578b\u7f16\u8f91\u6846\u67b6\uff09\u9996\u5148\u68c0\u7d22\u7ecf\u8fc7\u7f16\u8f91\u7684\u4e8b\u5b9e\uff0c\u7136\u540e\u901a\u8fc7\u4e0a\u4e0b\u6587\u5b66\u4e60\u6765\u4f18\u5316\u8bed\u8a00\u6a21\u578b\u3002\u57fa\u4e8e\u4e92\u4fe1\u606f\u6700\u5927\u5316\u7684\u68c0\u7d22\u65b9\u6cd5\u5229\u7528\u5927\u578b\u8bed\u8a00\u6a21\u578b\u7684\u63a8\u7406\u80fd\u529b\u6765\u8bc6\u522b\u4f20\u7edf\u57fa\u4e8e\u76f8\u4f3c\u6027\u7684\u641c\u7d22\u53ef\u80fd\u4f1a\u9519\u8fc7\u7684\u94fe\u5f0f\u4e8b\u5b9e\u3002\u6b64\u5916\u6846\u67b6\u5305\u62ec\u4e00\u79cd\u4fee\u526a\u7b56\u7565\uff0c\u4ee5\u4ece\u68c0\u7d22\u5230\u7684\u4e8b\u5b9e\u4e2d\u6d88\u9664\u5197\u4f59\u4fe1\u606f\uff0c\u8fd9\u63d0\u9ad8\u4e86\u7f16\u8f91\u51c6\u786e\u6027\u5e76\u51cf\u8f7b\u4e86\u5e7b\u89c9\u95ee\u9898\u3002<\/p>\n

<\/p>\n

(11) RAGCache\u3010\u4ed3\u50a8\u5458\u3011<\/h2>\n

\u4ed3\u50a8\u5458<\/strong>\uff1a\u50cf\u5927\u578b\u7269\u6d41\u4e2d\u5fc3\u4e00\u6837\uff0c\u628a\u5e38\u7528\u77e5\u8bc6\u653e\u5728\u6700\u5bb9\u6613\u53d6\u7684\u8d27\u67b6\u4e0a\u3002\u61c2\u5f97\u628a\u7ecf\u5e38\u7528\u7684\u5305\u88f9\u653e\u5728\u95e8\u53e3\uff0c\u628a\u4e0d\u5e38\u7528\u7684\u653e\u5728\u540e\u4ed3\uff0c\u8ba9\u53d6\u8d27\u6548\u7387\u6700\u5927\u5316\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aRAGCache: Efficient Knowledge Caching for Retrieval-Augmented Generation<\/p>\n

RAGCache\u662f\u4e00\u79cd\u4e3aRAG\u91cf\u8eab\u5b9a\u5236\u7684\u65b0\u578b\u591a\u7ea7\u52a8\u6001\u7f13\u5b58\u7cfb\u7edf\uff0c\u5b83\u5c06\u68c0\u7d22\u5230\u7684\u77e5\u8bc6\u7684\u4e2d\u95f4\u72b6\u6001\u7ec4\u7ec7\u5728\u77e5\u8bc6\u6811\u4e2d\uff0c\u5e76\u5728GPU\u548c\u4e3b\u673a\u5185\u5b58\u5c42\u6b21\u7ed3\u6784\u4e2d\u8fdb\u884c\u7f13\u5b58\u3002RAGCache\u63d0\u51fa\u4e86\u4e00\u79cd\u8003\u8651\u5230LLM\u63a8\u7406\u7279\u5f81\u548cRAG\u68c0\u7d22\u6a21\u5f0f\u7684\u66ff\u6362\u7b56\u7565\u3002\u5b83\u8fd8\u52a8\u6001\u5730\u91cd\u53e0\u68c0\u7d22\u548c\u63a8\u7406\u6b65\u9aa4\uff0c\u4ee5\u6700\u5c0f\u5316\u7aef\u5230\u7aef\u5ef6\u8fdf\u3002<\/p>\n

<\/p>\n

(12) GraphRAG\u3010\u793e\u533a\u6458\u8981\u3011<\/h2>\n

\u793e\u533a\u6458\u8981<\/strong>\uff1a\u5148\u628a\u5c0f\u533a\u5c45\u6c11\u7684\u5173\u7cfb\u7f51\u7406\u6e05\u695a\uff0c\u518d\u7ed9\u6bcf\u4e2a\u90bb\u91cc\u5708\u505a\u4e2a\u7b80\u4ecb\u3002\u6709\u4eba\u95ee\u8def\u65f6\uff0c\u5404\u4e2a\u90bb\u91cc\u5708\u63d0\u4f9b\u7ebf\u7d22\uff0c\u6700\u540e\u6574\u5408\u6210\u6700\u5b8c\u6574\u7684\u7b54\u6848\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aFrom Local to Global: A Graph RAG Approach to Query-Focused Summarization<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/microsoft\/graphrag<\/p>\n

GraphRAG\u5206\u4e24\u4e2a\u9636\u6bb5\u6784\u5efa\u57fa\u4e8e\u56fe\u7684\u6587\u672c\u7d22\u5f15\uff1a\u9996\u5148\u4ece\u6e90\u6587\u6863\u4e2d\u63a8\u5bfc\u51fa\u5b9e\u4f53\u77e5\u8bc6\u56fe\uff0c\u7136\u540e\u4e3a\u6240\u6709\u7d27\u5bc6\u76f8\u5173\u5b9e\u4f53\u7684\u7ec4\u9884\u751f\u6210\u793e\u533a\u6458\u8981\u3002\u7ed9\u5b9a\u4e00\u4e2a\u95ee\u9898\uff0c\u6bcf\u4e2a\u793e\u533a\u6458\u8981\u7528\u4e8e\u751f\u6210\u90e8\u5206\u54cd\u5e94\uff0c\u7136\u540e\u5728\u5411\u7528\u6237\u7684\u6700\u7ec8\u54cd\u5e94\u4e2d\u518d\u6b21\u603b\u7ed3\u6240\u6709\u90e8\u5206\u54cd\u5e94\u3002<\/p>\n

<\/p>\n

(13) R4\u3010\u7f16\u6392\u5927\u5e08\u3011<\/h2>\n

\u7f16\u6392\u5927\u5e08<\/strong>\uff1a\u50cf\u4e2a\u6392\u7248\u9ad8\u624b\uff0c\u901a\u8fc7\u4f18\u5316\u6750\u6599\u7684\u987a\u5e8f\u548c\u5448\u73b0\u65b9\u5f0f\u6765\u63d0\u5347\u8f93\u51fa\u8d28\u91cf\uff0c\u65e0\u9700\u6539\u52a8\u6838\u5fc3\u6a21\u578b\u5c31\u80fd\u8ba9\u5185\u5bb9\u66f4\u6709\u6761\u7406\uff0c\u91cd\u70b9\u66f4\u7a81\u51fa\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aR4: Reinforced Retriever-Reorder-Responder for Retrieval-Augmented Large Language Models<\/p>\n

R4\uff08Reinforced Retriever-Reorder-Responder\uff09\u7528\u4e8e\u4e3a\u68c0\u7d22\u589e\u5f3a\u578b\u5927\u8bed\u8a00\u6a21\u578b\u5b66\u4e60\u6587\u6863\u6392\u5e8f\uff0c\u4ece\u800c\u5728\u5927\u8bed\u8a00\u6a21\u578b\u7684\u5927\u91cf\u53c2\u6570\u4fdd\u6301\u51bb\u7ed3\u7684\u60c5\u51b5\u4e0b\u8fdb\u4e00\u6b65\u589e\u5f3a\u5176\u751f\u6210\u80fd\u529b\u3002\u91cd\u6392\u5e8f\u5b66\u4e60\u8fc7\u7a0b\u6839\u636e\u751f\u6210\u54cd\u5e94\u7684\u8d28\u91cf\u5206\u4e3a\u4e24\u4e2a\u6b65\u9aa4\uff1a\u6587\u6863\u987a\u5e8f\u8c03\u6574\u548c\u6587\u6863\u8868\u793a\u589e\u5f3a\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u6587\u6863\u987a\u5e8f\u8c03\u6574\u65e8\u5728\u57fa\u4e8e\u56fe\u6ce8\u610f\u529b\u5b66\u4e60\u5c06\u68c0\u7d22\u5230\u7684\u6587\u6863\u6392\u5e8f\u7ec4\u7ec7\u5230\u5f00\u5934\u3001\u4e2d\u95f4\u548c\u7ed3\u5c3e\u4f4d\u7f6e\uff0c\u4ee5\u6700\u5927\u5316\u54cd\u5e94\u8d28\u91cf\u7684\u5f3a\u5316\u5956\u52b1\u3002\u6587\u6863\u8868\u793a\u589e\u5f3a\u901a\u8fc7\u6587\u6863\u7ea7\u68af\u5ea6\u5bf9\u6297\u5b66\u4e60\u8fdb\u4e00\u6b65\u7ec6\u5316\u8d28\u91cf\u8f83\u5dee\u7684\u54cd\u5e94\u7684\u68c0\u7d22\u6587\u6863\u8868\u793a\u3002<\/p>\n

<\/p>\n

(14) IM-RAG\u3010\u81ea\u8a00\u81ea\u8bed\u3011<\/h2>\n

\u81ea\u8a00\u81ea\u8bed<\/strong>\uff1a\u9047\u5230\u95ee\u9898\u65f6\u4f1a\u5728\u5fc3\u91cc\u76d8\u7b97”\u6211\u9700\u8981\u67e5\u4ec0\u4e48\u8d44\u6599”\u3001”\u8fd9\u4e2a\u4fe1\u606f\u591f\u4e0d\u591f”\uff0c\u901a\u8fc7\u4e0d\u65ad\u7684\u5185\u5fc3\u5bf9\u8bdd\u6765\u5b8c\u5584\u7b54\u6848\uff0c\u8fd9\u79cd”\u72ec\u767d”\u80fd\u529b\u50cf\u4eba\u7c7b\u4e13\u5bb6\u4e00\u6837\uff0c\u80fd\u591f\u9010\u6b65\u6df1\u5165\u601d\u8003\u5e76\u89e3\u51b3\u590d\u6742\u95ee\u9898\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aIM-RAG: Multi-Round Retrieval-Augmented Generation Through Learning Inner Monologues<\/p>\n

IM-RAG\u901a\u8fc7\u5b66\u4e60\u5185\u90e8\u72ec\u767d\uff08Inner Monologues\uff09\u6765\u8fde\u63a5IR\u7cfb\u7edf\u4e0eLLMs\uff0c\u4ece\u800c\u652f\u6301\u591a\u8f6e\u68c0\u7d22\u589e\u5f3a\u751f\u6210\u3002\u8be5\u65b9\u6cd5\u5c06\u4fe1\u606f\u68c0\u7d22\u7cfb\u7edf\u4e0e\u5927\u578b\u8bed\u8a00\u6a21\u578b\u76f8\u6574\u5408\uff0c\u901a\u8fc7\u5b66\u4e60\u5185\u5fc3\u72ec\u767d\u6765\u652f\u6301\u591a\u8f6e\u68c0\u7d22\u589e\u5f3a\u751f\u6210\u3002\u5728\u5185\u5fc3\u72ec\u767d\u8fc7\u7a0b\u4e2d\uff0c\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5145\u5f53\u6838\u5fc3\u63a8\u7406\u6a21\u578b\uff0c\u5b83\u65e2\u53ef\u4ee5\u901a\u8fc7\u68c0\u7d22\u5668\u63d0\u51fa\u67e5\u8be2\u4ee5\u6536\u96c6\u66f4\u591a\u4fe1\u606f\uff0c\u4e5f\u53ef\u4ee5\u57fa\u4e8e\u5bf9\u8bdd\u4e0a\u4e0b\u6587\u63d0\u4f9b\u6700\u7ec8\u7b54\u6848\u3002\u6211\u4eec\u8fd8\u5f15\u5165\u4e86\u4e00\u4e2a\u4f18\u5316\u5668\uff0c\u5b83\u80fd\u5bf9\u68c0\u7d22\u5668\u7684\u8f93\u51fa\u8fdb\u884c\u6539\u8fdb\uff0c\u6709\u6548\u5730\u5f25\u5408\u63a8\u7406\u5668\u4e0e\u80fd\u529b\u5404\u5f02\u7684\u4fe1\u606f\u68c0\u7d22\u6a21\u5757\u4e4b\u95f4\u7684\u5dee\u8ddd\uff0c\u5e76\u4fc3\u8fdb\u591a\u8f6e\u901a\u4fe1\u3002\u6574\u4e2a\u5185\u5fc3\u72ec\u767d\u8fc7\u7a0b\u901a\u8fc7\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u8fdb\u884c\u4f18\u5316\uff0c\u5728\u6b64\u8fc7\u7a0b\u4e2d\u8fd8\u5f15\u5165\u4e86\u4e00\u4e2a\u8fdb\u5c55\u8ddf\u8e2a\u5668\u6765\u63d0\u4f9b\u4e2d\u95f4\u6b65\u9aa4\u5956\u52b1\uff0c\u5e76\u4e14\u7b54\u6848\u9884\u6d4b\u4f1a\u901a\u8fc7\u76d1\u7763\u5fae\u8c03\uff08SFT\uff09\u8fdb\u4e00\u6b65\u5355\u72ec\u4f18\u5316\u3002<\/p>\n

<\/p>\n

(15) AntGroup-GraphRAG\u3010\u767e\u5bb6\u4e4b\u957f\u3011<\/h2>\n

\u767e\u5bb6\u4e4b\u957f<\/strong>\uff1a\u6c47\u96c6\u884c\u4e1a\u767e\u5bb6\u4e4b\u957f\uff0c\u64c5\u7528\u591a\u79cd\u65b9\u5f0f\u5feb\u901f\u5b9a\u4f4d\u4fe1\u606f\uff0c\u65e2\u80fd\u63d0\u4f9b\u7cbe\u51c6\u68c0\u7d22\uff0c\u53c8\u80fd\u7406\u89e3\u81ea\u7136\u8bed\u8a00\u67e5\u8be2\uff0c\u8ba9\u590d\u6742\u7684\u77e5\u8bc6\u68c0\u7d22\u53d8\u5f97\u65e2\u7ecf\u6d4e\u53c8\u9ad8\u6548\u3002<\/p><\/blockquote>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/eosphoros-ai\/DB-GPT<\/p>\n

\u8682\u8681TuGraph\u56e2\u961f\u57fa\u4e8eDB-GPT\u6784\u5efa\u7684\u5f00\u6e90GraphRAG\u6846\u67b6\uff0c\u517c\u5bb9\u4e86\u5411\u91cf\u3001\u56fe\u8c31\u3001\u5168\u6587\u7b49\u591a\u79cd\u77e5\u8bc6\u5e93\u7d22\u5f15\u5e95\u5ea7\uff0c\u652f\u6301\u4f4e\u6210\u672c\u7684\u77e5\u8bc6\u62bd\u53d6\u3001\u6587\u6863\u7ed3\u6784\u56fe\u8c31\u3001\u56fe\u793e\u533a\u6458\u8981\u4e0e\u6df7\u5408\u68c0\u7d22\u4ee5\u89e3\u51b3QFS\u95ee\u7b54\u95ee\u9898\u3002\u53e6\u5916\u4e5f\u63d0\u4f9b\u4e86\u5173\u952e\u8bcd\u3001\u5411\u91cf\u548c\u81ea\u7136\u8bed\u8a00\u7b49\u591a\u6837\u5316\u7684\u68c0\u7d22\u80fd\u529b\u652f\u6301\u3002<\/p>\n

<\/p>\n

(16) Kotaemon\u3010\u4e50\u9ad8\u3011<\/h2>\n

\u4e50\u9ad8<\/strong>\uff1a\u4e00\u5957\u73b0\u6210\u7684\u95ee\u7b54\u79ef\u6728\u5957\u88c5\uff0c\u65e2\u80fd\u76f4\u63a5\u62ff\u6765\u7528\uff0c\u53c8\u80fd\u81ea\u7531\u62c6\u88c5\u6539\u9020\u3002\u7528\u6237\u8981\u7528\u5c31\u7528\uff0c\u5f00\u53d1\u8981\u6539\u5c31\u6539\uff0c\u968f\u5fc3\u6240\u6b32\u4e0d\u5931\u7ae0\u6cd5\u3002<\/p><\/blockquote>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/Cinnamon\/kotaemon<\/p>\n

\u4e00\u4e2a\u5f00\u6e90\u7684\u5e72\u51c0\u4e14\u53ef\u5b9a\u5236\u7684RAG UI\uff0c\u7528\u4e8e\u6784\u5efa\u548c\u5b9a\u5236\u81ea\u5df1\u7684\u6587\u6863\u95ee\u7b54\u7cfb\u7edf\u3002\u65e2\u8003\u8651\u4e86\u6700\u7ec8\u7528\u6237\u7684\u9700\u6c42\uff0c\u4e5f\u8003\u8651\u4e86\u5f00\u53d1\u8005\u7684\u9700\u6c42\u3002<\/p>\n

<\/p>\n

(17) FlashRAG\u3010\u767e\u5b9d\u7bb1\u3011<\/h2>\n

\u767e\u5b9d\u7bb1<\/strong>\uff1a\u628a\u5404\u8defRAG\u795e\u5668\u6253\u5305\u6210\u4e00\u4e2a\u5de5\u5177\u5305\uff0c\u8ba9\u7814\u7a76\u8005\u50cf\u6311\u9009\u79ef\u6728\u4e00\u6837\uff0c\u968f\u5fc3\u6240\u6b32\u5730\u642d\u5efa\u81ea\u5df1\u7684\u68c0\u7d22\u6a21\u578b\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aFlashRAG: A Modular Toolkit for Efficient Retrieval-Augmented Generation Research<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/RUC-NLPIR\/FlashRAG<\/p>\n

FlashRAG\u662f\u4e00\u4e2a\u9ad8\u6548\u4e14\u6a21\u5757\u5316\u7684\u5f00\u6e90\u5de5\u5177\u5305\uff0c\u65e8\u5728\u5e2e\u52a9\u7814\u7a76\u4eba\u5458\u5728\u7edf\u4e00\u6846\u67b6\u5185\u91cd\u73b0\u73b0\u6709\u7684RAG\u65b9\u6cd5\u5e76\u5f00\u53d1\u4ed6\u4eec\u81ea\u5df1\u7684RAG\u7b97\u6cd5\u3002\u6211\u4eec\u7684\u5de5\u5177\u5305\u5b9e\u73b0\u4e8612\u79cd\u5148\u8fdb\u7684RAG\u65b9\u6cd5\uff0c\u5e76\u6536\u96c6\u548c\u6574\u7406\u4e8632\u4e2a\u57fa\u51c6\u6570\u636e\u96c6\u3002<\/p>\n

<\/p>\n

(18) GRAG\u3010\u4fa6\u63a2\u3011<\/h2>\n

\u4fa6\u63a2<\/strong>\uff1a\u4e0d\u6ee1\u8db3\u4e8e\u8868\u9762\u7ebf\u7d22\uff0c\u6df1\u5165\u6316\u6398\u6587\u672c\u4e4b\u95f4\u7684\u5173\u8054\u7f51\u7edc\uff0c\u50cf\u7834\u6848\u4e00\u6837\u8ffd\u8e2a\u6bcf\u6761\u4fe1\u606f\u80cc\u540e\u7684\u771f\u76f8\uff0c\u8ba9\u7b54\u6848\u66f4\u51c6\u786e\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aGRAG: Graph Retrieval-Augmented Generation<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/HuieL\/GRAG<\/p>\n

\u4f20\u7edfRAG\u6a21\u578b\u5728\u5904\u7406\u590d\u6742\u7684\u56fe\u7ed3\u6784\u6570\u636e\u65f6\u5ffd\u89c6\u4e86\u6587\u672c\u4e4b\u95f4\u7684\u8054\u7cfb\u548c\u6570\u636e\u5e93\u7684\u62d3\u6251\u4fe1\u606f\uff0c\u4ece\u800c\u5bfc\u81f4\u4e86\u6027\u80fd\u74f6\u9888\u3002GRAG\u901a\u8fc7\u5f3a\u8c03\u5b50\u56fe\u7ed3\u6784\u7684\u91cd\u8981\u6027\uff0c\u663e\u8457\u63d0\u5347\u4e86\u68c0\u7d22\u548c\u751f\u6210\u8fc7\u7a0b\u7684\u6027\u80fd\u5e76\u964d\u4f4e\u5e7b\u89c9\u3002<\/p>\n

<\/p>\n

(19) Camel-GraphRAG\u3010\u5de6\u53f3\u5f00\u5f13\u3011<\/h2>\n

\u5de6\u53f3\u5f00\u5f13<\/strong>\uff1a\u4e00\u53ea\u773c\u775b\u7528Mistral\u626b\u63cf\u6587\u672c\u63d0\u53d6\u60c5\u62a5\uff0c\u53e6\u53ea\u773c\u775b\u7528Neo4j\u7f16\u7ec7\u5173\u7cfb\u7f51\u3002\u67e5\u627e\u65f6\u5de6\u53f3\u773c\u914d\u5408\uff0c\u65e2\u80fd\u627e\u76f8\u4f3c\u7684\uff0c\u53c8\u80fd\u987a\u7740\u7ebf\u7d22\u56fe\u8ffd\u8e2a\uff0c\u8ba9\u641c\u7d22\u66f4\u5168\u9762\u7cbe\u51c6\u3002<\/p><\/blockquote>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/camel-ai\/camel<\/p>\n

Camel-GraphRAG\u4f9d\u6258Mistral\u6a21\u578b\u63d0\u4f9b\u652f\u6301\uff0c\u4ece\u7ed9\u5b9a\u7684\u5185\u5bb9\u4e2d\u63d0\u53d6\u77e5\u8bc6\u5e76\u6784\u5efa\u77e5\u8bc6\u7ed3\u6784\uff0c\u7136\u540e\u5c06\u8fd9\u4e9b\u4fe1\u606f\u5b58\u50a8\u5728 Neo4j\u56fe\u6570\u636e\u5e93\u4e2d\u3002\u968f\u540e\u91c7\u7528\u4e00\u79cd\u6df7\u5408\u65b9\u6cd5\uff0c\u5c06\u5411\u91cf\u68c0\u7d22\u4e0e\u77e5\u8bc6\u56fe\u8c31\u68c0\u7d22\u76f8\u7ed3\u5408\uff0c\u6765\u67e5\u8be2\u548c\u63a2\u7d22\u6240\u5b58\u50a8\u7684\u77e5\u8bc6\u3002<\/p>\n

<\/p>\n

(20) G-RAG\u3010\u4e32\u95e8\u795e\u5668\u3011<\/h2>\n

\u4e32\u95e8\u795e\u5668<\/strong>\uff1a\u4e0d\u518d\u662f\u5355\u6253\u72ec\u6597\u5730\u67e5\u8d44\u6599\uff0c\u800c\u662f\u7ed9\u6bcf\u4e2a\u77e5\u8bc6\u70b9\u90fd\u5efa\u7acb\u4eba\u9645\u5173\u7cfb\u7f51\u3002\u50cf\u4e2a\u793e\u4ea4\u8fbe\u4eba\uff0c\u4e0d\u4ec5\u77e5\u9053\u6bcf\u4e2a\u670b\u53cb\u7684\u7279\u957f\uff0c\u8fd8\u6e05\u695a\u8c01\u548c\u8c01\u662f\u9152\u8089\u670b\u53cb\uff0c\u627e\u7b54\u6848\u65f6\u76f4\u63a5\u987a\u85e4\u6478\u74dc\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aDon’t Forget to Connect! Improving RAG with Graph-based Reranking<\/p>\n

RAG \u5728\u5904\u7406\u6587\u6863\u4e0e\u95ee\u9898\u4e0a\u4e0b\u6587\u7684\u5173\u7cfb\u65f6\u4ecd\u5b58\u5728\u6311\u6218\uff0c\u5f53\u6587\u6863\u4e0e\u95ee\u9898\u7684\u5173\u8054\u6027\u4e0d\u660e\u663e\u6216\u4ec5\u5305\u542b\u90e8\u5206\u4fe1\u606f\u65f6\uff0c\u6a21\u578b\u53ef\u80fd\u65e0\u6cd5\u6709\u6548\u5229\u7528\u8fd9\u4e9b\u6587\u6863\u3002\u6b64\u5916\uff0c\u5982\u4f55\u5408\u7406\u63a8\u65ad\u6587\u6863\u4e4b\u95f4\u7684\u5173\u8054\u4e5f\u662f\u4e00\u4e2a\u91cd\u8981\u95ee\u9898\u3002G-RAG\u5b9e\u73b0\u4e86RAG\u68c0\u7d22\u5668\u548c\u9605\u8bfb\u5668\u4e4b\u95f4\u57fa\u4e8e\u56fe\u795e\u7ecf\u7f51\u7edc\uff08GNN\uff09\u7684\u91cd\u6392\u5668\u3002\u8be5\u65b9\u6cd5\u7ed3\u5408\u4e86\u6587\u6863\u4e4b\u95f4\u7684\u8fde\u63a5\u4fe1\u606f\u548c\u8bed\u4e49\u4fe1\u606f\uff08\u901a\u8fc7\u62bd\u8c61\u8bed\u4e49\u8868\u793a\u56fe\uff09\uff0c\u4e3a RAG \u63d0\u4f9b\u4e86\u57fa\u4e8e\u4e0a\u4e0b\u6587\u7684\u6392\u5e8f\u5668\u3002<\/p>\n

<\/p>\n

(21) LLM-Graph-Builder\u3010\u642c\u8fd0\u5de5\u3011<\/h2>\n

\u642c\u8fd0\u5de5<\/strong>\uff1a\u7ed9\u6df7\u4e71\u7684\u6587\u5b57\u5b89\u4e2a\u660e\u767d\u7684\u5bb6\u3002\u4e0d\u662f\u7b80\u5355\u5730\u642c\u8fd0\uff0c\u800c\u662f\u50cf\u4e2a\u5f3a\u8feb\u75c7\u60a3\u8005\uff0c\u628a\u6bcf\u4e2a\u77e5\u8bc6\u70b9\u90fd\u8d34\u4e0a\u6807\u7b7e\uff0c\u753b\u4e0a\u5173\u7cfb\u7ebf\uff0c\u6700\u540e\u5728Neo4j\u7684\u6570\u636e\u5e93\u91cc\u76d6\u8d77\u4e00\u5ea7\u4e95\u4e95\u6709\u5e8f\u7684\u77e5\u8bc6\u5927\u53a6\u3002<\/p><\/blockquote>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/neo4j-labs\/llm-graph-builder<\/p>\n

Neo4j\u5f00\u6e90\u7684\u57fa\u4e8eLLM\u63d0\u53d6\u77e5\u8bc6\u56fe\u8c31\u7684\u751f\u6210\u5668\uff0c\u53ef\u4ee5\u628a\u975e\u7ed3\u6784\u5316\u6570\u636e\u8f6c\u6362\u6210Neo4j\u4e2d\u7684\u77e5\u8bc6\u56fe\u8c31\u3002\u5229\u7528\u5927\u6a21\u578b\u4ece\u975e\u7ed3\u6784\u5316\u6570\u636e\u4e2d\u63d0\u53d6\u8282\u70b9\u3001\u5173\u7cfb\u53ca\u5176\u5c5e\u6027\u3002<\/p>\n

<\/p>\n

(22) MRAG\u3010\u516b\u722a\u9c7c\u3011<\/h2>\n

\u516b\u722a\u9c7c<\/strong>\uff1a\u4e0d\u662f\u53ea\u957f\u4e00\u4e2a\u8111\u888b\u6b7b\u78d5\u95ee\u9898\uff0c\u800c\u662f\u50cf\u7ae0\u9c7c\u4e00\u6837\u957f\u51fa\u591a\u4e2a\u89e6\u89d2\uff0c\u6bcf\u4e2a\u89e6\u89d2\u8d1f\u8d23\u6293\u53d6\u4e00\u4e2a\u89d2\u5ea6\u3002\u7b80\u5355\u8bf4\uff0c\u8fd9\u5c31\u662fAI\u7248\u7684”\u4e00\u5fc3\u591a\u7528”\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aMulti-Head RAG: Solving Multi-Aspect Problems with LLMs<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/spcl\/MRAG<\/p>\n

\u73b0\u6709\u7684 RAG \u89e3\u51b3\u65b9\u6848\u5e76\u672a\u4e13\u6ce8\u4e8e\u53ef\u80fd\u9700\u8981\u83b7\u53d6\u5185\u5bb9\u5dee\u5f02\u663e\u8457\u7684\u591a\u4e2a\u6587\u6863\u7684\u67e5\u8be2\u3002\u6b64\u7c7b\u67e5\u8be2\u7ecf\u5e38\u51fa\u73b0\uff0c\u4f46\u5177\u6709\u6311\u6218\u6027\uff0c\u56e0\u4e3a\u8fd9\u4e9b\u6587\u6863\u7684\u5d4c\u5165\u5728\u5d4c\u5165\u7a7a\u95f4\u4e2d\u53ef\u80fd\u76f8\u8ddd\u8f83\u8fdc\uff0c\u4f7f\u5f97\u96be\u4ee5\u5168\u90e8\u68c0\u7d22\u5230\u5b83\u4eec\u3002\u672c\u6587\u4ecb\u7ecd\u4e86\u591a\u5934 RAG\uff08MRAG\uff09\uff0c\u8fd9\u662f\u4e00\u79cd\u65b0\u9896\u7684\u65b9\u6848\uff0c\u65e8\u5728\u901a\u8fc7\u4e00\u4e2a\u7b80\u5355\u800c\u5f3a\u5927\u7684\u60f3\u6cd5\u6765\u586b\u8865\u8fd9\u4e00\u7a7a\u767d\uff1a\u5229\u7528 Transformer<\/a> \u591a\u5934\u6ce8\u610f\u529b\u5c42\u7684\u6fc0\u6d3b\uff0c\u800c\u975e\u89e3\u7801\u5668\u5c42\uff0c\u4f5c\u4e3a\u83b7\u53d6\u591a\u65b9\u9762\u6587\u6863\u7684\u952e\u3002\u5176\u9a71\u52a8\u52a8\u673a\u662f\u4e0d\u540c\u7684\u6ce8\u610f\u529b\u5934\u53ef\u4ee5\u5b66\u4e60\u6355\u83b7\u4e0d\u540c\u7684\u6570\u636e\u65b9\u9762\u3002\u5229\u7528\u76f8\u5e94\u7684\u6fc0\u6d3b\u4f1a\u4ea7\u751f\u4ee3\u8868\u6570\u636e\u9879\u548c\u67e5\u8be2\u5404\u4e2a\u5c42\u9762\u7684\u5d4c\u5165\uff0c\u4ece\u800c\u63d0\u9ad8\u590d\u6742\u67e5\u8be2\u7684\u68c0\u7d22\u51c6\u786e\u6027\u3002<\/p>\n

<\/p>\n

(23) PlanRAG\u3010\u6218\u7565\u5bb6\u3011<\/h2>\n

\u6218\u7565\u5bb6<\/strong>\uff1a\u5148\u5236\u5b9a\u5b8c\u6574\u4f5c\u6218\u8ba1\u5212\uff0c\u518d\u6839\u636e\u89c4\u5219\u548c\u6570\u636e\u5206\u6790\u5c40\u52bf\uff0c\u6700\u540e\u505a\u51fa\u6700\u4f73\u6218\u672f\u51b3\u7b56\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aPlanRAG: A Plan-then-Retrieval Augmented Generation for Generative Large Language Models as Decision Makers<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/myeon9h\/PlanRAG<\/p>\n

PlanRAG\u7814\u7a76\u5982\u4f55\u5229\u7528\u5927\u578b\u8bed\u8a00\u6a21\u578b\u89e3\u51b3\u590d\u6742\u6570\u636e\u5206\u6790\u51b3\u7b56\u95ee\u9898\u7684\u65b9\u6848\uff0c\u901a\u8fc7\u5b9a\u4e49\u51b3\u7b56\u95ee\u7b54\uff08Decision QA\uff09\u4efb\u52a1\uff0c\u5373\u6839\u636e\u51b3\u7b56\u95ee\u9898Q\u3001\u4e1a\u52a1\u89c4\u5219R\u548c\u6570\u636e\u5e93D\uff0c\u786e\u5b9a\u6700\u4f73\u51b3\u7b56d\u3002PlanRAG\u9996\u5148\u751f\u6210\u51b3\u7b56\u8ba1\u5212\uff0c\u7136\u540e\u68c0\u7d22\u5668\u751f\u6210\u6570\u636e\u5206\u6790\u7684\u67e5\u8be2\u3002<\/p>\n

<\/p>\n

(24) FoRAG\u3010\u4f5c\u5bb6\u3011<\/h2>\n

\u4f5c\u5bb6<\/strong>\uff1a\u5148\u5217\u5199\u4f5c\u5927\u7eb2\u6784\u601d\u6587\u7ae0\u6846\u67b6\uff0c\u518d\u9010\u6bb5\u6269\u5145\u5b8c\u5584\u5185\u5bb9\u3002\u540c\u65f6\u8fd8\u914d\u5907\u4e86\u4e00\u4e2a”\u7f16\u8f91”\uff0c\u901a\u8fc7\u4ed4\u7ec6\u7684\u4e8b\u5b9e\u6838\u67e5\u548c\u4fee\u6539\u5efa\u8bae\uff0c\u5e2e\u52a9\u5b8c\u5584\u6bcf\u4e2a\u7ec6\u8282\uff0c\u786e\u4fdd\u4f5c\u54c1\u7684\u8d28\u91cf\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aFoRAG: Factuality-optimized Retrieval Augmented Generation for Web-enhanced Long-form Question Answering<\/p>\n

FoRAG\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u9896\u7684\u5927\u7eb2\u589e\u5f3a\u751f\u6210\u5668\uff0c\u5728\u7b2c\u4e00\u9636\u6bb5\u751f\u6210\u5668\u4f7f\u7528\u5927\u7eb2\u6a21\u677f\uff0c\u6839\u636e\u7528\u6237\u67e5\u8be2\u548c\u4e0a\u4e0b\u6587\u8349\u62df\u7b54\u6848\u5927\u7eb2\uff0c\u7b2c\u4e8c\u9636\u6bb5\u57fa\u4e8e\u751f\u6210\u7684\u5927\u7eb2\u6269\u5c55\u6bcf\u4e2a\u89c2\u70b9\uff0c\u6784\u5efa\u6700\u7ec8\u7b54\u6848\u3002\u540c\u65f6\u63d0\u51fa\u4e00\u79cd\u57fa\u4e8e\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u53cc\u7cbe\u7ec6\u7c92\u5ea6RLHF\u6846\u67b6\u7684\u4e8b\u5b9e\u6027\u4f18\u5316\u65b9\u6cd5\uff0c\u901a\u8fc7\u5728\u4e8b\u5b9e\u6027\u8bc4\u4f30\u548c\u5956\u52b1\u5efa\u6a21\u4e24\u4e2a\u6838\u5fc3\u6b65\u9aa4\u4e2d\u5f15\u5165\u7ec6\u7c92\u5ea6\u8bbe\u8ba1\uff0c\u63d0\u4f9b\u4e86\u66f4\u5bc6\u96c6\u7684\u5956\u52b1\u4fe1\u53f7\u3002<\/p>\n

<\/p>\n

(25) Multi-Meta-RAG\u3010\u5143\u7b5b\u9009\u5668\u3011<\/h2>\n

\u5143\u7b5b\u9009\u5668<\/strong>\uff1a\u50cf\u4e2a\u7ecf\u9a8c\u4e30\u5bcc\u7684\u8d44\u6599\u7ba1\u7406\u5458\uff0c\u901a\u8fc7\u591a\u91cd\u7b5b\u9009\u673a\u5236\uff0c\u4ece\u6d77\u91cf\u4fe1\u606f\u4e2d\u7cbe\u51c6\u5b9a\u4f4d\u6700\u76f8\u5173\u7684\u5185\u5bb9\u3002\u5b83\u4e0d\u53ea\u770b\u8868\u9762\uff0c\u8fd8\u4f1a\u6df1\u5165\u5206\u6790\u6587\u6863\u7684”\u8eab\u4efd\u6807\u7b7e”\uff08\u5143\u6570\u636e\uff09\uff0c\u786e\u4fdd\u627e\u5230\u7684\u6bcf\u4efd\u8d44\u6599\u90fd\u771f\u6b63\u5bf9\u9898\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aMulti-Meta-RAG: Improving RAG for Multi-Hop Queries using Database Filtering with LLM-Extracted Metadata<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/mxpoliakov\/multi-meta-rag<\/p>\n

Multi-Meta-RAG\u4f7f\u7528\u6570\u636e\u5e93\u8fc7\u6ee4\u548cLLM\u63d0\u53d6\u7684\u5143\u6570\u636e\u6765\u6539\u8fdbRAG\u4ece\u5404\u79cd\u6765\u6e90\u4e2d\u9009\u62e9\u4e0e\u95ee\u9898\u76f8\u5173\u7684\u76f8\u5173\u6587\u6863\u3002<\/p>\n

<\/p>\n

(26) RankRAG\u3010\u5168\u80fd\u9009\u624b\u3011<\/h2>\n

\u5168\u80fd\u9009\u624b<\/strong>\uff1a\u901a\u8fc7\u4e00\u70b9\u7279\u8bad\u5c31\u80fd\u5f53\u597d”\u8bc4\u59d4”\u548c”\u9009\u624b”\u53cc\u91cd\u89d2\u8272\u3002\u50cf\u4e2a\u5929\u8d4b\u5f02\u7980\u7684\u8fd0\u52a8\u5458\uff0c\u53ea\u9700\u8981\u5c11\u91cf\u6307\u5bfc\u5c31\u80fd\u5728\u591a\u4e2a\u9879\u76ee\u4e0a\u8d85\u8d8a\u4e13\u4e1a\u9009\u624b\uff0c\u8fd8\u80fd\u628a\u770b\u5bb6\u672c\u9886\u90fd\u878d\u4f1a\u8d2f\u901a\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aRankRAG: Unifying Context Ranking with Retrieval-Augmented Generation in LLMs<\/p>\n

RankRAG\u7684\u901a\u8fc7\u6307\u4ee4\u5fae\u8c03\u5355\u4e00\u7684LLM\uff0c\u4f7f\u5176\u540c\u65f6\u5177\u5907\u4e0a\u4e0b\u6587\u6392\u540d\u548c\u7b54\u6848\u751f\u6210\u7684\u53cc\u91cd\u529f\u80fd\u3002\u901a\u8fc7\u5728\u8bad\u7ec3\u6570\u636e\u4e2d\u52a0\u5165\u5c11\u91cf\u6392\u5e8f\u6570\u636e\uff0c\u7ecf\u8fc7\u6307\u4ee4\u5fae\u8c03\u7684\u5927\u8bed\u8a00\u6a21\u578b\u6548\u679c\u51fa\u5947\u5730\u597d\uff0c\u751a\u81f3\u8d85\u8fc7\u4e86\u73b0\u6709\u7684\u4e13\u5bb6\u6392\u5e8f\u6a21\u578b\uff0c\u5305\u62ec\u5728\u5927\u91cf\u6392\u5e8f\u6570\u636e\u4e0a\u4e13\u95e8\u5fae\u8c03\u7684\u76f8\u540c\u5927\u8bed\u8a00\u6a21\u578b\u3002\u8fd9\u79cd\u8bbe\u8ba1\u4e0d\u4ec5\u7b80\u5316\u4e86\u4f20\u7edfRAG\u7cfb\u7edf\u4e2d\u591a\u6a21\u578b\u7684\u590d\u6742\u6027\uff0c\u8fd8\u901a\u8fc7\u5171\u4eab\u6a21\u578b\u53c2\u6570\u589e\u5f3a\u4e86\u4e0a\u4e0b\u6587\u7684\u76f8\u5173\u6027\u5224\u65ad\u548c\u4fe1\u606f\u7684\u5229\u7528\u6548\u7387\u3002<\/p>\n

<\/p>\n

(27) GraphRAG-Local-UI\u3010\u6539\u88c5\u5e08\u3011<\/h2>\n

\u6539\u88c5\u5e08<\/strong>\uff1a\u628a\u8dd1\u8f66\u6539\u88c5\u6210\u9002\u5408\u672c\u5730\u9053\u8def\u7684\u5b9e\u7528\u8f66\uff0c\u52a0\u88c5\u4e86\u53cb\u597d\u7684\u4eea\u8868\u76d8\uff0c\u8ba9\u4eba\u4eba\u90fd\u80fd\u8f7b\u677e\u9a7e\u9a76\u3002<\/p><\/blockquote>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/severian42\/GraphRAG-Local-UI<\/p>\n

GraphRAG-Local-UI\u662f\u57fa\u4e8eMicrosoft\u7684GraphRAG\u7684\u672c\u5730\u6a21\u578b\u9002\u914d\u7248\u672c\uff0c\u5177\u6709\u4e30\u5bcc\u7684\u4ea4\u4e92\u5f0f\u7528\u6237\u754c\u9762\u751f\u6001\u7cfb\u7edf\u3002<\/p>\n

<\/p>\n

(28) ThinkRAG\u3010\u5c0f\u79d8\u4e66\u3011<\/h2>\n

\u5c0f\u79d8\u4e66<\/strong>\uff1a\u628a\u5e9e\u5927\u7684\u77e5\u8bc6\u4f53\u7cfb\u6d53\u7f29\u6210\u53e3\u888b\u7248\uff0c\u50cf\u4e2a\u968f\u8eab\u643a\u5e26\u7684\u5c0f\u79d8\u4e66\uff0c\u4e0d\u7528\u5927\u578b\u8bbe\u5907\u5c31\u80fd\u968f\u65f6\u5e2e\u4f60\u67e5\u627e\u89e3\u7b54\u3002<\/p><\/blockquote>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/wzdavid\/ThinkRAG<\/p>\n

ThinkRAG\u5927\u6a21\u578b\u68c0\u7d22\u589e\u5f3a\u751f\u6210\u7cfb\u7edf\uff0c\u53ef\u4ee5\u8f7b\u677e\u90e8\u7f72\u5728\u7b14\u8bb0\u672c\u7535\u8111\u4e0a\uff0c\u5b9e\u73b0\u672c\u5730\u77e5\u8bc6\u5e93\u667a\u80fd\u95ee\u7b54\u3002<\/p>\n

<\/p>\n

(29) Nano-GraphRAG\u3010\u8f7b\u88c5\u4e0a\u9635\u3011<\/h2>\n

\u8f7b\u88c5\u4e0a\u9635<\/strong>\uff1a\u50cf\u4e2a\u8f7b\u88c5\u4e0a\u9635\u7684\u8fd0\u52a8\u5458\uff0c\u628a\u7e41\u590d\u7684\u88c5\u5907\u90fd\u7b80\u5316\u4e86\uff0c\u4f46\u4fdd\u7559\u4e86\u6838\u5fc3\u80fd\u529b\u3002<\/p><\/blockquote>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/gusye1234\/nano-graphrag<\/p>\n

Nano-GraphRAG\u662f\u4e00\u4e2a\u66f4\u5c0f\u3001\u66f4\u5feb\u3001\u66f4\u7b80\u6d01\u7684 GraphRAG\uff0c\u540c\u65f6\u4fdd\u7559\u4e86\u6838\u5fc3\u529f\u80fd\u3002<\/p>\n

(30) RAGFlow-GraphRAG\u3010\u5bfc\u822a\u5458\u3011<\/h2>\n

\u5bfc\u822a\u5458<\/strong>\uff1a\u5728\u95ee\u7b54\u7684\u8ff7\u5bab\u91cc\u5f00\u8f9f\u6377\u5f84\uff0c\u5148\u753b\u5f20\u5730\u56fe\u628a\u77e5\u8bc6\u70b9\u90fd\u6807\u597d\uff0c\u91cd\u590d\u7684\u8def\u6807\u5408\u5e76\u6389\uff0c\u8fd8\u7279\u5730\u7ed9\u5730\u56fe\u7626\u8eab\uff0c\u8ba9\u95ee\u8def\u7684\u4eba\u4e0d\u4f1a\u7ed5\u8fdc\u8def\u3002<\/p><\/blockquote>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/infiniflow\/ragflow<\/p>\n

RAGFlow\u501f\u9274\u4e86GraphRAG\u7684\u5b9e\u73b0\uff0c\u5728\u6587\u6863\u9884\u5904\u7406\u9636\u6bb5\uff0c\u5f15\u5165\u77e5\u8bc6\u56fe\u8c31\u6784\u5efa\u4f5c\u4e3a\u53ef\u9009\u9879\uff0c\u670d\u52a1\u4e8eQFS\u95ee\u7b54\u573a\u666f\uff0c\u5e76\u5f15\u5165\u4e86\u5b9e\u4f53\u53bb\u91cd\u3001Token\u4f18\u5316\u7b49\u6539\u8fdb\u3002<\/p>\n

<\/p>\n

(31) Medical-Graph-RAG\u3010\u6570\u5b57\u533b\u751f\u3011<\/h2>\n

\u6570\u5b57\u533b\u751f<\/strong>\uff1a\u50cf\u4e2a\u7ecf\u9a8c\u4e30\u5bcc\u7684\u533b\u5b66\u987e\u95ee\uff0c\u7528\u56fe\u8c31\u628a\u590d\u6742\u7684\u533b\u7597\u77e5\u8bc6\u6574\u7406\u5f97\u6e05\u6e05\u695a\u695a\uff0c\u8bca\u65ad\u5efa\u8bae\u4e0d\u662f\u51ed\u7a7a\u60f3\u8c61\uff0c\u800c\u662f\u6709\u7406\u6709\u636e\uff0c\u8ba9\u533b\u751f\u548c\u60a3\u8005\u90fd\u80fd\u770b\u660e\u767d\u6bcf\u4e2a\u8bca\u65ad\u80cc\u540e\u7684\u4f9d\u636e\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aMedical Graph RAG: Towards Safe Medical Large Language Model via Graph Retrieval-Augmented Generation<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/SuperMedIntel\/Medical-Graph-RAG<\/p>\n

MedGraphRAG \u662f\u4e00\u4e2a\u6846\u67b6\uff0c\u65e8\u5728\u89e3\u51b3\u5728\u533b\u5b66\u4e2d\u5e94\u7528 LLM \u7684\u6311\u6218\u3002\u5b83\u4f7f\u7528\u57fa\u4e8e\u56fe\u8c31\u7684\u65b9\u6cd5\u6765\u63d0\u9ad8\u8bca\u65ad\u51c6\u786e\u6027\u3001\u900f\u660e\u5ea6\u5e76\u96c6\u6210\u5230\u4e34\u5e8a\u5de5\u4f5c\u6d41\u7a0b\u4e2d\u3002\u8be5\u7cfb\u7edf\u901a\u8fc7\u751f\u6210\u7531\u53ef\u9760\u6765\u6e90\u652f\u6301\u7684\u54cd\u5e94\u6765\u63d0\u9ad8\u8bca\u65ad\u51c6\u786e\u6027\uff0c\u89e3\u51b3\u4e86\u5728\u5927\u91cf\u533b\u7597\u6570\u636e\u4e2d\u7ef4\u62a4\u4e0a\u4e0b\u6587\u7684\u56f0\u96be\u3002<\/p>\n

<\/p>\n

(32) HybridRAG\u3010\u4e2d\u533b\u5408\u65b9\u3011<\/h2>\n

\u4e2d\u533b\u5408\u65b9<\/strong>\uff1a\u5c31\u50cf\u4e2d\u533b\u8bb2\u7a76\u7684”\u5408\u65b9”\uff0c\u5355\u5473\u836f\u4e0d\u5982\u51e0\u5473\u836f\u5408\u5728\u4e00\u8d77\u6548\u679c\u597d\u3002\u5411\u91cf\u6570\u636e\u5e93\u8d1f\u8d23\u5feb\u901f\u68c0\u7d22\uff0c\u77e5\u8bc6\u56fe\u8c31\u8865\u5145\u5173\u7cfb\u903b\u8f91\uff0c\u4e24\u8005\u4f18\u52bf\u4e92\u8865\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aHybridRAG: Integrating Knowledge Graphs and Vector Retrieval Augmented Generation for Efficient Information Extraction<\/p>\n

\u4e00\u79cd\u57fa\u4e8e\u77e5\u8bc6\u56fe\u8c31RAG\u6280\u672f\uff08GraphRAG\uff09\u548cVectorRAG\u6280\u672f\u76f8\u7ed3\u5408\u7684\u65b0\u65b9\u6cd5\uff0c\u79f0\u4e3aHybridRAG\uff0c\u4ee5\u589e\u5f3a\u4ece\u91d1\u878d\u6587\u6863\u4e2d\u63d0\u53d6\u4fe1\u606f\u7684\u95ee\u7b54\u7cfb\u7edf\uff0c\u8be5\u65b9\u6cd5\u88ab\u8bc1\u660e\u80fd\u591f\u751f\u6210\u51c6\u786e\u4e14\u4e0e\u4e0a\u4e0b\u6587\u76f8\u5173\u7684\u7b54\u6848\u3002\u5728\u68c0\u7d22\u548c\u751f\u6210\u9636\u6bb5\uff0c\u5c31\u68c0\u7d22\u51c6\u786e\u6027\u548c\u7b54\u6848\u751f\u6210\u800c\u8a00\uff0c\u4ece\u5411\u91cf\u6570\u636e\u5e93\u548c\u77e5\u8bc6\u56fe\u8c31\u4e2d\u68c0\u7d22\u4e0a\u4e0b\u6587\u7684HybridRAG\u4f18\u4e8e\u4f20\u7edf\u7684VectorRAG\u548cGraphRAG\u3002<\/p>\n

<\/p>\n

(33) W-RAG\u3010\u8fdb\u5316\u641c\u7d22\u3011<\/h2>\n

\u8fdb\u5316\u641c\u7d22<\/strong>\uff1a\u50cf\u4e2a\u5584\u4e8e\u81ea\u6211\u8fdb\u5316\u7684\u641c\u7d22\u5f15\u64ce\uff0c\u901a\u8fc7\u5927\u6a21\u578b\u5bf9\u6587\u7ae0\u6bb5\u843d\u7684\u8bc4\u5206\u6765\u5b66\u4e60\u4ec0\u4e48\u662f\u597d\u7b54\u6848\uff0c\u9010\u6b65\u63d0\u5347\u81ea\u5df1\u627e\u5230\u5173\u952e\u4fe1\u606f\u7684\u80fd\u529b\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aW-RAG: Weakly Supervised Dense Retrieval in RAG for Open-domain Question Answering<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/jmnian\/weak_label_for_rag<\/p>\n

\u5f00\u653e\u57df\u95ee\u7b54\u4e2d\u7684\u5f31\u76d1\u7763\u5bc6\u96c6\u68c0\u7d22\u6280\u672f\uff0c\u5229\u7528\u5927\u578b\u8bed\u8a00\u6a21\u578b\u7684\u6392\u5e8f\u80fd\u529b\u4e3a\u8bad\u7ec3\u5bc6\u96c6\u68c0\u7d22\u5668\u521b\u5efa\u5f31\u6807\u6ce8\u6570\u636e\u3002\u901a\u8fc7\u8bc4\u4f30\u5927\u578b\u8bed\u8a00\u6a21\u578b\u57fa\u4e8e\u95ee\u9898\u548c\u6bcf\u4e2a\u6bb5\u843d\u751f\u6210\u6b63\u786e\u7b54\u6848\u7684\u6982\u7387\uff0c\u5bf9\u901a\u8fc7 BM25<\/a> \u68c0\u7d22\u5230\u7684\u524d K \u4e2a\u6bb5\u843d\u8fdb\u884c\u91cd\u65b0\u6392\u5e8f\u3002\u6392\u540d\u6700\u9ad8\u7684\u6bb5\u843d\u968f\u540e\u88ab\u7528\u4f5c\u5bc6\u96c6\u68c0\u7d22\u7684\u6b63\u8bad\u7ec3\u793a\u4f8b\u3002<\/p>\n

<\/p>\n

(34) RAGChecker\u3010\u8d28\u68c0\u5458\u3011<\/h2>\n

\u8d28\u68c0\u5458<\/strong>\uff1a\u4e0d\u53ea\u7b80\u5355\u5730\u5224\u65ad\u7b54\u6848\u5bf9\u9519\uff0c\u800c\u662f\u4f1a\u6df1\u5165\u68c0\u67e5\u6574\u4e2a\u56de\u7b54\u8fc7\u7a0b\u4e2d\u7684\u6bcf\u4e2a\u73af\u8282\uff0c\u4ece\u8d44\u6599\u67e5\u627e\u5230\u6700\u7ec8\u7b54\u6848\u751f\u6210\uff0c\u5c31\u50cf\u4e00\u4e2a\u4e25\u683c\u7684\u8003\u5b98\uff0c\u65e2\u7ed9\u51fa\u8be6\u7ec6\u7684\u8bc4\u5206\u62a5\u544a\uff0c\u8fd8\u4f1a\u6307\u51fa\u5177\u4f53\u54ea\u91cc\u9700\u8981\u6539\u8fdb\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aRAGChecker: A Fine-grained Framework for Diagnosing Retrieval-Augmented Generation<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/amazon-science\/RAGChecker<\/p>\n

RAGChecker \u7684\u8bca\u65ad\u5de5\u5177\u4e3a RAG \u7cfb\u7edf\u63d0\u4f9b\u7ec6\u7c92\u5ea6\u3001\u5168\u9762\u3001\u53ef\u9760\u7684\u8bca\u65ad\u62a5\u544a\uff0c\u5e76\u4e3a\u8fdb\u4e00\u6b65\u63d0\u5347\u6027\u80fd\uff0c\u63d0\u4f9b\u53ef\u64cd\u4f5c\u7684\u65b9\u5411\u3002\u5b83\u4e0d\u4ec5\u80fd\u8bc4\u4f30\u7cfb\u7edf\u7684\u6574\u4f53\u8868\u73b0\uff0c\u8fd8\u80fd\u6df1\u5165\u5206\u6790\u68c0\u7d22\u548c\u751f\u6210\u4e24\u5927\u6838\u5fc3\u6a21\u5757\u7684\u6027\u80fd\u3002<\/p>\n

<\/p>\n

(35) Meta-Knowledge-RAG\u3010\u5b66\u8005\u3011<\/h2>\n

\u5b66\u8005<\/strong>\uff1a\u50cf\u4e2a\u5b66\u672f\u754c\u7684\u8d44\u6df1\u7814\u7a76\u5458\uff0c\u4e0d\u4ec5\u6536\u96c6\u8d44\u6599\uff0c\u8fd8\u4f1a\u4e3b\u52a8\u601d\u8003\u95ee\u9898\uff0c\u4e3a\u6bcf\u4efd\u6587\u6863\u505a\u6279\u6ce8\u548c\u603b\u7ed3\uff0c\u751a\u81f3\u9884\u5148\u8bbe\u60f3\u53ef\u80fd\u7684\u95ee\u9898\u3002\u5b83\u4f1a\u628a\u76f8\u5173\u7684\u77e5\u8bc6\u70b9\u4e32\u8054\u8d77\u6765\uff0c\u5f62\u6210\u77e5\u8bc6\u7f51\u7edc\uff0c\u8ba9\u67e5\u8be2\u53d8\u5f97\u66f4\u6709\u6df1\u5ea6\u548c\u5e7f\u5ea6\uff0c\u5c31\u50cf\u6709\u4e00\u4e2a\u5b66\u8005\u5728\u5e2e\u4f60\u505a\u7814\u7a76\u7efc\u8ff0\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aMeta Knowledge for Retrieval Augmented Large Language Models<\/p>\n

Meta-Knowledge-RAG\uff08MK Summary\uff09\u5f15\u5165\u4e86\u4e00\u79cd\u65b0\u9896\u7684\u4ee5\u6570\u636e\u4e3a\u4e2d\u5fc3\u7684 RAG \u5de5\u4f5c\u6d41\u7a0b\uff0c\u5c06\u4f20\u7edf\u7684 \u201c\u68c0\u7d22-\u8bfb\u53d6\u201d \u7cfb\u7edf\u8f6c\u53d8\u4e3a\u66f4\u5148\u8fdb\u7684 \u201c\u51c6\u5907-\u91cd\u5199-\u68c0\u7d22-\u8bfb\u53d6\u201d \u6846\u67b6\uff0c\u4ee5\u5b9e\u73b0\u5bf9\u77e5\u8bc6\u5e93\u7684\u66f4\u9ad8\u9886\u57df\u4e13\u5bb6\u7ea7\u7406\u89e3\u3002\u6211\u4eec\u7684\u65b9\u6cd5\u4f9d\u8d56\u4e8e\u4e3a\u6bcf\u4e2a\u6587\u6863\u751f\u6210\u5143\u6570\u636e\u548c\u5408\u6210\u7684\u95ee\u9898\u4e0e\u7b54\u6848\u4ee5\u53ca\u4e3a\u57fa\u4e8e\u5143\u6570\u636e\u7684\u6587\u6863\u96c6\u7fa4\u5f15\u5165\u5143\u77e5\u8bc6\u6458\u8981\u7684\u65b0\u6982\u5ff5\u3002\u6240\u63d0\u51fa\u7684\u521b\u65b0\u5b9e\u73b0\u4e86\u4e2a\u6027\u5316\u7684\u7528\u6237\u67e5\u8be2\u589e\u5f3a\u548c\u8de8\u77e5\u8bc6\u5e93\u7684\u6df1\u5ea6\u4fe1\u606f\u68c0\u7d22\u3002<\/p>\n

<\/p>\n

(36) CommunityKG-RAG\u3010\u793e\u7fa4\u63a2\u7d22\u3011<\/h2>\n

\u793e\u7fa4\u63a2\u7d22<\/strong>\uff1a\u50cf\u4e2a\u719f\u6089\u793e\u533a\u5173\u7cfb\u7f51\u7edc\u7684\u5411\u5bfc\uff0c\u5584\u4e8e\u5229\u7528\u77e5\u8bc6\u95f4\u7684\u5173\u8054\u548c\u7fa4\u7ec4\u7279\u5f81\uff0c\u5728\u4e0d\u9700\u8981\u7279\u522b\u5b66\u4e60\u7684\u60c5\u51b5\u4e0b\uff0c\u5c31\u80fd\u51c6\u786e\u5730\u627e\u5230\u76f8\u5173\u4fe1\u606f\uff0c\u5e76\u9a8c\u8bc1\u5176\u53ef\u9760\u6027\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aCommunityKG-RAG: Leveraging Community Structures in Knowledge Graphs for Advanced Retrieval-Augmented Generation in Fact-Checking<\/p>\n

CommunityKG-RAG\u662f\u4e00\u79cd\u65b0\u9896\u7684\u96f6\u6837\u672c\u6846\u67b6\uff0c\u5b83\u5c06\u77e5\u8bc6\u56fe\u8c31\u4e2d\u7684\u793e\u533a\u7ed3\u6784\u4e0eRAG\u7cfb\u7edf\u76f8\u7ed3\u5408\uff0c\u4ee5\u589e\u5f3a\u4e8b\u5b9e\u6838\u67e5\u8fc7\u7a0b\u3002CommunityKG-RAG\u80fd\u591f\u5728\u65e0\u9700\u989d\u5916\u8bad\u7ec3\u7684\u60c5\u51b5\u4e0b\u9002\u5e94\u65b0\u7684\u9886\u57df\u548c\u67e5\u8be2\uff0c\u5b83\u5229\u7528\u77e5\u8bc6\u56fe\u8c31\u4e2d\u793e\u533a\u7ed3\u6784\u7684\u591a\u8df3\u6027\u8d28\uff0c\u663e\u8457\u63d0\u9ad8\u4fe1\u606f\u68c0\u7d22\u7684\u51c6\u786e\u6027\u548c\u76f8\u5173\u6027\u3002<\/p>\n

<\/p>\n

(37) TC-RAG\u3010\u8bb0\u5fc6\u672f\u58eb\u3011<\/h2>\n

\u8bb0\u5fc6\u672f\u58eb<\/strong>\uff1a\u7ed9LLM\u88c5\u4e86\u4e2a\u5e26\u81ea\u52a8\u6e05\u7406\u529f\u80fd\u7684\u5927\u8111\u3002\u5c31\u50cf\u6211\u4eec\u89e3\u9898\uff0c\u4f1a\u628a\u91cd\u8981\u6b65\u9aa4\u5199\u5728\u8349\u7a3f\u7eb8\u4e0a\uff0c\u505a\u5b8c\u5c31\u5212\u6389\u3002\u5b83\u4e0d\u662f\u6b7b\u8bb0\u786c\u80cc\uff0c\u8be5\u8bb0\u7684\u8bb0\u4f4f\uff0c\u8be5\u5fd8\u7684\u53ca\u65f6\u6e05\u7a7a\uff0c\u50cf\u4e2a\u4f1a\u6536\u62fe\u623f\u95f4\u7684\u5b66\u9738\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aTC-RAG: Turing-Complete RAG’s Case study on Medical LLM Systems<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/Artessay\/TC-RAG<\/p>\n

\u901a\u8fc7\u5f15\u5165\u56fe\u7075\u5b8c\u5907\u7684\u7cfb\u7edf\u6765\u7ba1\u7406\u72b6\u6001\u53d8\u91cf\uff0c\u4ece\u800c\u5b9e\u73b0\u66f4\u9ad8\u6548\u3001\u51c6\u786e\u7684\u77e5\u8bc6\u68c0\u7d22\u3002\u901a\u8fc7\u5229\u7528\u5177\u6709\u81ea\u9002\u5e94\u68c0\u7d22\u3001\u63a8\u7406\u548c\u89c4\u5212\u80fd\u529b\u7684\u8bb0\u5fc6\u5806\u6808\u7cfb\u7edf\uff0cTC-RAG\u4e0d\u4ec5\u786e\u4fdd\u4e86\u68c0\u7d22\u8fc7\u7a0b\u7684\u53d7\u63a7\u505c\u6b62\uff0c\u8fd8\u901a\u8fc7Push\u548cPop\u64cd\u4f5c\u51cf\u8f7b\u4e86\u9519\u8bef\u77e5\u8bc6\u7684\u79ef\u7d2f\u3002<\/p>\n

<\/p>\n

(38) RAGLAB\u3010\u7ade\u6280\u573a\u3011<\/h2>\n

\u7ade\u6280\u573a<\/strong>\uff1a\u8ba9\u5404\u79cd\u7b97\u6cd5\u53ef\u4ee5\u5728\u76f8\u540c\u7684\u89c4\u5219\u4e0b\u8fdb\u884c\u516c\u5e73\u7ade\u4e89\u548c\u6bd4\u8f83\uff0c\u5c31\u50cf\u79d1\u5b66\u5b9e\u9a8c\u5ba4\u91cc\u7684\u6807\u51c6\u5316\u6d4b\u8bd5\u6d41\u7a0b\uff0c\u786e\u4fdd\u6bcf\u4e2a\u65b0\u65b9\u6cd5\u90fd\u80fd\u5f97\u5230\u5ba2\u89c2\u900f\u660e\u7684\u8bc4\u4f30\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aRAGLAB: A Modular and Research-Oriented Unified Framework for Retrieval-Augmented Generation<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/fate-ubw\/RAGLab<\/p>\n

\u65b0\u578bRAG\u7b97\u6cd5\u4e4b\u95f4\u8d8a\u6765\u8d8a\u7f3a\u4e4f\u5168\u9762\u548c\u516c\u5e73\u7684\u6bd4\u8f83\uff0c\u5f00\u6e90\u5de5\u5177\u7684\u9ad8\u7ea7\u62bd\u8c61\u5bfc\u81f4\u7f3a\u4e4f\u900f\u660e\u5ea6\uff0c\u5e76\u9650\u5236\u4e86\u5f00\u53d1\u65b0\u7b97\u6cd5\u548c\u8bc4\u4f30\u6307\u6807\u7684\u80fd\u529b\u3002RAGLAB\u662f\u4e00\u4e2a\u6a21\u5757\u5316\u3001\u7814\u7a76\u5bfc\u5411\u7684\u5f00\u6e90\u5e93\uff0c\u91cd\u73b06\u79cd\u7b97\u6cd5\u5e76\u6784\u5efa\u5168\u9762\u7814\u7a76\u751f\u6001\u3002\u501f\u52a9RAGLAB\uff0c\u6211\u4eec\u572810\u4e2a\u57fa\u51c6\u4e0a\u516c\u5e73\u5bf9\u6bd46\u79cd\u7b97\u6cd5\uff0c\u52a9\u529b\u7814\u7a76\u4eba\u5458\u9ad8\u6548\u8bc4\u4f30\u548c\u521b\u65b0\u7b97\u6cd5\u3002<\/p>\n

<\/p>\n

(39) MemoRAG\u3010\u8fc7\u76ee\u4e0d\u5fd8\u3011<\/h2>\n

\u8fc7\u76ee\u4e0d\u5fd8<\/strong>\uff1a\u5b83\u4e0d\u53ea\u662f\u6309\u9700\u67e5\u627e\u8d44\u6599\uff0c\u800c\u662f\u5df2\u7ecf\u628a\u6574\u4e2a\u77e5\u8bc6\u5e93\u90fd\u6df1\u5165\u7406\u89e3\u5e76\u8bb0\u5728\u5fc3\u91cc\u3002\u5f53\u4f60\u95ee\u95ee\u9898\u65f6\uff0c\u5b83\u80fd\u5feb\u901f\u4ece\u8fd9\u4e2a”\u8d85\u7ea7\u5927\u8111”\u4e2d\u8c03\u53d6\u76f8\u5173\u8bb0\u5fc6\uff0c\u7ed9\u51fa\u65e2\u51c6\u786e\u53c8\u5bcc\u6709\u89c1\u5730\u7684\u7b54\u6848\uff0c\u5c31\u50cf\u4e00\u4e2a\u535a\u5b66\u591a\u8bc6\u7684\u4e13\u5bb6\u3002<\/p><\/blockquote>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/qhjqhj00\/MemoRAG<\/p>\n

MemoRAG\u662f\u4e00\u4e2a\u521b\u65b0\u7684\u68c0\u7d22\u589e\u5f3a\u751f\u6210\uff08RAG\uff09\u6846\u67b6\uff0c\u6784\u5efa\u5728\u4e00\u4e2a\u9ad8\u6548\u7684\u8d85\u957f\u8bb0\u5fc6\u6a21\u578b\u4e4b\u4e0a\u3002\u4e0e\u4e3b\u8981\u5904\u7406\u5177\u6709\u660e\u786e\u4fe1\u606f\u9700\u6c42\u67e5\u8be2\u7684\u6807\u51c6RAG\u4e0d\u540c\uff0cMemoRAG\u5229\u7528\u5176\u8bb0\u5fc6\u6a21\u578b\u5b9e\u73b0\u5bf9\u6574\u4e2a\u6570\u636e\u5e93\u7684\u5168\u5c40\u7406\u89e3\u3002\u901a\u8fc7\u4ece\u8bb0\u5fc6\u4e2d\u53ec\u56de\u7279\u5b9a\u4e8e\u67e5\u8be2\u7684\u7ebf\u7d22\uff0cMemoRAG\u589e\u5f3a\u4e86\u8bc1\u636e\u68c0\u7d22\uff0c\u4ece\u800c\u4ea7\u751f\u66f4\u51c6\u786e\u4e14\u5177\u6709\u4e30\u5bcc\u4e0a\u4e0b\u6587\u7684\u54cd\u5e94\u751f\u6210\u3002<\/p>\n

<\/p>\n

(40) OP-RAG\u3010\u6ce8\u610f\u529b\u7ba1\u7406\u3011<\/h2>\n

\u6ce8\u610f\u529b\u7ba1\u7406<\/strong>\uff1a\u5c31\u50cf\u770b\u4e00\u672c\u7279\u522b\u539a\u7684\u4e66\uff0c\u4f60\u4e0d\u53ef\u80fd\u628a\u6bcf\u4e2a\u7ec6\u8282\u90fd\u8bb0\u4f4f\uff0c\u4f46\u61c2\u5f97\u5728\u5173\u952e\u7ae0\u8282\u505a\u597d\u6807\u8bb0\u7684\u4eba\u624d\u662f\u9ad8\u624b\u3002\u5b83\u4e0d\u662f\u6f2b\u65e0\u76ee\u7684\u5730\u770b\uff0c\u800c\u662f\u50cf\u4e2a\u8d44\u6df1\u8bfb\u4e66\u4eba\uff0c\u8fb9\u8bfb\u8fb9\u5728\u91cd\u70b9\u5904\u753b\u4e0b\u91cd\u70b9\uff0c\u9700\u8981\u7684\u65f6\u5019\u76f4\u63a5\u7ffb\u5230\u6807\u8bb0\u9875\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aIn Defense of RAG in the Era of Long-Context Language Models<\/p>\n

LLM\u4e2d\u7684\u6781\u957f\u8bed\u5883\u4f1a\u5bfc\u81f4\u5bf9\u76f8\u5173\u4fe1\u606f\u7684\u5173\u6ce8\u5ea6\u964d\u4f4e\uff0c\u5e76\u5bfc\u81f4\u7b54\u6848\u8d28\u91cf\u7684\u6f5c\u5728\u4e0b\u964d\u3002\u91cd\u65b0\u5ba1\u89c6\u957f\u4e0a\u4e0b\u6587\u7b54\u6848\u751f\u6210\u4e2d\u7684RAG\u3002\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u987a\u5e8f\u4fdd\u7559\u68c0\u7d22\u589e\u5f3a\u751f\u6210\u673a\u5236OP-RAG\uff0c\u663e\u8457\u63d0\u9ad8\u4e86RAG\u5728\u957f\u4e0a\u4e0b\u6587\u95ee\u7b54\u5e94\u7528\u4e2d\u7684\u6027\u80fd\u3002<\/p>\n

<\/p>\n

(41) AgentRE\u3010\u667a\u80fd\u62bd\u53d6\u3011<\/h2>\n

\u667a\u80fd\u62bd\u53d6<\/strong>\uff1a\u50cf\u4e2a\u5584\u4e8e\u89c2\u5bdf\u4eba\u9645\u5173\u7cfb\u7684\u793e\u4f1a\u5b66\u5bb6\uff0c\u4e0d\u4ec5\u80fd\u8bb0\u4f4f\u5173\u952e\u4fe1\u606f\uff0c\u8fd8\u4f1a\u4e3b\u52a8\u67e5\u8bc1\u5e76\u6df1\u5165\u601d\u8003\uff0c\u4ece\u800c\u51c6\u786e\u7406\u89e3\u590d\u6742\u7684\u5173\u7cfb\u7f51\u7edc\u3002\u5373\u4f7f\u9762\u5bf9\u9519\u7efc\u590d\u6742\u7684\u5173\u7cfb\uff0c\u4e5f\u80fd\u901a\u8fc7\u591a\u89d2\u5ea6\u5206\u6790\uff0c\u7406\u6e05\u5176\u4e2d\u7684\u8109\u7edc\uff0c\u907f\u514d\u671b\u6587\u751f\u4e49\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aAgentRE: An Agent-Based Framework for Navigating Complex Information Landscapes in Relation Extraction<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/Lightblues\/AgentRE<\/p>\n

AgentRE\u901a\u8fc7\u6574\u5408\u5927\u578b\u8bed\u8a00\u6a21\u578b\u7684\u8bb0\u5fc6\u3001\u68c0\u7d22\u548c\u53cd\u601d\u80fd\u529b\uff0c\u6709\u6548\u5e94\u5bf9\u590d\u6742\u573a\u666f\u5173\u7cfb\u62bd\u53d6\u4e2d\u5173\u7cfb\u7c7b\u578b\u591a\u6837\u4ee5\u53ca\u5355\u4e2a\u53e5\u5b50\u4e2d\u5b9e\u4f53\u4e4b\u95f4\u5173\u7cfb\u6a21\u7cca\u7684\u6311\u6218\u3002AgentRE \u5305\u542b\u4e09\u5927\u6a21\u5757\uff0c\u52a9\u529b\u4ee3\u7406\u9ad8\u6548\u83b7\u53d6\u5e76\u5904\u7406\u4fe1\u606f\uff0c\u663e\u8457\u63d0\u5347 RE \u6027\u80fd\u3002<\/p>\n

<\/p>\n

(42) iText2KG\u3010\u5efa\u7b51\u5e08\u3011<\/h2>\n

\u5efa\u7b51\u5e08<\/strong>\uff1a\u50cf\u4e2a\u6709\u6761\u7406\u7684\u5de5\u7a0b\u5e08\uff0c\u901a\u8fc7\u5206\u6b65\u9aa4\u63d0\u70bc\u3001\u63d0\u53d6\u548c\u6574\u5408\u4fe1\u606f\uff0c\u9010\u6b65\u5c06\u96f6\u6563\u6587\u6863\u8f6c\u5316\u4e3a\u7cfb\u7edf\u7684\u77e5\u8bc6\u7f51\u7edc\uff0c\u800c\u4e14\u4e0d\u9700\u8981\u4e8b\u5148\u51c6\u5907\u8be6\u7ec6\u7684\u5efa\u7b51\u56fe\u7eb8\uff0c\u53ef\u4ee5\u6839\u636e\u9700\u8981\u7075\u6d3b\u5730\u6269\u5efa\u548c\u5b8c\u5584\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aiText2KG: Incremental Knowledge Graphs Construction Using Large Language Models<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/AuvaLab\/itext2kg<\/p>\n

iText2KG\uff08\u589e\u91cf\u5f0f\u77e5\u8bc6\u56fe\u8c31\u6784\u5efa\uff09\u5229\u7528\u5927\u578b\u8bed\u8a00\u6a21\u578b (LLM) \u4ece\u539f\u59cb\u6587\u6863\u4e2d\u6784\u5efa\u77e5\u8bc6\u56fe\u8c31\uff0c\u5e76\u901a\u8fc7\u56db\u4e2a\u6a21\u5757\uff08\u6587\u6863\u63d0\u70bc\u5668\u3001\u589e\u91cf\u5b9e\u4f53\u63d0\u53d6\u5668\u3001\u589e\u91cf\u5173\u7cfb\u63d0\u53d6\u5668\u548c\u56fe\u8c31\u96c6\u6210\u5668\uff09\u5b9e\u73b0\u589e\u91cf\u5f0f\u77e5\u8bc6\u56fe\u8c31\u6784\u5efa\uff0c\u65e0\u9700\u4e8b\u5148\u5b9a\u4e49\u672c\u4f53\u6216\u8fdb\u884c\u5927\u91cf\u7684\u76d1\u7763\u8bad\u7ec3\u3002<\/p>\n

<\/p>\n

(43) GraphInsight\u3010\u56fe\u8c31\u89e3\u8bfb\u3011<\/h2>\n

\u56fe\u8c31\u89e3\u8bfb<\/strong>\uff1a\u50cf\u4e2a\u64c5\u957f\u4fe1\u606f\u56fe\u8868\u5206\u6790\u7684\u4e13\u5bb6\uff0c\u77e5\u9053\u628a\u91cd\u8981\u4fe1\u606f\u653e\u5728\u6700\u663e\u773c\u7684\u4f4d\u7f6e\uff0c\u540c\u65f6\u5728\u9700\u8981\u65f6\u67e5\u9605\u53c2\u8003\u8d44\u6599\u6765\u8865\u5145\u7ec6\u8282\uff0c\u5e76\u80fdstep by step\u5730\u63a8\u7406\u590d\u6742\u56fe\u8868\uff0c\u8ba9AI\u65e2\u80fd\u628a\u63e1\u5168\u5c40\u53c8\u4e0d\u9057\u6f0f\u7ec6\u8282\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aGraphInsight: Unlocking Insights in Large Language Models for Graph Structure Understanding<\/p>\n

GraphInsight\u65e8\u5728\u63d0\u5347LLMs\u5bf9\u5b8f\u89c2\u548c\u5fae\u89c2\u5c42\u9762\u56fe\u5f62\u4fe1\u606f\u7406\u89e3\u7684\u65b0\u6846\u67b6\u3002GraphInsight\u57fa\u4e8e\u4e24\u5927\u5173\u952e\u7b56\u7565\uff1a1\uff09\u5c06\u5173\u952e\u56fe\u5f62\u4fe1\u606f\u7f6e\u4e8eLLMs\u8bb0\u5fc6\u6027\u80fd\u8f83\u5f3a\u7684\u4f4d\u7f6e\uff1b2\uff09\u501f\u9274\u68c0\u7d22\u589e\u5f3a\u751f\u6210\uff08RAG\uff09\u7684\u601d\u60f3\uff0c\u5bf9\u8bb0\u5fc6\u6027\u80fd\u8f83\u5f31\u7684\u533a\u57df\u5f15\u5165\u8f7b\u91cf\u7ea7\u5916\u90e8\u77e5\u8bc6\u5e93\u3002\u6b64\u5916\uff0cGraphInsight\u63a2\u7d22\u5c06\u8fd9\u4e24\u79cd\u7b56\u7565\u6574\u5408\u5230LLM\u4ee3\u7406\u8fc7\u7a0b\u4e2d\uff0c\u4ee5\u5e94\u5bf9\u9700\u8981\u591a\u6b65\u63a8\u7406\u7684\u590d\u5408\u56fe\u4efb\u52a1\u3002<\/p>\n

<\/p>\n

(44) LA-RAG\u3010\u65b9\u8a00\u901a\u3011<\/h2>\n

\u65b9\u8a00\u901a<\/strong>\uff1a\u50cf\u4e2a\u7cbe\u901a\u5404\u5730\u65b9\u8a00\u7684\u8bed\u8a00\u4e13\u5bb6\uff0c\u901a\u8fc7\u7ec6\u81f4\u7684\u8bed\u97f3\u5206\u6790\u548c\u4e0a\u4e0b\u6587\u7406\u89e3\uff0c\u4e0d\u4ec5\u80fd\u51c6\u786e\u8bc6\u522b\u6807\u51c6\u666e\u901a\u8bdd\uff0c\u8fd8\u80fd\u542c\u61c2\u5e26\u6709\u5730\u65b9\u7279\u8272\u7684\u53e3\u97f3\uff0c\u8ba9AI\u4e0e\u4e0d\u540c\u5730\u533a\u7684\u4eba\u90fd\u80fd\u65e0\u969c\u788d\u4ea4\u6d41\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aLA-RAG:Enhancing LLM-based ASR Accuracy with Retrieval-Augmented Generation<\/p>\n

LA-RAG\uff0c\u662f\u4e00\u79cd\u57fa\u4e8eLLM\u7684ASR\u7684\u65b0\u578b\u68c0\u7d22\u589e\u5f3a\u751f\u6210\uff08RAG\uff09\u8303\u4f8b\u3002LA-RAG \u5229\u7528\u7ec6\u7c92\u5ea6\u6807\u8bb0\u7ea7\u8bed\u97f3\u6570\u636e\u5b58\u50a8\u548c\u8bed\u97f3\u5230\u8bed\u97f3\u68c0\u7d22\u673a\u5236\uff0c\u901a\u8fc7 LLM \u4e0a\u4e0b\u6587\u5b66\u4e60 (ICL) \u529f\u80fd\u63d0\u9ad8 ASR \u51c6\u786e\u6027\u3002<\/p>\n

<\/p>\n

(45) SFR-RAG\u3010\u7cbe\u7b80\u68c0\u7d22\u3011<\/h2>\n

\u7cbe\u7b80\u68c0\u7d22<\/strong>\uff1a\u50cf\u4e2a\u7cbe\u7ec3\u7684\u53c2\u8003\u987e\u95ee\uff0c\u4f53\u79ef\u867d\u5c0f\u4f46\u529f\u80fd\u7cbe\u51c6\uff0c\u65e2\u80fd\u7406\u89e3\u9700\u6c42\u53c8\u61c2\u5f97\u5bfb\u6c42\u5916\u90e8\u5e2e\u52a9\uff0c\u4fdd\u8bc1\u56de\u7b54\u65e2\u51c6\u786e\u53c8\u9ad8\u6548\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aSFR-RAG: Towards Contextually Faithful LLMs<\/p>\n

SFR-RAG\u662f\u4e00\u4e2a\u7ecf\u8fc7\u6307\u4ee4\u5fae\u8c03\u7684\u5c0f\u578b\u8bed\u8a00\u6a21\u578b\uff0c\u91cd\u70b9\u662f\u57fa\u4e8e\u4e0a\u4e0b\u6587\u7684\u751f\u6210\u548c\u6700\u5c0f\u5316\u5e7b\u89c9\u3002\u901a\u8fc7\u4e13\u6ce8\u4e8e\u5728\u4fdd\u6301\u9ad8\u6027\u80fd\u7684\u540c\u65f6\u51cf\u5c11\u53c2\u6570\u6570\u91cf\uff0cSFR-RAG\u6a21\u578b\u5305\u542b\u51fd\u6570\u8c03\u7528\u529f\u80fd\uff0c\u4f7f\u5176\u80fd\u591f\u4e0e\u5916\u90e8\u5de5\u5177\u52a8\u6001\u4ea4\u4e92\u4ee5\u68c0\u7d22\u9ad8\u8d28\u91cf\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\u3002<\/p>\n

<\/p>\n

(46) FlexRAG\u3010\u538b\u7f29\u4e13\u5bb6\u3011<\/h2>\n

\u538b\u7f29\u4e13\u5bb6<\/strong>\uff1a\u628a\u957f\u7bc7\u5927\u8bba\u6d53\u7f29\u6210\u7cbe\u534e\u6458\u8981\uff0c\u800c\u4e14\u538b\u7f29\u6bd4\u4f8b\u53ef\u4ee5\u6839\u636e\u9700\u8981\u7075\u6d3b\u8c03\u6574\uff0c\u65e2\u4e0d\u4e22\u5931\u5173\u952e\u4fe1\u606f\uff0c\u53c8\u80fd\u8282\u7701\u5b58\u50a8\u548c\u5904\u7406\u6210\u672c\u3002\u5c31\u50cf\u628a\u4e00\u672c\u539a\u4e66\u7cbe\u70bc\u6210\u4e00\u4efd\u7b80\u660e\u627c\u8981\u7684\u8bfb\u4e66\u7b14\u8bb0\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aLighter And Better: Towards Flexible Context Adaptation For Retrieval Augmented Generation<\/p>\n

FlexRAG\u68c0\u7d22\u5230\u7684\u4e0a\u4e0b\u6587\u5728\u88abLLMs\u7f16\u7801\u4e4b\u524d\u88ab\u538b\u7f29\u4e3a\u7d27\u51d1\u7684\u5d4c\u5165\u3002\u540c\u65f6\u8fd9\u4e9b\u538b\u7f29\u540e\u7684\u5d4c\u5165\u7ecf\u8fc7\u4f18\u5316\u4ee5\u63d0\u5347\u4e0b\u6e38RAG\u7684\u6027\u80fd\u3002FlexRAG\u7684\u4e00\u4e2a\u5173\u952e\u7279\u6027\u662f\u5176\u7075\u6d3b\u6027\uff0c\u5b83\u80fd\u591f\u6709\u6548\u652f\u6301\u4e0d\u540c\u7684\u538b\u7f29\u6bd4\uff0c\u5e76\u9009\u62e9\u6027\u5730\u4fdd\u7559\u91cd\u8981\u4e0a\u4e0b\u6587\u3002\u5f97\u76ca\u4e8e\u8fd9\u4e9b\u6280\u672f\u8bbe\u8ba1\uff0cFlexRAG\u5728\u663e\u8457\u964d\u4f4e\u8fd0\u884c\u6210\u672c\u7684\u540c\u65f6\u5b9e\u73b0\u4e86\u5353\u8d8a\u7684\u751f\u6210\u8d28\u91cf\u3002\u5728\u5404\u79cd\u95ee\u7b54\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u7684\u5168\u9762\u5b9e\u9a8c\u9a8c\u8bc1\u4e86\u6211\u4eec\u7684\u65b9\u6cd5\u662fRAG\u7cfb\u7edf\u7684\u4e00\u79cd\u5177\u6709\u6210\u672c\u6548\u76ca\u4e14\u7075\u6d3b\u7684\u89e3\u51b3\u65b9\u6848\u3002<\/p>\n

<\/p>\n

(47) CoTKR\u3010\u56fe\u8c31\u7ffb\u8bd1\u3011<\/h2>\n

\u56fe\u8c31\u7ffb\u8bd1<\/strong>\uff1a\u50cf\u4e2a\u8010\u5fc3\u7684\u8001\u5e08\uff0c\u5148\u7406\u89e3\u77e5\u8bc6\u7684\u6765\u9f99\u53bb\u8109\uff0c\u518d\u4e00\u6b65\u6b65\u8bb2\u89e3\uff0c\u4e0d\u662f\u7b80\u5355\u590d\u8ff0\u800c\u662f\u6df1\u5165\u6d45\u51fa\u5730\u8f6c\u8ff0\u3002\u540c\u65f6\u901a\u8fc7\u4e0d\u65ad\u6536\u96c6”\u5b66\u751f”\u7684\u53cd\u9988\u6765\u6539\u8fdb\u81ea\u5df1\u7684\u8bb2\u89e3\u65b9\u5f0f\uff0c\u8ba9\u77e5\u8bc6\u4f20\u9012\u66f4\u52a0\u6e05\u6670\u6709\u6548\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aCoTKR: Chain-of-Thought Enhanced Knowledge Rewriting for Complex Knowledge Graph Question Answering<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/wuyike2000\/CoTKR<\/p>\n

CoTKR\uff08Chain-of-Thought Enhanced Knowledge Rewriting\uff09\u65b9\u6cd5\u4ea4\u66ff\u751f\u6210\u63a8\u7406\u8def\u5f84\u548c\u76f8\u5e94\u77e5\u8bc6\uff0c\u4ece\u800c\u514b\u670d\u4e86\u5355\u6b65\u77e5\u8bc6\u6539\u5199\u7684\u9650\u5236\u3002\u6b64\u5916\uff0c\u4e3a\u4e86\u5f25\u5408\u77e5\u8bc6\u6539\u5199\u5668\u548c\u95ee\u7b54\u6a21\u578b\u4e4b\u95f4\u7684\u504f\u597d\u5dee\u5f02\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u8bad\u7ec3\u7b56\u7565\uff0c\u5373\u4ece\u95ee\u7b54\u53cd\u9988\u4e2d\u5bf9\u9f50\u504f\u597d\u901a\u8fc7\u5229\u7528QA\u6a21\u578b\u7684\u53cd\u9988\u8fdb\u4e00\u6b65\u4f18\u5316\u77e5\u8bc6\u6539\u5199\u5668\u3002<\/p>\n

<\/p>\n

(48) Open-RAG\u3010\u667a\u56ca\u56e2\u3011<\/h2>\n

\u667a\u56ca\u56e2<\/strong>\uff1a\u628a\u5e9e\u5927\u7684\u8bed\u8a00\u6a21\u578b\u5206\u89e3\u6210\u4e13\u5bb6\u5c0f\u7ec4\uff0c\u8ba9\u5b83\u4eec\u65e2\u80fd\u72ec\u7acb\u601d\u8003\u53c8\u80fd\u534f\u540c\u5de5\u4f5c\uff0c\u8fd8\u7279\u522b\u4f1a\u5206\u8fa8\u771f\u5047\u4fe1\u606f\uff0c\u5173\u952e\u65f6\u523b\u77e5\u9053\u8be5\u4e0d\u8be5\u67e5\u8d44\u6599\uff0c\u50cf\u4e2a\u7ecf\u9a8c\u4e30\u5bcc\u7684\u667a\u56ca\u56e2\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aOpen-RAG: Enhanced Retrieval-Augmented Reasoning with Open-Source Large Language Models<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/ShayekhBinIslam\/openrag<\/p>\n

Open-RAG\u901a\u8fc7\u5f00\u6e90\u5927\u8bed\u8a00\u6a21\u578b\u63d0\u9ad8RAG\u4e2d\u7684\u63a8\u7406\u80fd\u529b\uff0c\u5c06\u4efb\u610f\u5bc6\u96c6\u7684\u5927\u8bed\u8a00\u6a21\u578b\u8f6c\u6362\u4e3a\u53c2\u6570\u9ad8\u6548\u7684\u7a00\u758f\u4e13\u5bb6\u6df7\u5408\uff08MoE\uff09\u6a21\u578b\uff0c\u8be5\u6a21\u578b\u80fd\u591f\u5904\u7406\u590d\u6742\u7684\u63a8\u7406\u4efb\u52a1\uff0c\u5305\u62ec\u5355\u8df3\u548c\u591a\u8df3\u67e5\u8be2\u3002OPEN-RAG\u72ec\u7279\u5730\u8bad\u7ec3\u6a21\u578b\u4ee5\u5e94\u5bf9\u90a3\u4e9b\u770b\u4f3c\u76f8\u5173\u4f46\u5177\u6709\u8bef\u5bfc\u6027\u7684\u6311\u6218\u6027\u5e72\u6270\u9879\u3002<\/p>\n

<\/p>\n

(49) TableRAG\u3010Excel\u4e13\u5bb6\u3011<\/h2>\n

Excel\u4e13\u5bb6<\/strong>\uff1a\u4e0d\u53ea\u7b80\u5355\u5730\u67e5\u770b\u8868\u683c\u6570\u636e\uff0c\u800c\u662f\u61c2\u5f97\u4ece\u8868\u5934\u548c\u5355\u5143\u683c\u4e24\u4e2a\u7ef4\u5ea6\u53bb\u7406\u89e3\u548c\u68c0\u7d22\u6570\u636e\uff0c\u5c31\u50cf\u719f\u7ec3\u4f7f\u7528\u6570\u636e\u900f\u89c6\u8868\u4e00\u6837\uff0c\u80fd\u5feb\u901f\u5b9a\u4f4d\u548c\u63d0\u53d6\u6240\u9700\u7684\u5173\u952e\u4fe1\u606f\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aTableRAG: Million-Token Table Understanding with Language Models<\/p>\n

TableRAG\u4e13\u4e3a\u8868\u683c\u7406\u89e3\u8bbe\u8ba1\u4e86\u68c0\u7d22\u589e\u5f3a\u751f\u6210\u6846\u67b6\uff0c\u901a\u8fc7\u67e5\u8be2\u6269\u5c55\u7ed3\u5408Schema\u548c\u5355\u5143\u683c\u68c0\u7d22\uff0c\u80fd\u591f\u5728\u63d0\u4f9b\u4fe1\u606f\u7ed9\u8bed\u8a00\u6a21\u578b\u4e4b\u524d\u7cbe\u51c6\u5b9a\u4f4d\u5173\u952e\u6570\u636e\uff0c\u4ece\u800c\u5b9e\u73b0\u66f4\u9ad8\u6548\u7684\u6570\u636e\u7f16\u7801\u548c\u7cbe\u786e\u68c0\u7d22\uff0c\u5927\u5e45\u7f29\u77ed\u63d0\u793a\u957f\u5ea6\u5e76\u51cf\u5c11\u4fe1\u606f\u4e22\u5931\u3002<\/p>\n

<\/p>\n

(50) LightRAG\u3010\u8718\u86db\u4fa0\u3011<\/h2>\n

\u8718\u86db\u4fa0<\/strong>\uff1a\u5728\u77e5\u8bc6\u7684\u7f51\u4e2d\u7075\u6d3b\u7a7f\u68ad\uff0c\u65e2\u80fd\u6293\u4f4f\u77e5\u8bc6\u70b9\u4e4b\u95f4\u7684\u4e1d\uff0c\u53c8\u80fd\u501f\u7f51\u987a\u85e4\u6478\u74dc\u3002\u50cf\u4e2a\u957f\u4e86\u5343\u91cc\u773c\u7684\u56fe\u4e66\u7ba1\u7406\u5458\uff0c\u4e0d\u4ec5\u77e5\u9053\u6bcf\u672c\u4e66\u5728\u54ea\uff0c\u8fd8\u77e5\u9053\u54ea\u4e9b\u4e66\u8be5\u4e00\u8d77\u770b\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aLightRAG: Simple and Fast Retrieval-Augmented Generation<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/HKUDS\/LightRAG<\/p>\n

\u8be5\u6846\u67b6\u5c06\u56fe\u7ed3\u6784\u878d\u5165\u6587\u672c\u7d22\u5f15\u548c\u68c0\u7d22\u8fc7\u7a0b\u4e2d\u3002\u8fd9\u4e00\u521b\u65b0\u6846\u67b6\u91c7\u7528\u4e86\u4e00\u4e2a\u53cc\u5c42\u68c0\u7d22\u7cfb\u7edf\uff0c\u4ece\u4f4e\u7ea7\u548c\u9ad8\u7ea7\u77e5\u8bc6\u53d1\u73b0\u4e2d\u589e\u5f3a\u5168\u9762\u7684\u4fe1\u606f\u68c0\u7d22\u3002\u6b64\u5916\uff0c\u5c06\u56fe\u7ed3\u6784\u4e0e\u5411\u91cf\u8868\u793a\u76f8\u7ed3\u5408\uff0c\u4fbf\u4e8e\u9ad8\u6548\u68c0\u7d22\u76f8\u5173\u5b9e\u4f53\u53ca\u5176\u5173\u7cfb\uff0c\u663e\u8457\u63d0\u9ad8\u4e86\u54cd\u5e94\u65f6\u95f4\uff0c\u540c\u65f6\u4fdd\u6301\u4e86\u4e0a\u4e0b\u6587\u76f8\u5173\u6027\u3002\u8fd9\u4e00\u80fd\u529b\u901a\u8fc7\u589e\u91cf\u66f4\u65b0\u7b97\u6cd5\u5f97\u5230\u4e86\u8fdb\u4e00\u6b65\u589e\u5f3a\uff0c\u8be5\u7b97\u6cd5\u786e\u4fdd\u4e86\u65b0\u6570\u636e\u7684\u53ca\u65f6\u6574\u5408\uff0c\u4f7f\u7cfb\u7edf\u80fd\u591f\u5728\u5feb\u901f\u53d8\u5316\u7684\u6570\u636e\u73af\u5883\u4e2d\u4fdd\u6301\u6709\u6548\u6027\u548c\u54cd\u5e94\u6027\u3002<\/p>\n

<\/p>\n

(51) AstuteRAG\u3010\u660e\u667a\u5224\u5b98\u3011<\/h2>\n

\u660e\u667a\u5224\u5b98<\/strong>\uff1a\u5bf9\u5916\u90e8\u4fe1\u606f\u4fdd\u6301\u8b66\u60d5\uff0c\u4e0d\u8f7b\u4fe1\u68c0\u7d22\u7ed3\u679c\uff0c\u5584\u7528\u81ea\u8eab\u79ef\u7d2f\u7684\u77e5\u8bc6\uff0c\u7504\u522b\u4fe1\u606f\u771f\u4f2a\uff0c\u50cf\u8d44\u6df1\u6cd5\u5b98\u4e00\u6837\uff0c\u6743\u8861\u591a\u65b9\u8bc1\u636e\u5b9a\u8bba\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aAstute RAG: Overcoming Imperfect Retrieval Augmentation and Knowledge Conflicts for Large Language Models<\/p>\n

\u901a\u8fc7\u9002\u5e94\u6027\u5730\u4eceLLMs\u5185\u90e8\u77e5\u8bc6\u4e2d\u63d0\u53d6\u4fe1\u606f\uff0c\u7ed3\u5408\u5916\u90e8\u68c0\u7d22\u7ed3\u679c\uff0c\u5e76\u6839\u636e\u4fe1\u606f\u7684\u53ef\u9760\u6027\u6765\u6700\u7ec8\u786e\u5b9a\u7b54\u6848\uff0c\u4ece\u800c\u63d0\u9ad8\u7cfb\u7edf\u7684\u9c81\u68d2\u6027\u548c\u53ef\u4fe1\u5ea6\u3002<\/p>\n

<\/p>\n

(52) TurboRAG\u3010\u901f\u8bb0\u9ad8\u624b\u3011<\/h2>\n

\u901f\u8bb0\u9ad8\u624b<\/strong>\uff1a\u63d0\u524d\u628a\u529f\u8bfe\u505a\u597d\uff0c\u628a\u7b54\u6848\u90fd\u8bb0\u5728\u5c0f\u672c\u672c\u91cc\u3002\u50cf\u4e2a\u8003\u524d\u7a81\u51fb\u7684\u5b66\u9738\uff0c\u4e0d\u662f\u4e34\u573a\u62b1\u4f5b\u811a\uff0c\u800c\u662f\u628a\u5e38\u8003\u9898\u63d0\u524d\u6574\u7406\u6210\u9519\u9898\u672c\u3002\u9700\u8981\u7684\u65f6\u5019\u76f4\u63a5\u7ffb\u51fa\u6765\u7528\uff0c\u7701\u5f97\u6bcf\u6b21\u90fd\u8981\u73b0\u573a\u63a8\u5bfc\u4e00\u904d\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aTurboRAG: Accelerating Retrieval-Augmented Generation with Precomputed KV Caches for Chunked Text<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/MooreThreads\/TurboRAG<\/p>\n

TurboRAG\u901a\u8fc7\u79bb\u7ebf\u9884\u8ba1\u7b97\u548c\u5b58\u50a8\u6587\u6863\u7684KV\u7f13\u5b58\u6765\u4f18\u5316RAG\u7cfb\u7edf\u7684\u63a8\u7406\u8303\u5f0f\u3002\u4e0e\u4f20\u7edf\u65b9\u6cd5\u4e0d\u540c\uff0cTurboRAG\u5728\u6bcf\u6b21\u63a8\u7406\u65f6\u4e0d\u518d\u8ba1\u7b97\u8fd9\u4e9bKV\u7f13\u5b58\uff0c\u800c\u662f\u68c0\u7d22\u9884\u5148\u8ba1\u7b97\u7684\u7f13\u5b58\u4ee5\u8fdb\u884c\u9ad8\u6548\u7684\u9884\u586b\u5145\uff0c\u4ece\u800c\u6d88\u9664\u4e86\u91cd\u590d\u5728\u7ebf\u8ba1\u7b97\u7684\u9700\u8981\u3002\u8fd9\u79cd\u65b9\u6cd5\u663e\u8457\u51cf\u5c11\u4e86\u8ba1\u7b97\u5f00\u9500\uff0c\u52a0\u5feb\u4e86\u54cd\u5e94\u65f6\u95f4\uff0c\u540c\u65f6\u4fdd\u6301\u4e86\u51c6\u786e\u6027\u3002<\/p>\n

<\/p>\n

(53) StructRAG\u3010\u6536\u7eb3\u5e08\u3011<\/h2>\n

\u6536\u7eb3\u5e08<\/strong>\uff1a\u628a\u6742\u4e71\u65e0\u7ae0\u7684\u4fe1\u606f\u50cf\u6536\u7eb3\u8863\u67dc\u4e00\u6837\u5206\u95e8\u522b\u7c7b\u5730\u6574\u7406\u597d\u3002\u50cf\u4e2a\u6a21\u4eff\u4eba\u7c7b\u601d\u7ef4\u7684\u5b66\u9738\uff0c\u4e0d\u662f\u6b7b\u8bb0\u786c\u80cc\uff0c\u800c\u662f\u5148\u753b\u4e2a\u601d\u7ef4\u5bfc\u56fe\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aStructRAG: Boosting Knowledge Intensive Reasoning of LLMs via Inference-time Hybrid Information Structurization<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/Li-Z-Q\/StructRAG<\/p>\n

\u53d7\u4eba\u7c7b\u5728\u5904\u7406\u77e5\u8bc6\u5bc6\u96c6\u578b\u63a8\u7406\u65f6\u5c06\u539f\u59cb\u4fe1\u606f\u8f6c\u6362\u4e3a\u7ed3\u6784\u5316\u77e5\u8bc6\u7684\u8ba4\u77e5\u7406\u8bba\u542f\u53d1\uff0c\u8be5\u6846\u67b6\u5f15\u5165\u4e86\u4e00\u79cd\u6df7\u5408\u4fe1\u606f\u7ed3\u6784\u5316\u673a\u5236\uff0c\u8be5\u673a\u5236\u6839\u636e\u624b\u5934\u4efb\u52a1\u7684\u7279\u5b9a\u8981\u6c42\u4ee5\u6700\u5408\u9002\u7684\u683c\u5f0f\u6784\u5efa\u548c\u5229\u7528\u7ed3\u6784\u5316\u77e5\u8bc6\u3002\u901a\u8fc7\u6a21\u4eff\u7c7b\u4eba\u7684\u601d\u7ef4\u8fc7\u7a0b\uff0c\u63d0\u9ad8\u4e86LLM\u5728\u77e5\u8bc6\u5bc6\u96c6\u578b\u63a8\u7406\u4efb\u52a1\u4e0a\u7684\u8868\u73b0\u3002<\/p>\n

<\/p>\n

(54) VisRAG\u3010\u706b\u773c\u91d1\u775b\u3011<\/h2>\n

\u706b\u773c\u91d1\u775b<\/strong>\uff1a\u7ec8\u4e8e\u609f\u51fa\u6587\u5b57\u4e0d\u8fc7\u662f\u56fe\u50cf\u7684\u4e00\u79cd\u7279\u6b8a\u8868\u73b0\u5f62\u5f0f\u3002\u50cf\u4e2a\u5f00\u4e86\u5929\u773c\u7684\u9605\u8bfb\u8005\uff0c\u4e0d\u518d\u6267\u7740\u4e8e\u9010\u5b57\u89e3\u6790\uff0c\u800c\u662f\u76f4\u63a5”\u770b”\u900f\u5168\u5c40\u3002\u7528\u7167\u76f8\u673a\u4ee3\u66ff\u4e86OCR\uff0c\u61c2\u5f97\u4e86”\u4e00\u56fe\u80dc\u5343\u8a00”\u7684\u7cbe\u9ad3\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aVisRAG: Vision-based Retrieval-augmented Generation on Multi-modality Documents<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/openbmb\/visrag<\/p>\n

\u901a\u8fc7\u6784\u5efa\u57fa\u4e8e\u89c6\u89c9-\u8bed\u8a00\u6a21\u578b (VLM) \u7684RAG\u6d41\u7a0b\uff0c\u76f4\u63a5\u5c06\u6587\u6863\u4f5c\u4e3a\u56fe\u50cf\u5d4c\u5165\u5e76\u68c0\u7d22\uff0c\u4ece\u800c\u589e\u5f3a\u751f\u6210\u6548\u679c\u3002\u76f8\u6bd4\u4f20\u7edf\u6587\u672cRAG\uff0cVisRAG\u907f\u514d\u4e86\u89e3\u6790\u8fc7\u7a0b\u4e2d\u7684\u4fe1\u606f\u635f\u5931\uff0c\u66f4\u5168\u9762\u5730\u4fdd\u7559\u4e86\u539f\u59cb\u6587\u6863\u7684\u4fe1\u606f\u3002\u5b9e\u9a8c\u663e\u793a\uff0cVisRAG\u5728\u68c0\u7d22\u548c\u751f\u6210\u9636\u6bb5\u5747\u8d85\u8d8a\u4f20\u7edfRAG\uff0c\u7aef\u5230\u7aef\u6027\u80fd\u63d0\u5347\u8fbe25-39%\u3002VisRAG\u4e0d\u4ec5\u6709\u6548\u5229\u7528\u8bad\u7ec3\u6570\u636e\uff0c\u8fd8\u5c55\u73b0\u51fa\u5f3a\u5927\u7684\u6cdb\u5316\u80fd\u529b\uff0c\u6210\u4e3a\u591a\u6a21\u6001\u6587\u6863RAG\u7684\u7406\u60f3\u9009\u62e9\u3002<\/p>\n

<\/p>\n

(55) AGENTiGraph\u3010\u77e5\u8bc6\u7ba1\u5bb6\u3011<\/h2>\n

\u77e5\u8bc6\u7ba1\u5bb6<\/strong>\uff1a\u50cf\u4e2a\u5584\u4e8e\u5bf9\u8bdd\u7684\u56fe\u4e66\u7ba1\u7406\u5458\uff0c\u901a\u8fc7\u65e5\u5e38\u4ea4\u6d41\u5e2e\u4f60\u6574\u7406\u548c\u5c55\u793a\u77e5\u8bc6\uff0c\u5e26\u7740\u4e00\u961f\u52a9\u624b\u968f\u65f6\u51c6\u5907\u89e3\u7b54\u95ee\u9898\u3001\u66f4\u65b0\u8d44\u6599\uff0c\u8ba9\u77e5\u8bc6\u7ba1\u7406\u53d8\u5f97\u7b80\u5355\u81ea\u7136\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aAGENTiGraph: An Interactive Knowledge Graph Platform for LLM-based Chatbots Utilizing Private Data<\/p>\n

AGENTiGraph\u901a\u8fc7\u81ea\u7136\u8bed\u8a00\u4ea4\u4e92\u8fdb\u884c\u77e5\u8bc6\u7ba1\u7406\u7684\u5e73\u53f0\u3002\u5b83\u96c6\u6210\u4e86\u77e5\u8bc6\u63d0\u53d6\u3001\u96c6\u6210\u548c\u5b9e\u65f6\u53ef\u89c6\u5316\u3002AGENTiGraph \u91c7\u7528\u591a\u667a\u80fd\u4f53\u67b6\u6784\u6765\u52a8\u6001\u89e3\u91ca\u7528\u6237\u610f\u56fe\u3001\u7ba1\u7406\u4efb\u52a1\u548c\u96c6\u6210\u65b0\u77e5\u8bc6\uff0c\u786e\u4fdd\u80fd\u591f\u9002\u5e94\u4e0d\u65ad\u53d8\u5316\u7684\u7528\u6237\u9700\u6c42\u548c\u6570\u636e\u4e0a\u4e0b\u6587\u3002<\/p>\n

<\/p>\n

(56) RuleRAG\u3010\u5faa\u89c4\u8e48\u77e9\u3011<\/h2>\n

\u5faa\u89c4\u8e48\u77e9<\/strong>\uff1a\u7528\u89c4\u77e9\u6765\u6559AI\u505a\u4e8b\uff0c\u5c31\u50cf\u5e26\u65b0\u4eba\u5165\u804c\uff0c\u5148\u7ed9\u672c\u5458\u5de5\u624b\u518c\u3002\u4e0d\u662f\u6f2b\u65e0\u76ee\u7684\u5730\u5b66\uff0c\u800c\u662f\u50cf\u4e2a\u4e25\u683c\u7684\u8001\u5e08\uff0c\u5148\u628a\u89c4\u77e9\u548c\u8303\u4f8b\u90fd\u8bb2\u660e\u767d\uff0c\u7136\u540e\u518d\u8ba9\u5b66\u751f\u81ea\u5df1\u52a8\u624b\u3002\u505a\u591a\u4e86\uff0c\u8fd9\u4e9b\u89c4\u77e9\u5c31\u53d8\u6210\u4e86\u808c\u8089\u8bb0\u5fc6\uff0c\u4e0b\u6b21\u9047\u5230\u7c7b\u4f3c\u95ee\u9898\u81ea\u7136\u77e5\u9053\u600e\u4e48\u5904\u7406\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aRuleRAG: Rule-guided retrieval-augmented generation with language models for question answering<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/chenzhongwu20\/RuleRAG_ICL_FT<\/p>\n

RuleRAG\u63d0\u51fa\u4e86\u57fa\u4e8e\u8bed\u8a00\u6a21\u578b\u7684\u89c4\u5219\u5f15\u5bfc\u68c0\u7d22\u589e\u5f3a\u751f\u6210\u65b9\u6cd5\uff0c\u8be5\u65b9\u6cd5\u660e\u786e\u5f15\u5165\u7b26\u53f7\u89c4\u5219\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u5b66\u4e60\uff08RuleRAG – ICL\uff09\u7684\u793a\u4f8b\uff0c\u4ee5\u5f15\u5bfc\u68c0\u7d22\u5668\u6309\u7167\u89c4\u5219\u65b9\u5411\u68c0\u7d22\u903b\u8f91\u76f8\u5173\u7684\u6587\u6863\uff0c\u5e76\u7edf\u4e00\u5f15\u5bfc\u751f\u6210\u5668\u5728\u540c\u4e00\u7ec4\u89c4\u5219\u7684\u6307\u5bfc\u4e0b\u751f\u6210\u6709\u4f9d\u636e\u7684\u7b54\u6848\u3002\u6b64\u5916\uff0c\u67e5\u8be2\u548c\u89c4\u5219\u7684\u7ec4\u5408\u53ef\u8fdb\u4e00\u6b65\u7528\u4f5c\u6709\u76d1\u7763\u7684\u5fae\u8c03\u6570\u636e\uff0c\u7528\u4ee5\u66f4\u65b0\u68c0\u7d22\u5668\u548c\u751f\u6210\u5668\uff08RuleRAG – FT\uff09\uff0c\u4ece\u800c\u5b9e\u73b0\u66f4\u597d\u7684\u57fa\u4e8e\u89c4\u5219\u7684\u6307\u4ee4\u9075\u5faa\u80fd\u529b\uff0c\u8fdb\u800c\u68c0\u7d22\u5230\u66f4\u5177\u652f\u6301\u6027\u7684\u7ed3\u679c\u5e76\u751f\u6210\u66f4\u53ef\u63a5\u53d7\u7684\u7b54\u6848\u3002<\/p>\n

<\/p>\n

(57) Class-RAG\u3010\u6cd5\u5b98\u3011<\/h2>\n

\u6cd5\u5b98<\/strong>\uff1a\u4e0d\u662f\u9760\u6b7b\u677f\u7684\u6761\u6587\u5224\u6848\uff0c\u800c\u662f\u901a\u8fc7\u4e0d\u65ad\u6269\u5145\u7684\u5224\u4f8b\u5e93\u6765\u7814\u5224\u3002\u50cf\u4e2a\u7ecf\u9a8c\u8001\u5230\u7684\u6cd5\u5b98\uff0c\u624b\u63e1\u6d3b\u9875\u6cd5\u5178\uff0c\u968f\u65f6\u7ffb\u9605\u6700\u65b0\u6848\u4f8b\uff0c\u8ba9\u5224\u51b3\u65e2\u6709\u6e29\u5ea6\u53c8\u6709\u5c3a\u5ea6\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aClass-RAG: Content Moderation with Retrieval Augmented Generation<\/p>\n

\u5185\u5bb9\u5ba1\u6838\u5206\u7c7b\u5668\u5bf9\u751f\u6210\u5f0f AI \u7684\u5b89\u5168\u6027\u81f3\u5173\u91cd\u8981\u3002\u7136\u800c\uff0c\u5b89\u5168\u4e0e\u4e0d\u5b89\u5168\u5185\u5bb9\u95f4\u7684\u7ec6\u5fae\u5dee\u522b\u5e38\u4ee4\u4eba\u96be\u4ee5\u533a\u5206\u3002\u968f\u7740\u6280\u672f\u5e7f\u6cdb\u5e94\u7528\uff0c\u6301\u7eed\u5fae\u8c03\u6a21\u578b\u4ee5\u5e94\u5bf9\u98ce\u9669\u53d8\u5f97\u6108\u53d1\u56f0\u96be\u4e14\u6602\u8d35\u3002\u4e3a\u6b64\uff0c\u6211\u4eec\u63d0\u51fa Class-RAG \u65b9\u6cd5\uff0c\u901a\u8fc7\u52a8\u6001\u66f4\u65b0\u68c0\u7d22\u5e93\uff0c\u5b9e\u73b0\u5373\u65f6\u98ce\u9669\u7f13\u89e3\u3002\u4e0e\u4f20\u7edf\u5fae\u8c03\u6a21\u578b\u76f8\u6bd4\uff0cClass-RAG \u66f4\u5177\u7075\u6d3b\u6027\u4e0e\u900f\u660e\u5ea6\uff0c\u4e14\u5728\u5206\u7c7b\u4e0e\u6297\u653b\u51fb\u65b9\u9762\u8868\u73b0\u66f4\u4f73\u3002\u7814\u7a76\u8fd8\u8868\u660e\uff0c\u6269\u5927\u68c0\u7d22\u5e93\u80fd\u6709\u6548\u63d0\u5347\u5ba1\u6838\u6027\u80fd\uff0c\u6210\u672c\u4f4e\u5ec9\u3002<\/p>\n

<\/p>\n

(58) Self-RAG\u3010\u53cd\u601d\u8005\u3011<\/h2>\n

\u53cd\u601d\u8005<\/strong>\uff1a\u5728\u56de\u7b54\u95ee\u9898\u65f6\uff0c\u4e0d\u4ec5\u4f1a\u67e5\u9605\u8d44\u6599\uff0c\u8fd8\u4f1a\u4e0d\u65ad\u601d\u8003\u548c\u68c0\u67e5\u81ea\u5df1\u7684\u7b54\u6848\u662f\u5426\u51c6\u786e\u5b8c\u6574\u3002\u901a\u8fc7”\u8fb9\u8bf4\u8fb9\u60f3”\u7684\u65b9\u5f0f\uff0c\u50cf\u4e00\u4e2a\u8c28\u614e\u7684\u5b66\u8005\u4e00\u6837\uff0c\u786e\u4fdd\u6bcf\u4e2a\u89c2\u70b9\u90fd\u6709\u53ef\u9760\u7684\u4f9d\u636e\u652f\u6301\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aSelf-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/AkariAsai\/self-rag<\/p>\n

Self-RAG\u901a\u8fc7\u68c0\u7d22\u548c\u81ea\u6211\u53cd\u601d\u6765\u63d0\u5347\u8bed\u8a00\u6a21\u578b\u7684\u8d28\u91cf\u548c\u51c6\u786e\u6027\u3002\u6846\u67b6\u8bad\u7ec3\u4e00\u4e2a\u5355\u4e00\u7684\u4efb\u610f\u8bed\u8a00\u6a21\u578b\uff0c\u8be5\u6a21\u578b\u80fd\u6309\u9700\u81ea\u9002\u5e94\u5730\u68c0\u7d22\u6587\u6bb5\uff0c\u5e76\u4f7f\u7528\u88ab\u79f0\u4e3a\u53cd\u601d\u6807\u8bb0\u7684\u7279\u6b8a\u6807\u8bb0\u6765\u5bf9\u68c0\u7d22\u5230\u7684\u6587\u6bb5\u53ca\u5176\u81ea\u8eab\u751f\u6210\u7684\u5185\u5bb9\u8fdb\u884c\u751f\u6210\u548c\u53cd\u601d\u3002\u751f\u6210\u53cd\u601d\u6807\u8bb0\u4f7f\u5f97\u8bed\u8a00\u6a21\u578b\u5728\u63a8\u7406\u9636\u6bb5\u5177\u5907\u53ef\u63a7\u6027\uff0c\u4f7f\u5176\u80fd\u591f\u6839\u636e\u4e0d\u540c\u7684\u4efb\u52a1\u8981\u6c42\u8c03\u6574\u81ea\u8eab\u884c\u4e3a\u3002<\/p>\n

<\/p>\n

(59) SimRAG\u3010\u81ea\u5b66\u6210\u624d\u3011<\/h2>\n

\u81ea\u5b66\u6210\u624d<\/strong>\uff1a\u9762\u5bf9\u4e13\u4e1a\u9886\u57df\u65f6\uff0c\u5148\u81ea\u5df1\u63d0\u95ee\u518d\u81ea\u5df1\u56de\u7b54\uff0c\u901a\u8fc7\u4e0d\u65ad\u7ec3\u4e60\u6765\u63d0\u5347\u4e13\u4e1a\u77e5\u8bc6\u50a8\u5907\uff0c\u5c31\u50cf\u5b66\u751f\u901a\u8fc7\u53cd\u590d\u505a\u4e60\u9898\u6765\u719f\u6089\u4e13\u4e1a\u77e5\u8bc6\u4e00\u6837\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aSimRAG: Self-Improving Retrieval-Augmented Generation for Adapting Large Language Models to Specialized Domains<\/p>\n

SimRAG\u662f\u4e00\u79cd\u81ea\u8bad\u7ec3\u65b9\u6cd5\uff0c\u4f7fLLM\u5177\u5907\u95ee\u7b54\u548c\u95ee\u9898\u751f\u6210\u7684\u8054\u5408\u80fd\u529b\u4ee5\u9002\u5e94\u7279\u5b9a\u9886\u57df\u3002\u53ea\u6709\u771f\u6b63\u7406\u89e3\u4e86\u77e5\u8bc6\uff0c\u624d\u80fd\u63d0\u51fa\u597d\u7684\u95ee\u9898\u3002\u8fd9\u4e24\u4e2a\u80fd\u529b\u76f8\u8f85\u76f8\u6210\uff0c\u53ef\u4ee5\u5e2e\u52a9\u6a21\u578b\u66f4\u597d\u5730\u7406\u89e3\u4e13\u4e1a\u77e5\u8bc6\u3002\u9996\u5148\u5728\u6307\u4ee4\u9075\u5faa\u3001\u95ee\u7b54\u548c\u641c\u7d22\u76f8\u5173\u6570\u636e\u4e0a\u5bf9LLM\u8fdb\u884c\u5fae\u8c03\u3002\u7136\u540e\uff0c\u5b83\u4fc3\u4f7f\u540c\u4e00LLM\u4ece\u65e0\u6807\u7b7e\u8bed\u6599\u5e93\u4e2d\u751f\u6210\u5404\u79cd\u4e0e\u9886\u57df\u76f8\u5173\u7684\u95ee\u9898\uff0c\u5e76\u91c7\u7528\u989d\u5916\u7684\u8fc7\u6ee4\u7b56\u7565\u6765\u4fdd\u7559\u9ad8\u8d28\u91cf\u7684\u5408\u6210\u793a\u4f8b\u3002\u901a\u8fc7\u5229\u7528\u8fd9\u4e9b\u5408\u6210\u793a\u4f8b\uff0cLLM\u53ef\u4ee5\u63d0\u9ad8\u5176\u5728\u7279\u5b9a\u9886\u57dfRAG\u4efb\u52a1\u4e0a\u7684\u6027\u80fd\u3002<\/p>\n

<\/p>\n

(60) ChunkRAG\u3010\u6458\u6284\u8fbe\u4eba\u3011<\/h2>\n

\u6458\u6284\u8fbe\u4eba<\/strong>\uff1a\u5148\u628a\u957f\u6587\u7ae0\u5206\u6210\u5c0f\u6bb5\u843d\uff0c\u518d\u7528\u4e13\u4e1a\u773c\u5149\u6311\u51fa\u6700\u76f8\u5173\u7684\u7247\u6bb5\uff0c\u65e2\u4e0d\u9057\u6f0f\u91cd\u70b9\uff0c\u53c8\u4e0d\u88ab\u65e0\u5173\u5185\u5bb9\u5e72\u6270\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aChunkRAG: Novel LLM-Chunk Filtering Method for RAG Systems<\/p>\n

ChunkRAG\u63d0\u51faLLM\u9a71\u52a8\u7684\u5757\u8fc7\u6ee4\u65b9\u6cd5\uff0c\u901a\u8fc7\u5728\u5757\u7ea7\u522b\u8bc4\u4f30\u548c\u8fc7\u6ee4\u68c0\u7d22\u5230\u7684\u4fe1\u606f\u6765\u589e\u5f3aRAG\u7cfb\u7edf\u7684\u6846\u67b6\uff0c\u5176\u4e2d \u201c\u5757\u201d \u4ee3\u8868\u6587\u6863\u4e2d\u8f83\u5c0f\u7684\u8fde\u8d2f\u90e8\u5206\u3002\u6211\u4eec\u7684\u65b9\u6cd5\u91c7\u7528\u8bed\u4e49\u5206\u5757\u5c06\u6587\u6863\u5212\u5206\u4e3a\u8fde\u8d2f\u7684\u90e8\u5206\uff0c\u5e76\u5229\u7528\u57fa\u4e8e\u5927\u8bed\u8a00\u6a21\u578b\u7684\u76f8\u5173\u6027\u8bc4\u5206\u6765\u8bc4\u4f30\u6bcf\u4e2a\u5757\u4e0e\u7528\u6237\u67e5\u8be2\u7684\u5339\u914d\u7a0b\u5ea6\u3002\u901a\u8fc7\u5728\u751f\u6210\u9636\u6bb5\u4e4b\u524d\u8fc7\u6ee4\u6389\u4e0d\u592a\u76f8\u5173\u7684\u5757\uff0c\u6211\u4eec\u663e\u8457\u51cf\u5c11\u4e86\u5e7b\u89c9\u5e76\u63d0\u9ad8\u4e86\u4e8b\u5b9e\u51c6\u786e\u6027\u3002<\/p>\n

<\/p>\n

(61) FastGraphRAG\u3010\u96f7\u8fbe\u3011<\/h2>\n

\u96f7\u8fbe<\/strong>\uff1a\u50cf\u8c37\u6b4c\u7f51\u9875\u6392\u540d\u4e00\u6837\uff0c\u7ed9\u77e5\u8bc6\u70b9\u4e5f\u6392\u51fa\u4e2a\u70ed\u5ea6\u699c\u3002\u5c31\u597d\u6bd4\u793e\u4ea4\u7f51\u7edc\u4e2d\u7684\u610f\u89c1\u9886\u8896\uff0c\u8d8a\u591a\u4eba\u5173\u6ce8\u5c31\u8d8a\u5bb9\u6613\u88ab\u770b\u89c1\u3002\u5b83\u4e0d\u662f\u6f2b\u65e0\u76ee\u7684\u5730\u641c\u7d22\uff0c\u800c\u662f\u50cf\u4e2a\u5e26\u7740\u96f7\u8fbe\u7684\u4fa6\u5bdf\u5175\uff0c\u54ea\u91cc\u7684\u4fe1\u53f7\u5f3a\u5c31\u5f80\u54ea\u91cc\u770b\u3002<\/p><\/blockquote>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/circlemind-ai\/fast-graphrag<\/p>\n

FastGraphRAG\u63d0\u4f9b\u4e86\u4e00\u4e2a\u9ad8\u6548\u3001\u53ef\u89e3\u91ca\u4e14\u7cbe\u5ea6\u9ad8\u7684\u5feb\u901f\u56fe\u68c0\u7d22\u589e\u5f3a\u751f\u6210\uff08FastGraphRAG\uff09\u6846\u67b6\u3002\u5b83\u5c06PageRank\u7b97\u6cd5\u5e94\u7528\u4e8e\u77e5\u8bc6\u56fe\u8c31\u7684\u904d\u5386\u8fc7\u7a0b\uff0c\u5feb\u901f\u5b9a\u4f4d\u6700\u76f8\u5173\u7684\u77e5\u8bc6\u8282\u70b9\u3002\u901a\u8fc7\u8ba1\u7b97\u8282\u70b9\u7684\u91cd\u8981\u6027\u5f97\u5206\uff0cPageRank\u4f7fGraphRAG\u80fd\u591f\u66f4\u667a\u80fd\u5730\u7b5b\u9009\u548c\u6392\u5e8f\u77e5\u8bc6\u56fe\u8c31\u4e2d\u7684\u4fe1\u606f\u3002\u8fd9\u5c31\u50cf\u662f\u4e3aGraphRAG\u88c5\u4e0a\u4e86\u4e00\u4e2a”\u91cd\u8981\u6027\u96f7\u8fbe”\uff0c\u80fd\u591f\u5728\u6d69\u5982\u70df\u6d77\u7684\u6570\u636e\u4e2d\u5feb\u901f\u5b9a\u4f4d\u5173\u952e\u4fe1\u606f\u3002<\/p>\n

<\/p>\n

(62) AutoRAG\u3010\u8c03\u97f3\u5e08\u3011<\/h2>\n

\u8c03\u97f3\u5e08<\/strong>\uff1a\u4e00\u4f4d\u7ecf\u9a8c\u4e30\u5bcc\u7684\u8c03\u97f3\u5e08\uff0c\u4e0d\u662f\u9760\u731c\u6d4b\u8c03\u97f3\uff0c\u800c\u662f\u901a\u8fc7\u79d1\u5b66\u6d4b\u8bd5\u627e\u5230\u6700\u4f73\u97f3\u6548\u3002\u5b83\u4f1a\u81ea\u52a8\u5c1d\u8bd5\u5404\u79cdRAG\u7ec4\u5408\uff0c\u5c31\u50cf\u8c03\u97f3\u5e08\u6d4b\u8bd5\u4e0d\u540c\u7684\u97f3\u54cd\u8bbe\u5907\u642d\u914d\uff0c\u6700\u7ec8\u627e\u5230\u6700\u548c\u8c10\u7684”\u6f14\u594f\u65b9\u6848”\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aAutoRAG: Automated Framework for optimization of Retrieval Augmented Generation Pipeline<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/Marker-Inc-Korea\/AutoRAG_ARAGOG_Paper<\/p>\n

AutoRAG\u6846\u67b6\u80fd\u591f\u81ea\u52a8\u4e3a\u7ed9\u5b9a\u6570\u636e\u96c6\u8bc6\u522b\u5408\u9002\u7684RAG\u6a21\u5757\uff0c\u5e76\u63a2\u7d22\u548c\u903c\u8fd1\u8be5\u6570\u636e\u96c6\u7684RAG\u6a21\u5757\u7684\u6700\u4f18\u7ec4\u5408\u3002\u901a\u8fc7\u7cfb\u7edf\u8bc4\u4f30\u4e0d\u540c\u7684RAG\u8bbe\u7f6e\u6765\u4f18\u5316\u6280\u672f\u9009\u62e9\uff0c\u8be5\u6846\u67b6\u7c7b\u4f3c\u4e8e\u4f20\u7edf\u673a\u5668\u5b66\u4e60\u4e2d\u7684AutoML\u5b9e\u8df5\uff0c\u901a\u8fc7\u5e7f\u6cdb\u5b9e\u9a8c\u6765\u4f18\u5316RAG\u6280\u672f\u7684\u9009\u62e9\uff0c\u63d0\u9ad8RAG\u7cfb\u7edf\u7684\u6548\u7387\u548c\u53ef\u6269\u5c55\u6027\u3002<\/p>\n

<\/p>\n

(63) Plan\u00d7RAG\u3010\u9879\u76ee\u7ecf\u7406\u3011<\/h2>\n

\u9879\u76ee\u7ecf\u7406<\/strong>\uff1a\u5148\u89c4\u5212\u540e\u884c\u52a8\uff0c\u628a\u5927\u4efb\u52a1\u5206\u89e3\u6210\u5c0f\u4efb\u52a1\uff0c\u5b89\u6392\u591a\u4e2a”\u4e13\u5bb6”\u5e76\u884c\u5de5\u4f5c\u3002\u6bcf\u4e2a\u4e13\u5bb6\u8d1f\u8d23\u81ea\u5df1\u7684\u9886\u57df\uff0c\u6700\u540e\u7531\u9879\u76ee\u7ecf\u7406\u7edf\u7b79\u6c47\u603b\u7ed3\u679c\u3002\u8fd9\u79cd\u65b9\u5f0f\u4e0d\u4ec5\u66f4\u5feb\u3001\u66f4\u51c6\uff0c\u8fd8\u80fd\u6e05\u695a\u4ea4\u4ee3\u6bcf\u4e2a\u7ed3\u8bba\u7684\u6765\u6e90\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aPlan\u00d7RAG: Planning-guided Retrieval Augmented Generation<\/p>\n

Plan\u00d7RAG\u662f\u4e00\u4e2a\u65b0\u9896\u7684\u6846\u67b6\uff0c\u5b83\u5c06\u73b0\u6709RAG\u6846\u67b6\u7684 \u201c\u68c0\u7d22 – \u63a8\u7406\u201d \u8303\u5f0f\u6269\u5145\u4e3a \u201c\u8ba1\u5212 – \u68c0\u7d22\u201d\u8303\u5f0f\u3002Plan\u00d7RAG \u5c06\u63a8\u7406\u8ba1\u5212\u5236\u5b9a\u4e3a\u6709\u5411\u65e0\u73af\u56fe\uff08DAG\uff09\uff0c\u5c06\u67e5\u8be2\u5206\u89e3\u4e3a\u76f8\u4e92\u5173\u8054\u7684\u539f\u5b50\u5b50\u67e5\u8be2\u3002\u7b54\u6848\u751f\u6210\u9075\u5faa DAG \u7ed3\u6784\uff0c\u901a\u8fc7\u5e76\u884c\u68c0\u7d22\u548c\u751f\u6210\u663e\u8457\u63d0\u9ad8\u6548\u7387\u3002\u867d\u7136\u6700\u5148\u8fdb\u7684RAG\u89e3\u51b3\u65b9\u6848\u9700\u8981\u5927\u91cf\u7684\u6570\u636e\u751f\u6210\u548c\u8bed\u8a00\u6a21\u578b\uff08LMs\uff09\u7684\u5fae\u8c03\uff0c\u4f46Plan\u00d7RAG\u7eb3\u5165\u4e86\u51bb\u7ed3\u7684LMs\u4f5c\u4e3a\u5373\u63d2\u5373\u7528\u7684\u4e13\u5bb6\u6765\u751f\u6210\u9ad8\u8d28\u91cf\u7684\u7b54\u6848\u3002<\/p>\n

<\/p>\n

(64) SubgraphRAG\u3010\u5b9a\u4f4d\u4eea\u3011<\/h2>\n

\u5b9a\u4f4d\u4eea<\/strong>\uff1a\u4e0d\u662f\u6f2b\u65e0\u76ee\u7684\u5730\u5927\u6d77\u635e\u9488\uff0c\u800c\u662f\u7cbe\u51c6\u7ed8\u5236\u4e00\u5f20\u5c0f\u578b\u77e5\u8bc6\u5730\u56fe\uff0c\u8ba9 AI \u80fd\u5feb\u901f\u627e\u5230\u7b54\u6848\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aSimple is Effective: The Roles of Graphs and Large Language Models in Knowledge-Graph-Based Retrieval-Augmented Generation<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/Graph-COM\/SubgraphRAG<\/p>\n

SubgraphRAG\u6269\u5c55\u4e86\u57fa\u4e8eKG\u7684RAG\u6846\u67b6\uff0c\u901a\u8fc7\u68c0\u7d22\u5b50\u56fe\u5e76\u5229\u7528LLM\u8fdb\u884c\u63a8\u7406\u548c\u7b54\u6848\u9884\u6d4b\u3002\u5c06\u8f7b\u91cf\u7ea7\u591a\u5c42\u611f\u77e5\u5668\u4e0e\u5e76\u884c\u4e09\u5143\u7ec4\u8bc4\u5206\u673a\u5236\u76f8\u7ed3\u5408\uff0c\u4ee5\u5b9e\u73b0\u9ad8\u6548\u7075\u6d3b\u7684\u5b50\u56fe\u68c0\u7d22\uff0c\u540c\u65f6\u7f16\u7801\u6709\u5411\u7ed3\u6784\u8ddd\u79bb\u4ee5\u63d0\u9ad8\u68c0\u7d22\u6709\u6548\u6027\u3002\u68c0\u7d22\u5230\u7684\u5b50\u56fe\u5927\u5c0f\u53ef\u4ee5\u7075\u6d3b\u8c03\u6574\uff0c\u4ee5\u5339\u914d\u67e5\u8be2\u9700\u6c42\u548c\u4e0b\u6e38LLM\u7684\u80fd\u529b\u3002\u8fd9\u79cd\u8bbe\u8ba1\u5728\u6a21\u578b\u590d\u6742\u6027\u548c\u63a8\u7406\u80fd\u529b\u4e4b\u95f4\u53d6\u5f97\u4e86\u5e73\u8861\uff0c\u5b9e\u73b0\u4e86\u53ef\u6269\u5c55\u4e14\u901a\u7528\u7684\u68c0\u7d22\u8fc7\u7a0b\u3002<\/p>\n

<\/p>\n

(65) RuRAG\u3010\u70bc\u91d1\u672f\u58eb\u3011<\/h2>\n

\u70bc\u91d1\u672f\u58eb<\/strong>\uff1a\u50cf\u4e2a\u70bc\u91d1\u672f\u58eb\uff0c\u80fd\u5c06\u6d77\u91cf\u6570\u636e\u63d0\u70bc\u6210\u6e05\u6670\u7684\u903b\u8f91\u89c4\u5219\uff0c\u5e76\u7528\u901a\u4fd7\u6613\u61c2\u7684\u8bed\u8a00\u8868\u8fbe\u51fa\u6765\uff0c\u8ba9AI\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u66f4\u6709\u667a\u6167\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aRuAG: Learned-rule-augmented Generation for Large Language Models<\/p>\n

\u65e8\u5728\u901a\u8fc7\u5c06\u5927\u91cf\u79bb\u7ebf\u6570\u636e\u81ea\u52a8\u84b8\u998f\u6210\u53ef\u89e3\u91ca\u7684\u4e00\u9636\u903b\u8f91\u89c4\u5219\uff0c\u5e76\u6ce8\u5165\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u4e2d\uff0c\u4ee5\u63d0\u5347\u5176\u63a8\u7406\u80fd\u529b\u3002\u8be5\u6846\u67b6\u4f7f\u7528\u8499\u7279\u5361\u6d1b\u6811\u641c\u7d22\uff08MCTS\uff09\u6765\u53d1\u73b0\u903b\u8f91\u89c4\u5219\uff0c\u5e76\u5c06\u8fd9\u4e9b\u89c4\u5219\u8f6c\u5316\u4e3a\u81ea\u7136\u8bed\u8a00\uff0c\u5b9e\u73b0\u9488\u5bf9LLM\u4e0b\u6e38\u4efb\u52a1\u7684\u77e5\u8bc6\u6ce8\u5165\u548c\u65e0\u7f1d\u96c6\u6210\u3002\u8be5\u8bba\u6587\u5728\u516c\u5171\u548c\u79c1\u6709\u5de5\u4e1a\u4efb\u52a1\u4e0a\u8bc4\u4f30\u4e86\u8be5\u6846\u67b6\u7684\u6709\u6548\u6027\uff0c\u8bc1\u660e\u4e86\u5176\u5728\u591a\u6837\u5316\u4efb\u52a1\u4e2d\u589e\u5f3aLLM\u80fd\u529b\u7684\u6f5c\u529b\u3002<\/p>\n

<\/p>\n

(66) RAGViz\u3010\u900f\u89c6\u773c\u3011<\/h2>\n

\u900f\u89c6\u773c<\/strong>\uff1a\u8ba9RAG\u7cfb\u7edf\u53d8\u900f\u660e\uff0c\u770b\u5f97\u89c1\u6a21\u578b\u5728\u8bfb\u54ea\u53e5\u8bdd\uff0c\u50cf\u533b\u751f\u770bX\u5149\u7247\u4e00\u6837\uff0c\u54ea\u91cc\u4e0d\u5bf9\u4e00\u76ee\u4e86\u7136\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aRAGViz: Diagnose and Visualize Retrieval-Augmented Generation<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/cxcscmu\/RAGViz<\/p>\n

RAGViz\u63d0\u4f9b\u4e86\u5bf9\u68c0\u7d22\u6587\u6863\u548c\u6a21\u578b\u6ce8\u610f\u529b\u7684\u53ef\u89c6\u5316\uff0c\u5e2e\u52a9\u7528\u6237\u7406\u89e3\u751f\u6210\u7684\u6807\u8bb0\u4e0e\u68c0\u7d22\u6587\u6863\u4e4b\u95f4\u7684\u4ea4\u4e92\uff0c\u53ef\u7528\u4e8e\u8bca\u65ad\u548c\u53ef\u89c6\u5316RAG\u7cfb\u7edf\u3002<\/p>\n

<\/p>\n

(67) AgenticRAG\u3010\u667a\u80fd\u52a9\u624b\u3011<\/h2>\n

\u667a\u80fd\u52a9\u624b<\/strong>\uff1a\u4e0d\u518d\u662f\u7b80\u5355\u7684\u67e5\u627e\u590d\u5236\uff0c\u800c\u662f\u914d\u4e86\u4e2a\u80fd\u5f53\u673a\u8981\u79d8\u4e66\u7684\u52a9\u624b\u3002\u50cf\u4e2a\u5f97\u529b\u7684\u884c\u653f\u5b98\uff0c\u4e0d\u5149\u4f1a\u67e5\u8d44\u6599\uff0c\u8fd8\u77e5\u9053\u4ec0\u4e48\u65f6\u5019\u8be5\u6253\u7535\u8bdd\uff0c\u4ec0\u4e48\u65f6\u5019\u8be5\u5f00\u4f1a\uff0c\u4ec0\u4e48\u65f6\u5019\u8be5\u8bf7\u793a\u9886\u5bfc\u3002<\/p><\/blockquote>\n

AgenticRAG\u63cf\u8ff0\u4e86\u57fa\u4e8eAI\u667a\u80fd\u4f53\u5b9e\u73b0\u7684RAG\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u5b83\u5c06AI\u667a\u80fd\u4f53\u7eb3\u5165RAG\u6d41\u7a0b\u4e2d\uff0c\u4ee5\u534f\u8c03\u5176\u7ec4\u4ef6\u5e76\u6267\u884c\u8d85\u51fa\u7b80\u5355\u4fe1\u606f\u68c0\u7d22\u548c\u751f\u6210\u7684\u989d\u5916\u884c\u52a8\uff0c\u4ee5\u514b\u670d\u975e\u667a\u80fd\u4f53\u6d41\u7a0b\u7684\u5c40\u9650\u6027\u3002<\/p>\n

<\/p>\n

(68) HtmlRAG\u3010\u6392\u7248\u5e08\u3011<\/h2>\n

\u6392\u7248\u5e08<\/strong>\uff1a\u628a\u77e5\u8bc6\u4e0d\u662f\u5f53\u4f5c\u6d41\u6c34\u8d26\u6765\u8bb0\uff0c\u800c\u662f\u50cf\u6392\u7248\u6742\u5fd7\u4e00\u6837\uff0c\u8be5\u52a0\u7c97\u7684\u52a0\u7c97\uff0c\u8be5\u6807\u7ea2\u7684\u6807\u7ea2\u3002\u5c31\u50cf\u4e00\u4e2a\u6311\u5254\u7684\u7f8e\u7f16\uff0c\u89c9\u5f97\u5149\u6709\u5185\u5bb9\u4e0d\u591f\uff0c\u8fd8\u5f97\u8bb2\u7a76\u6392\u7248\uff0c\u8fd9\u6837\u91cd\u70b9\u624d\u80fd\u4e00\u76ee\u4e86\u7136\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aHtmlRAG: HTML is Better Than Plain Text for Modeling Retrieved Knowledge in RAG Systems<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/plageon\/HtmlRAG<\/p>\n

HtmlRAG\u5728RAG\u4e2d\u4f7f\u7528HTML\u800c\u4e0d\u662f\u7eaf\u6587\u672c\u4f5c\u4e3a\u68c0\u7d22\u77e5\u8bc6\u7684\u683c\u5f0f\uff0c\u5728\u5bf9\u5916\u90e8\u6587\u6863\u4e2d\u7684\u77e5\u8bc6\u8fdb\u884c\u5efa\u6a21\u65f6\uff0cHTML\u6bd4\u7eaf\u6587\u672c\u66f4\u597d\uff0c\u5e76\u4e14\u5927\u591a\u6570LLM\u5177\u6709\u5f3a\u5927\u7684\u7406\u89e3HTML\u7684\u80fd\u529b\u3002HtmlRAG\u63d0\u51fa\u4e86HTML\u6e05\u7406\u3001\u538b\u7f29\u548c\u4fee\u526a\u7b56\u7565\uff0c\u4ee5\u7f29\u77edHTML\u540c\u65f6\u6700\u5c0f\u5316\u4fe1\u606f\u635f\u5931\u3002<\/p>\n

<\/p>\n

(69) M3DocRAG\u3010\u611f\u5b98\u8fbe\u4eba\u3011<\/h2>\n

\u611f\u5b98\u8fbe\u4eba<\/strong>\uff1a\u4e0d\u662f\u53ea\u4f1a\u8bfb\u4e66\uff0c\u8fd8\u80fd\u770b\u56fe\u8bc6\u56fe\uff0c\u542c\u58f0\u8fa8\u4f4d\u3002\u50cf\u4e2a\u7efc\u827a\u8282\u76ee\u91cc\u7684\u5168\u80fd\u9009\u624b\uff0c\u56fe\u7247\u80fd\u770b\u61c2\uff0c\u6587\u5b57\u80fd\u7406\u89e3\uff0c\u8be5\u8df3\u8dc3\u601d\u7ef4\u65f6\u5c31\u8df3\u8dc3\uff0c\u8be5\u4e13\u6ce8\u7ec6\u8282\u65f6\u5c31\u4e13\u6ce8\uff0c\u5404\u79cd\u6311\u6218\u90fd\u96be\u4e0d\u5012\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aM3DocRAG: Multi-modal Retrieval is What You Need for Multi-page Multi-document Understanding<\/p>\n

M3DocRAG\u662f\u4e00\u79cd\u65b0\u9896\u7684\u591a\u6a21\u6001RAG\u6846\u67b6\uff0c\u80fd\u591f\u7075\u6d3b\u9002\u5e94\u5404\u79cd\u6587\u6863\u4e0a\u4e0b\u6587\uff08\u5c01\u95ed\u57df\u548c\u5f00\u653e\u57df\uff09\u3001\u95ee\u9898\u8df3\u8f6c\uff08\u5355\u8df3\u548c\u591a\u8df3\uff09\u548c\u8bc1\u636e\u6a21\u5f0f\uff08\u6587\u672c\u3001\u56fe\u8868\u3001\u56fe\u5f62\u7b49\uff09\u3002M3DocRAG\u4f7f\u7528\u591a\u6a21\u6001\u68c0\u7d22\u5668\u548cMLM\u67e5\u627e\u76f8\u5173\u6587\u6863\u5e76\u56de\u7b54\u95ee\u9898\uff0c\u56e0\u6b64\u5b83\u53ef\u4ee5\u6709\u6548\u5730\u5904\u7406\u5355\u4e2a\u6216\u591a\u4e2a\u6587\u6863\uff0c\u540c\u65f6\u4fdd\u7559\u89c6\u89c9\u4fe1\u606f\u3002<\/p>\n

<\/p>\n

(70) KAG\u3010\u903b\u8f91\u5927\u5e08\u3011<\/h2>\n

\u903b\u8f91\u5927\u5e08<\/strong>\uff1a\u4e0d\u5149\u9760\u611f\u89c9\u627e\u76f8\u4f3c\u7684\u7b54\u6848\uff0c\u8fd8\u5f97\u8bb2\u7a76\u77e5\u8bc6\u95f4\u7684\u56e0\u679c\u5173\u7cfb\u3002\u50cf\u4e2a\u4e25\u8c28\u7684\u6570\u5b66\u8001\u5e08\uff0c\u4e0d\u4ec5\u8981\u77e5\u9053\u7b54\u6848\u662f\u4ec0\u4e48\uff0c\u8fd8\u5f97\u89e3\u91ca\u6e05\u695a\u8fd9\u7b54\u6848\u662f\u600e\u4e48\u4e00\u6b65\u6b65\u63a8\u5bfc\u51fa\u6765\u7684\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aKAG: Boosting LLMs in Professional Domains via Knowledge Augmented Generation<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/OpenSPG\/KAG<\/p>\n

RAG\u4e2d\u5411\u91cf\u76f8\u4f3c\u6027\u4e0e\u77e5\u8bc6\u63a8\u7406\u7684\u76f8\u5173\u6027\u4e4b\u95f4\u7684\u5dee\u8ddd\uff0c\u4ee5\u53ca\u5bf9\u77e5\u8bc6\u903b\u8f91\uff08\u5982\u6570\u503c\u3001\u65f6\u95f4\u5173\u7cfb\u3001\u4e13\u5bb6\u89c4\u5219\u7b49\uff09\u4e0d\u654f\u611f\u963b\u788d\u4e86\u4e13\u4e1a\u77e5\u8bc6\u670d\u52a1\u7684\u6709\u6548\u6027\u3002KAG\u7684\u8bbe\u8ba1\u76ee\u7684\u662f\u5145\u5206\u5229\u7528\u77e5\u8bc6\u56fe\u8c31\uff08KG\uff09\u548c\u5411\u91cf\u68c0\u7d22\u7684\u4f18\u52bf\u6765\u5e94\u5bf9\u4e0a\u8ff0\u6311\u6218\uff0c\u5e76\u901a\u8fc7\u4e94\u4e2a\u5173\u952e\u65b9\u9762\u53cc\u5411\u589e\u5f3a\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u548c\u77e5\u8bc6\u56fe\u8c31\u6765\u63d0\u9ad8\u751f\u6210\u548c\u63a8\u7406\u6027\u80fd\uff1a\uff081\uff09\u5bf9LLM\u53cb\u597d\u7684\u77e5\u8bc6\u8868\u793a\uff0c\uff082\uff09\u77e5\u8bc6\u56fe\u8c31\u4e0e\u539f\u59cb\u5757\u4e4b\u95f4\u7684\u76f8\u4e92\u7d22\u5f15\uff0c\uff083\uff09\u903b\u8f91\u5f62\u5f0f\u5f15\u5bfc\u7684\u6df7\u5408\u63a8\u7406\u5f15\u64ce\uff0c\uff084\uff09\u4e0e\u8bed\u4e49\u63a8\u7406\u7684\u77e5\u8bc6\u5bf9\u9f50\uff0c\uff085\uff09KAG\u7684\u6a21\u578b\u80fd\u529b\u589e\u5f3a\u3002<\/p>\n

<\/p>\n

(71) FILCO\u3010\u7b5b\u9009\u5e08\u3011<\/h2>\n

\u7b5b\u9009\u5e08<\/strong>\uff1a\u50cf\u4e2a\u4e25\u8c28\u7684\u7f16\u8f91\uff0c\u5584\u4e8e\u4ece\u5927\u91cf\u6587\u672c\u4e2d\u8bc6\u522b\u5e76\u4fdd\u7559\u6700\u6709\u4ef7\u503c\u7684\u4fe1\u606f\uff0c\u786e\u4fdd\u4f20\u9012\u7ed9AI\u7684\u6bcf\u6bb5\u5185\u5bb9\u90fd\u7cbe\u51c6\u4e14\u76f8\u5173\u3002<\/p><\/blockquote>\n

\u2022 \u8bba\u6587\uff1aLearning to Filter Context for Retrieval-Augmented Generation<\/p>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/zorazrw\/filco<\/p>\n

FILCO\u901a\u8fc7\u57fa\u4e8e\u8bcd\u6cd5\u548c\u4fe1\u606f\u8bba\u65b9\u6cd5\u8bc6\u522b\u6709\u7528\u7684\u4e0a\u4e0b\u6587\uff0c\u4ee5\u53ca\u8bad\u7ec3\u4e0a\u4e0b\u6587\u8fc7\u6ee4\u6a21\u578b\uff0c\u4ee5\u8fc7\u6ee4\u68c0\u7d22\u5230\u7684\u4e0a\u4e0b\u6587\uff0c\u6765\u63d0\u9ad8\u63d0\u4f9b\u7ed9\u751f\u6210\u5668\u7684\u4e0a\u4e0b\u6587\u8d28\u91cf\u3002<\/p>\n

<\/p>\n

(72) LazyGraphRAG\u3010\u7cbe\u7b97\u5e08\u3011<\/h2>\n

\u7cbe\u7b97\u5e08<\/strong>\uff1a\u80fd\u7701\u4e00\u6b65\u662f\u4e00\u6b65\uff0c\u628a\u8d35\u7684\u5927\u6a21\u578b\u7528\u5728\u5200\u5203\u4e0a\u3002\u5c31\u50cf\u4e2a\u4f1a\u8fc7\u65e5\u5b50\u7684\u4e3b\u5987\uff0c\u4e0d\u662f\u770b\u5230\u8d85\u5e02\u6253\u6298\u5c31\u4e70\uff0c\u800c\u662f\u8d27\u6bd4\u4e09\u5bb6\u540e\u624d\u51b3\u5b9a\u5728\u54ea\u91cc\u82b1\u94b1\u6700\u503c\u3002<\/p><\/blockquote>\n

\u2022 \u9879\u76ee\uff1ahttps:\/\/github.com\/microsoft\/graphrag<\/p>\n

\u4e00\u79cd\u65b0\u578b\u7684\u56fe\u8c31\u589e\u5f3a\u751f\u6210\u589e\u5f3a\u68c0\u7d22\uff08RAG\uff09\u65b9\u6cd5\u3002\u8fd9\u79cd\u65b9\u6cd5\u663e\u8457\u964d\u4f4e\u4e86\u7d22\u5f15\u548c\u67e5\u8be2\u6210\u672c\uff0c\u540c\u65f6\u5728\u56de\u7b54\u8d28\u91cf\u4e0a\u4fdd\u6301\u6216\u8d85\u8d8a\u7ade\u4e89\u5bf9\u624b\uff0c\u4f7f\u5176\u5728\u591a\u79cd\u7528\u4f8b\u4e2d\u5177\u6709\u9ad8\u5ea6\u7684\u53ef\u6269\u5c55\u6027\u548c\u9ad8\u6548\u6027\u3002LazyGraphRAG\u63a8\u8fdf\u4e86\u5bf9LLM\u7684\u4f7f\u7528\u3002\u5728\u7d22\u5f15\u9636\u6bb5\uff0cLazyGraphRAG\u4ec5\u4f7f\u7528\u8f7b\u91cf\u7ea7\u7684NLP\u6280\u672f\u6765\u5904\u7406\u6587\u672c\uff0c\u5c06LLM\u7684\u8c03\u7528\u5ef6\u8fdf\u5230\u5b9e\u9645\u67e5\u8be2\u65f6\u3002\u8fd9\u79cd\u201c\u61d2\u60f0\u201d\u7684\u7b56\u7565\u907f\u514d\u4e86\u524d\u671f\u9ad8\u6602\u7684\u7d22\u5f15\u6210\u672c\uff0c\u5b9e\u73b0\u4e86\u9ad8\u6548\u7684\u8d44\u6e90\u5229\u7528\u3002<\/p>\n\n\n\n\n\n\n\n\n\n\n
<\/td>\n\u4f20\u7edfGraphRAG<\/strong><\/td>\nLazyGraphRAG<\/strong><\/td>\n<\/tr>\n<\/thead>\n
\u7d22\u5f15\u9636\u6bb5<\/strong><\/td>\n– \u4f7f\u7528LLM\u63d0\u53d6\u5e76\u63cf\u8ff0\u5b9e\u4f53\u548c\u5173\u7cfb
\n– \u4e3a\u6bcf\u4e2a\u5b9e\u4f53\u548c\u5173\u7cfb\u751f\u6210\u6458\u8981
\n– \u5229\u7528LLM\u603b\u7ed3\u793e\u533a\u5185\u5bb9
\n– \u751f\u6210\u5d4c\u5165\u5411\u91cf
\n– \u751f\u6210Parquet\u6587\u4ef6<\/td>\n		– \u4f7f\u7528NLP\u6280\u672f\u63d0\u53d6\u6982\u5ff5\u548c\u5171\u73b0\u5173\u7cfb
\n– \u6784\u5efa\u6982\u5ff5\u56fe
\n– \u63d0\u53d6\u793e\u533a\u7ed3\u6784
\n– \u7d22\u5f15\u9636\u6bb5\u4e0d\u4f7f\u7528LLM<\/td>\n<\/tr>\n
\u67e5\u8be2\u9636\u6bb5<\/strong><\/td>\n– \u76f4\u63a5\u4f7f\u7528\u793e\u533a\u6458\u8981\u56de\u7b54\u67e5\u8be2
\n– \u7f3a\u4e4f\u5bf9\u67e5\u8be2\u7684\u7ec6\u5316\u548c\u5bf9\u76f8\u5173\u4fe1\u606f\u7684\u805a\u7126<\/td>\n		– \u4f7f\u7528LLM\u7ec6\u5316\u67e5\u8be2\u5e76\u751f\u6210\u5b50\u67e5\u8be2
\n– \u6839\u636e\u76f8\u5173\u6027\u9009\u62e9\u6587\u672c\u7247\u6bb5\u548c\u793e\u533a
\n– \u4f7f\u7528LLM\u63d0\u53d6\u548c\u751f\u6210\u7b54\u6848
\n– \u66f4\u52a0\u805a\u96c6\u4e8e\u76f8\u5173\u5185\u5bb9\uff0c\u56de\u7b54\u66f4\u7cbe\u786e<\/td>\n<\/tr>\n
LLM\u8c03\u7528<\/strong><\/td>\n– \u5728\u7d22\u5f15\u9636\u6bb5\u548c\u67e5\u8be2\u9636\u6bb5\u90fd\u5927\u91cf\u4f7f\u7528<\/td>\n– \u5728\u7d22\u5f15\u9636\u6bb5\u4e0d\u4f7f\u7528LLM
\n– \u4ec5\u5728\u67e5\u8be2\u9636\u6bb5\u8c03\u7528LLM
\n– LLM\u7684\u4f7f\u7528\u66f4\u52a0\u9ad8\u6548<\/td>\n<\/tr>\n
\u6210\u672c\u6548\u7387<\/strong><\/td>\n– \u7d22\u5f15\u6210\u672c\u9ad8\uff0c\u8017\u65f6\u957f
\n– \u67e5\u8be2\u6027\u80fd\u53d7\u9650\u4e8e\u7d22\u5f15\u8d28\u91cf<\/td>\n		– \u7d22\u5f15\u6210\u672c\u4ec5\u4e3a\u4f20\u7edfGraphRAG\u76840.1%
\n– \u67e5\u8be2\u6548\u7387\u9ad8\uff0c\u7b54\u6848\u8d28\u91cf\u597d<\/td>\n<\/tr>\n
\u6570\u636e\u5b58\u50a8<\/strong><\/td>\n– \u7d22\u5f15\u6570\u636e\u751f\u6210 Parquet \u6587\u4ef6\uff0c\u9002\u5408\u5927\u89c4\u6a21\u6570\u636e\u7684\u5b58\u50a8\u548c\u5904\u7406<\/td>\n– \u7d22\u5f15\u6570\u636e\u5b58\u50a8\u4e3a\u8f7b\u91cf\u7ea7\u683c\u5f0f\uff08\u5982 JSON\u3001CSV\uff09\uff0c\u66f4\u9002\u5408\u5feb\u901f\u5f00\u53d1\u548c\u5c0f\u89c4\u6a21\u6570\u636e<\/td>\n<\/tr>\n
\u4f7f\u7528\u573a\u666f<\/strong><\/td>\n– \u9002\u7528\u4e8e\u5bf9\u8ba1\u7b97\u8d44\u6e90\u548c\u65f6\u95f4\u4e0d\u654f\u611f\u7684\u573a\u666f
\n– \u9700\u8981\u63d0\u524d\u6784\u5efa\u5b8c\u6574\u7684\u77e5\u8bc6\u56fe\u8c31\uff0c\u5e76\u5b58\u50a8\u4e3aParquet\u6587\u4ef6\uff0c\u65b9\u4fbf\u540e\u7eed\u5bfc\u5165\u6570\u636e\u5e93\u8fdb\u884c\u590d\u6742\u5206\u6790<\/td>\n		– \u9002\u7528\u4e8e\u9700\u8981\u5feb\u901f\u7d22\u5f15\u548c\u54cd\u5e94\u7684\u573a\u666f
\n– \u9002\u5408\u4e00\u6b21\u6027\u67e5\u8be2\u3001\u63a2\u7d22\u6027\u5206\u6790\u548c\u6d41\u5f0f\u6570\u636e\u5904\u7406<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

 <\/p>\n

RAG Survey<\/h2>\n