{"id":9488,"date":"2025-11-11T17:31:13","date_gmt":"2025-11-11T09:31:13","guid":{"rendered":"\/?p=9488"},"modified":"2025-11-11T17:31:13","modified_gmt":"2025-11-11T09:31:13","slug":"text2sql-nl2sql%ef%bc%88%e5%9b%9b%ef%bc%89%e8%af%84%e4%bc%b0%e6%8c%87%e6%a0%87%ef%bc%9a%e5%a5%bd%e4%b8%8d%e5%a5%bd%e8%b0%81%e8%af%b4%e4%ba%86%e7%ae%97","status":"publish","type":"post","link":"\/?p=9488","title":{"rendered":"Text2SQL \/ NL2SQL\uff08\u56db\uff09\u8bc4\u4f30\u6307\u6807\uff1a\u597d\u4e0d\u597d\u8c01\u8bf4\u4e86\u7b97"},"content":{"rendered":"

\u8fd9\u662f Text2SQL \u7cfb\u5217\u7b2c\u56db\u7bc7\uff0c\u4e4b\u524d\u5206\u522b\u5199\u4e86\u6570\u636e\u96c6<\/strong>\u3001\u5f53\u4e0b SOTA \u7684 LLM-based \u7684\u5b9e\u73b0\u65b9\u5f0f\u53ca\u8036\u9c81\u5927\u5b66 2024 \u5e74\u6570\u636e\u96c6 Spider 2<\/strong>\uff0c\u6392\u884c\u699c TOP1 \u7684\u5b9e\u73b0\u3001\u600e\u4e48\u9009\u62e9\u9002\u5408\u81ea\u5df1\u7684**\u5f00\u6e90\u9879\u76ee<\/p>\n

\u5199\u5728\u524d\u9762<\/h2>\n

\u5efa\u8bae\u8bfb\u8005\u8010\u5fc3\u770b\u5b8c\uff0c\u8fd9\u6837\u5bf9\u672c\u6587\u7684\u80cc\u666f\u4f1a\u6709\u66f4\u6df1\u523b\u7684\u4e86\u89e3\u3002<\/em><\/p>\n

\u5047\u8bbe\u6211\u4eec\u6709\u4e86\u4e00\u4e2a\u7cfb\u7edf\uff0c\u7cfb\u7edf\u5728\u4e0d\u505c\u5730\u8fed\u4ee3\uff0c\u5982\u4f55\u8bf4\u670d\u81ea\u5df1\u548c\u5ba1\u7a3f\u4eba\u7cfb\u7edf\u8d8a\u6765\u8d8a\u597d\uff08\u53d1\u8bba\u6587\u5fc5\u5907\u8981\u7d20\uff09\uff0c\u5982\u4f55\u5411\u5ba2\u6237\u8bc1\u660e\u7cfb\u7edf\u7684\u6709\u6548\u6027\u548c\u53ef\u9760\u6027\u3002\u8fd9\u65f6\u5019\u5c31\u9700\u8981\u6709\u91cf\u5316\u7684\u8bc4\u4f30\u6307\u6807\u3002<\/p>\n

\u4e3e\u4e2a\u4f8b\u5b50\uff1a\u7206\u706b\u7684 DeepSeek-R1\uff0c\u6210\u672c\u662f OpenAI o1 \u7684 1\/10\uff0c\u600e\u4e48\u8bf4\u660e\u4f60\u7684\u6548\u679c\u5dee\u4e0d\u591a\uff0c\u751a\u81f3\u66f4\u597d\u3002\u4ece\u4e2d\uff08DeepSeek-R1 \u53d1\u5e03\uff0c\u6027\u80fd\u5bf9\u6807 OpenAI o1 \u6b63\u5f0f\u7248<\/a>\uff09\u53ef\u4ee5\u770b\u5230\uff0c\u6709 2 \u4e2a\u8981\u7d20\uff0c\u6570\u636e\u96c6<\/strong>\u548c\u8bc4\u4f30\u6307\u6807<\/strong>\uff0c\u7cfb\u5217\u6587\u7ae0\u7684\u7b2c\u4e00\u7bc7\u4ecb\u7ecd\u8fc7\u4e1a\u754c\u516c\u5f00\u7684\u6570\u636e\u96c6\uff0c\u540e\u9762\u4f1a\u4e13\u95e8\u5199\u4e00\u7bc7\u5982\u4f55\u6784\u5efa\u81ea\u5df1\u4e1a\u52a1\u7684\u6570\u636e\u96c6\uff1b\u672c\u7bc7\u5185\u5bb9\u7740\u91cd\u4ecb\u7ecd\u8bc4\u4f30\u6307\u6807\u90e8\u5206\u3002<\/p>\n

\u53e6\u5916\uff0c\u4ee5\u4e0a\u6240\u8bf4\u7684\u90fd\u662f\u5728\u5df2\u7ecf\u77e5\u9053\u7528\u6237\u63d0\u95ee\u5bf9\u5e94\u7684\u6b63\u786e SQL \u8bed\u53e5\uff08Gold SQL\uff09\u7684\u524d\u63d0\u4e0b\u8fdb\u884c\u8bc4\u4f30\uff0c\u5728\u5b9e\u9645\u4ea7\u4e1a\u5e94\u7528\u4e2d\uff0c\u5fc5\u4e0d\u53ef\u5c11\u7684\u662f\u5728\u7ebf\u5b9e\u65f6\u8bc4\u4f30\u7cfb\u7edf<\/strong>\uff0c\u8fd9\u65f6\u4f60\u6ca1\u6cd5\u9884\u6d4b\u7528\u6237\u4f1a\u63d0\u51fa\u4ec0\u4e48\u95ee\u9898\uff0c\u66f4\u6ca1\u6cd5\u9884\u6d4b Gold SQL\uff0c\u4f46\u662f\u9700\u8981\u53ca\u65f6\u611f\u77e5\u7cfb\u7edf\u751f\u6210\u7684 SQL \u662f\u5426\u6b63\u786e\uff0c\u6765\u8c03\u6574\u7cfb\u7edf\u7ed9\u7528\u6237\u66f4\u597d\u7684\u4f53\u9a8c\u3002\u8fd9\u662f\u53e6\u4e00\u4e2a\u8bdd\u9898\uff0c\u9884\u8ba1\u4f1a\u5728\u540e\u9762\u8be6\u7ec6\u5199\u4e00\u7bc7\uff0c\u6709\u5174\u8da3\u7684\u5c0f\u4f19\u4f34\u6b22\u8fce\u7559\u8a00\u8ba8\u8bba\uff0c\u5173\u6ce8\u672c\u516c\u4f17\u53f7\u201cAI\u7231\u597d\u8005\u5c0f\u7ad9\u201c\u3002<\/p>\n

\n

\u672a\u5b8c\u5f85\u7eed\uff0c\u4e0b\u671f\u9884\u544a\u5546\u4e1a\u843d\u5730\u524d\u77bb\u6216\u8005\u5728\u7ebf\u5b9e\u65f6\u8bc4\u4f30\u7cfb\u7edf\uff0c\u66f4\u591a\u66f4\u65b0\u5185\u5bb9\u8bf7\u5173\u6ce8\u672c\u516c\u4f17\u53f7\uff01<\/strong><\/p>\n

\n

\u8bc4\u4ef7\u6307\u6807\u8bbe\u8ba1\u6709\u4e00\u4e2a\u6f5c\u89c4\u5219\uff1a\u6307\u6807\u6570\u91cf\u4e0d\u5b9c\u592a\u591a\uff0c\u4e0d\u7136\u4f1a\u5931\u6389\u91cd\u5fc3\uff08\u4e0d\u8d85\u8fc7 5 \u4e2a\uff09\u3002\u7279\u6b8a\u60c5\u51b5\u4e0b\u6709\u591a\u4e2a\u7ec6\u5206\u6307\u6807\uff0c\u9700\u8981\u5408\u5e76\u4e3a\u4e00\u4e2a\u6307\u6807\uff0c\u4fbf\u4e8e\u6bd4\u8f83\u3002<\/p>\n

\u672c\u6587\u5c06\u603b\u7ed3\u4ecb\u7ecd 5 \u79cd\u6807\u51c6\u8bc4\u4f30\u6307\u6807\u548c\u4e00\u4e9b\u8bc4\u4f30\u6307\u6807\u7684\u63a2\u7d22\u3002<\/p>\n

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

5\u79cd\u6807\u51c6\u8bc4\u4f30\u6307\u6807<\/strong><\/h2>\n

\ud83c\udfaf\u5148\u603b\u7ed3\u4e00\u4e0b\uff0c\u6709\u4e2a\u5370\u8c61<\/strong><\/p>\n

CM\u3001EM\u3001EX<\/strong> \u90fd\u662f\u5728 2018 \u5e74 EMNLP \u4f1a\u8bae\u4e0a\uff0c\u7531\u8036\u9c81\u5927\u5b66\u548c Spider \u6570\u636e\u96c6\u4e00\u8d77\u63d0\u51fa\u7684\uff0c\u5df2\u7ecf\u6210\u4e3a\u4e8b\u5b9e\u4e0a\u7684\u6807\u51c6\u8bc4\u4f30\u6307\u6807\uff1b<\/p>\n

test suite accuracy<\/strong> \u5728 2020 \u5e74 EMNLP \u4f1a\u8bae\u4e0a\u7531\u8036\u9c81\u5927\u5b66\u548c\u4f2f\u514b\u5229\u5927\u5b66\u63d0\u51fa\uff0c\u8bba\u6587\u540d\uff1aSemantic Evaluation for Text-to-SQL with Distilled Test Suites<\/p>\n

VES<\/strong> \u662f\u5728 2023 \u5e74 \u7531\u9999\u6e2f\u5927\u5b66\u548c\u963f\u91cc\u8fbe\u6469\u9662\u5728 NeurIPS \u4f1a\u8bae\u4e0a\u548c Bird \u6570\u636e\u96c6\u4e00\u8d77\u63d0\u51fa\u7684\uff0c\u76ee\u7684\u662f\u5728 EX \u7684\u57fa\u7840\u4e0a\uff0c\u6d4b\u91cf\u67e5\u8be2\u7684\u6548\u7387\uff0c\u53ef\u4ee5\u540c\u65f6\u8bc4\u4f30\u6548\u7387\u548c\u51c6\u786e\u6027\u3002<\/p>\n

\u53ef\u4ee5\u770b\u51fa\u4ee5\u4e0a 5 \u4e2a\u6307\u6807\u7684\u6743\u5a01\u6027\uff0c\u9047\u5230 text2sql \u76f8\u5173\u7684\u8bba\u6587\u3001\u4ea7\u54c1\u3001\u6587\u7ae0\uff0c\u5148\u76ef\u7740\u8fd9\u51e0\u4e2a\u6307\u6807\u770b\u51c6\u6ca1\u9519\u3002\u5173\u4e8e\u8fd9\u4e9b\u6307\u6807\u7684\u5f00\u6e90\u4ee3\u7801\uff0c\u5747\u5217\u5728\u4e0b\u6587\u7684\u8be6\u7ec6\u4ecb\u7ecd\u91cc\u9762\uff0c\u8bf7\u6ce8\u610f\u67e5\u770b\u3002<\/p>\n

\u4e0b\u6587\u7684\u516c\u5f0f\u548c\u5b9a\u4e49\u539f\u6587\uff0c\u5747\u53ef\u5728\u4e0b\u9762\u7684\u8bba\u6587\u91cc\u627e\u5230\uff1a<\/p>\n

\n

Spider 2018<\/strong>\uff1ahttps:\/\/aclanthology.org\/D18-1425\/<\/a><\/p>\n

Semantic Evaluation for Text-to-SQL with Distilled Test Suites<\/strong>\uff1ahttps:\/\/arxiv.org\/pdf\/2010.02840<\/a><\/p>\n

Bird 2023<\/strong>\uff1ahttps:\/\/arxiv.org\/pdf\/2305.03111<\/a><\/p>\n<\/blockquote>\n

\ud83c\udfaf\u57fa\u4e8e\u5185\u5bb9\u5339\u914d\u7684\u6307\u6807\uff1aCM\u3001EM\u3001test suite accuracy<\/strong><\/p>\n

\u8fd9\u91cc\u628a CM \u548c EM \u653e\u5230\u4e00\u7c7b\uff0c\u56e0\u4e3a\u90fd\u662f\u57fa\u4e8e\u5bf9 SQL \u8bed\u53e5\u548c\u7ed3\u6784\u7684\u5206\u6790\uff0c<\/p>\n

\u8be6\u60c5\u5982\u4e0b\uff1a<\/p>\n

\u7ec4\u4ef6\u5339\u914d\uff08**<\/strong>Component Matching, CM\uff09**<\/p>\n

\u4e5f\u53eb partial_match\uff0c\u8bc4\u4f30\u811a\u672c\u8bf7\u67e5\u770b\u5f00\u6e90\u4ee3\u7801\uff1a<\/p>\n

\n

CM\uff1a<\/strong>https:\/\/github.com\/taoyds\/spider\/blob\/master\/evaluation.py<\/a><\/p>\n<\/blockquote>\n

\u4e3a\u4e86\u4e86\u89e3\u6a21\u578b\u5728\u4e0d\u540c SQL \u7ec4\u4ef6\u4e0a\u7684\u6027\u80fd\uff0c\u8fdb\u884c\u6bcf\u4e2a\u90e8\u5206\u8bc4\u4f30\u3002\u901a\u8fc7\u4f7f\u7528 F1 \u5206\u6570\u6d4b\u91cf\u9884\u6d4b SQL \u7ec4\u4ef6\uff08SELECT\u3001WHERE\u3001GROUP BY\u3001ORDER BY \u548c KEYWORDS\uff09\u4e0e\u771f\u5b9e SQL \u7ec4\u4ef6\uff08GROUP BY\u3001ORDER BY \u548c KEYWORDS\uff09\u4e4b\u95f4\u7684\u7cbe\u786e\u5339\u914d\uff0c\u8bc4\u4f30\u6587\u672c\u5230 SQL \u7cfb\u7edf\u7684\u6027\u80fd\u3002\u6bcf\u4e2a\u7ec4\u4ef6\u88ab\u5206\u89e3\u4e3a\u82e5\u5e72\u5b50\u7ec4\u4ef6\u96c6\uff0c\u5e76\u5728\u8003\u8651\u65e0\u987a\u5e8f\u9650\u5236\u7684 SQL \u7ec4\u4ef6\u7684\u60c5\u51b5\u4e0b\uff0c\u6bd4\u8f83\u662f\u5426\u5b8c\u5168\u5339\u914d\u3002<\/p>\n

\u7cbe\u786e\u5339\u914d\uff08Exact Matching, EM\uff09<\/strong><\/p>\n

\u4e5f\u53eb Exact Set Match without Values\uff0c\u8bc4\u4f30\u811a\u672c\u8bf7\u67e5\u770b\u5f00\u6e90\u4ee3\u7801\uff1a<\/p>\n

\n

EM\uff1a<\/strong>https:\/\/github.com\/taoyds\/spider\/blob\/master\/evaluation.py<\/a><\/p>\n

EM\uff1a<\/strong>https:\/\/github.com\/PaddlePaddle\/Research\/blob\/master\/NLP\/Text2SQL-BASELINE\/tools\/evaluation\/text2sql_evaluation.py<\/a><\/p>\n<\/blockquote>\n

\u9996\u5148\u6309\u7167 CM \u63cf\u8ff0\u7684\u5bf9 SQL \u5b50\u53e5\u8fdb\u884c\u8bc4\u4f30\u3002\u53ea\u6709\u5f53\u6240\u6709\u7ec4\u4ef6\u90fd\u6b63\u786e\u65f6\uff0c\u9884\u6d4b\u7684\u67e5\u8be2\u624d\u662f\u6b63\u786e\u7684\u3002\u6bcf\u4e2a\u5b50\u53e5\u4e2d\u662f\u8fdb\u884c\u96c6\u5408\u6bd4\u8f83\uff0c\u6240\u4ee5\u8fd9\u79cd\u7cbe\u786e\u5339\u914d\u5ea6\u91cf\u53ef\u4ee5\u5904\u7406 \u201c\u987a\u5e8f\u95ee\u9898\u201d\u3002<\/p>\n

test suite accuracy<\/strong><\/p>\n

\u8bc4\u4f30\u811a\u672c\u8bf7\u67e5\u770b\u5f00\u6e90\u4ee3\u7801\uff1a<\/p>\n

\n

https:\/\/github.com\/taoyds\/test-suite-sql-eval<\/a><\/p>\n<\/blockquote>\n

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

\u9488\u5bf9 EM \u7684 False Negative \u95ee\u9898\uff08\u5de6\u4e0b\u89d2\uff09\u548c EX \u7684 False Positive \u95ee\u9898\uff08\u53f3\u4e0a\u89d2\uff09\uff0c\u901a\u8fc7\u8fd1\u90bb\u67e5\u8be2\u8861\u91cf\u6d4b\u8bd5\u5957\u4ef6\u4ee3\u7801\u8986\u76d6\u7387\u3002\u5c06\u76ee\u6807\u8bbe\u5b9a\u4e3a\u627e\u5230\u80fd\u533a\u5206\u6240\u6709\u8fd1\u90bb\u67e5\u8be2\u7684\u5c0f\u578b\u6d4b\u8bd5\u5957\u4ef6\u3002\u8fd1\u90bb\u67e5\u8be2\u901a\u8fc7\u4fee\u6539\u9ec4\u91d1\u67e5\u8be2\u7684\u6574\u6570\u3001\u5b57\u7b26\u4e32\u3001\u6bd4\u8f83\u8fd0\u7b97\u7b26\u7b49\u65b9\u9762\u751f\u6210\uff0c\u5177\u6709\u8bed\u4e49\u4e0d\u540c\u4e14\u80fd\u68c0\u9a8c\u6d4b\u8bd5\u5957\u4ef6\u8986\u76d6\u5ea6\u7684\u7279\u70b9\u3002<\/p>\n

\u8fd9\u91cc\u4e0d\u5c55\u5f00\u8bb2\u6280\u672f\u7ec6\u8282\uff0c\u6709\u5174\u8da3\u7684\u5c0f\u4f19\u4f34\u53ef\u4ee5\u901a\u8fc7\u94fe\u63a5\uff08https:\/\/arxiv.org\/abs\/2010.02840\uff09\u4e0b\u8f7d\u5168\u6587\u7814\u7a76\u3002\u53ef\u4ee5\u7b80\u5355\u7406\u89e3\u4e3a\u5bf9<\/a> EM \u6307\u6807\u7684\u4f18\u5316\u3002<\/p>\n

\ud83c\udfaf\u57fa\u4e8e\u6267\u884c\u7ed3\u679c\u7684\u6307\u6807\uff1aEX\u3001VES<\/strong><\/p>\n

\u8fd9\u91cc\u628a EX \u548c VES \u653e\u5230\u4e00\u7c7b\uff0c\u56e0\u4e3a\u90fd\u662f\u57fa\u4e8e\u6267\u884c\u7ed3\u679c\u3002\u8fd9\u7c7b\u6307\u6807\u6709\u4e00\u4e2a\u6bd4\u8f83\u9ebb\u70e6\u7684\u4e8b\u60c5\uff0c\u5c31\u662f\u9700\u8981\u6709\u5b9e\u9645\u7684\u6570\u636e\u5e93\u548c\u6570\u636e\u6765\u6267\u884c\u8bed\u53e5\uff0c\u8be6\u60c5\u5982\u4e0b\uff1a<\/p>\n

\u7ec4\u4ef6\u5339\u914d\uff08**<\/strong>Execution Accuracy, EX\uff09**<\/p>\n

\u8bc4\u4f30\u811a\u672c\u8bf7\u67e5\u770b\u5f00\u6e90\u4ee3\u7801\uff1a<\/p>\n

\n

EX\uff1a<\/strong>https:\/\/github.com\/AlibabaResearch\/DAMO-ConvAI\/blob\/main\/bird\/llm\/src\/evaluation.py<\/a><\/p>\n<\/blockquote>\n

\u901a\u8fc7\u5728\u76f8\u5e94\u6570\u636e\u5e93\u4e2d\u6267\u884c\u9884\u6d4b\u7684 SQL \u67e5\u8be2\uff0c\u5e76\u5c06\u6267\u884c\u7ed3\u679c\u4e0e\u57fa\u672c\u771f\u5b9e\u67e5\u8be2\u6240\u83b7\u7ed3\u679c\u8fdb\u884c\u6bd4\u8f83\uff0c\u6765\u8861\u91cf\u8be5\u67e5\u8be2\u7684\u6b63\u786e\u6027\u3002\u516c\u5f0f\u5982\u4e0b\uff1a<\/p>\n

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

\u6709\u6548\u6548\u7387\u5206\u6570\uff08**<\/strong>Valid Efficiency Score, VES\uff09**<\/p>\n

\u8bc4\u4f30\u811a\u672c\u8bf7\u67e5\u770b\u5f00\u6e90\u4ee3\u7801\uff1a<\/p>\n

\n

VES\uff1a<\/strong>https:\/\/github.com\/AlibabaResearch\/DAMO-ConvAI\/blob\/main\/bird\/llm\/src\/evaluation_ves.py<\/a><\/p>\n<\/blockquote>\n

\u6267\u884c\u7ed3\u679c\u4e0e\u771f\u5b9e\u7ed3\u679c\u5b8c\u5168\u4e00\u81f4\uff0c\u540c\u65f6\u52a0\u4e0a\u5bf9\u6267\u884c\u6548\u7387\uff08\u8fd0\u884c\u65f6\u95f4\uff09\u7684\u6307\u6807<\/p>\n

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

\u76f8\u5bf9\u6548\u7387\u51fd\u6570 R (\u30fb) \u7528\u4e8e\u8861\u91cf\u9884\u6d4b SQL \u4e0e\u57fa\u51c6 SQL \u7684\u6267\u884c\u6548\u7387\u6bd4\u503c\uff0c\u8be5\u6307\u6807\u901a\u8fc7\u5f15\u5165\u673a\u5668\u72b6\u6001\u4e0d\u786e\u5b9a\u6027\u6a21\u578b\u6765\u6d88\u9664\u786c\u4ef6\u73af\u5883\u6ce2\u52a8\u7684\u5f71\u54cd\uff1b<\/p>\n

\u7edd\u5bf9\u6548\u7387\u51fd\u6570 E (\u30fb) \u5219\u57fa\u4e8e\u7279\u5b9a\u73af\u5883\u4e0b\u7684\u5b9e\u6d4b\u6570\u636e\uff0c\u76f4\u63a5\u53cd\u6620\u5355\u6761 SQL \u7684\u6267\u884c\u6548\u80fd\uff1b<\/p>\n

\u7279\u522b\u5730\uff0c\u6211\u4eec\u5f15\u5165\u5e73\u65b9\u6839\u53d8\u6362\u673a\u5236\uff0c\u901a\u8fc7\u6570\u5b66\u65b9\u6cd5\u6291\u5236\u56e0\u968f\u673a\u56e0\u7d20\u5bfc\u81f4\u7684\u5f02\u5e38\u5feb \/ \u6162\u6267\u884c\u6848\u4f8b\u5bf9\u6574\u4f53\u8bc4\u4f30\u7684\u5e72\u6270\uff1b<\/p>\n

\u672c\u7cfb\u7edf\u5f53\u524d\u4e3b\u8981\u4ee5\u8fd0\u884c\u65f6\u95f4\u4f5c\u4e3a\u6838\u5fc3\u6548\u7387\u6307\u6807\uff0c\u8be5\u6307\u6807\u53ef\u6269\u5c55\u652f\u6301\u541e\u5410\u91cf\u3001\u5185\u5b58\u5360\u7528\u7b49\u5355\u9879\u6216\u590d\u5408\u8bc4\u4f30\u7ef4\u5ea6\u3002<\/p>\n

\u63a2\u7d22\u6027\u8bc4\u4f30\u6307\u6807<\/strong><\/h2>\n

\ud83c\udfafLLM-based \u6253\u5206\u673a\u5236\u8bc4\u4f30\u6307\u6807<\/strong><\/p>\n

\u8fd9\u79cd\u6307\u6807\u7684\u601d\u8def\u662f\u501f\u52a9\u63d0\u793a\u8bcd\u5de5\u7a0b\uff08Prompt Engineering\uff09\u3001\u68c0\u7d22\u589e\u5f3a\u751f\u6210\uff08Retrieval-Augmented Generation, RAG\uff09\u7b49\u6280\u672f\uff0c\u7528\u5927\u6a21\u578b\u7684\u63a8\u7406\u5206\u6790\u80fd\u529b\uff0c\u6765\u6253\u5206\uff0c\u4f18\u70b9\u662f\u9002\u7528\u6027\u5f3a\u800c\u4e14\u53ef\u4ee5\u4e0d\u65ad\u4f18\u5316\uff0c\u7f3a\u70b9\u662f\u4e0d\u7a33\u5b9a\uff0c\u4e14\u516c\u5171\u8ba4\u53ef\u5ea6\u4e0d\u9ad8\uff0c\u653e\u5728\u8bba\u6587\u91cc\u4e0d\u592a\u5408\u9002\u3002<\/p>\n

Prompt \u6a21\u677f\u793a\u610f\u5982\u4e0b\uff1a<\/p>\n

\"image.png\"<\/p>\n

\ud83c\udfaf\u8bad\u7ec3\u65b0\u7684\u6a21\u578b\u6765\u8bc4\u4f30<\/strong><\/p>\n

\u8fd9\u4e2a\u6307\u6807\u662f\u6700\u8fd1\u624d\u51fa\u6765\u7684\uff0c\u7531\u5b57\u8282\u56e2\u961f\u63d0\u51fa\uff0c\u4e5f\u4f5c\u4e3a session \u8bdd\u9898\u51fa\u73b0\u5728 GTC 2025 \u7684\u8ba8\u8bba\u4e2d\uff0c\u524d\u4e24\u5929\u6211\u7684\u4e00\u7bc7\u63a8\u6587\u6709\u8be6\u60c5\uff1aGTC25 | \u5b57\u8282\u53c2\u4f1a\u6587\u7ae0 Text2SQL \u65b0\u8bc4\u4f30\u6307\u6807-FuncEvalGMN<\/a><\/p>\n

\u603b\u7ed3\u6765\u8bf4\uff0c\u4e5f\u662f\u57fa\u4e8e\u7ec4\u4ef6\u7684\u65b9\u5f0f\uff0c\u4e0d\u9700\u8981\u6267\u884c\u7ed3\u679c\u548c\u5927\u91cf\u7684\u6570\u636e\u5e93\u51c6\u5907\u3002\u601d\u8def\u57fa\u4e8e\u56fe\u548c\u81ea\u6ce8\u610f\u673a\u5236\u8bad\u7ec3\u4e86\u65b0\u7684\u6a21\u578b\u7528\u4e8e\u8bc4\u4f30<\/strong>\uff0c\u8fd9\u4e2a\u601d\u8def\u53ef\u4ee5\u6301\u7eed\u5173\u6ce8\uff0c\u6709\u53ef\u80fd\u662f\u672a\u6765\u7684\u65b9\u5411\uff0c\u7f3a\u70b9\u662f\u5c1a\u672a\u5f97\u5230\u884c\u4e1a\u666e\u904d\u8ba4\u53ef\u548c\u5927\u89c4\u6a21\u7684\u5e94\u7528\u3002<\/p>\n

\u627e\u4e00\u4e9b\u6392\u884c\u699c\u548c\u8bba\u6587\u770b\u770b<\/strong><\/h2>\n

\u627e\u4e00\u4e9b\u6392\u884c\u699c\u548c\u8bba\u6587\u770b\u770b\uff0c\u4ed6\u4eec\u90fd\u7528\u54ea\u4e9b\u6307\u6807<\/p>\n

Bird<\/strong> \u4f7f\u7528\u7684\u662f EX \u548c VES \u6307\u6807\uff1ahttps:\/\/bird-bench.github.io\/<\/a><\/p>\n

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

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

Spider<\/strong> \u4f7f\u7528 EX \u548c EM\uff1ahttps:\/\/yale-lily.github.io\/spider<\/a><\/p>\n

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

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

2023 TPAMI \u8bba\u6587<\/strong>\uff1aA Heterogeneous Graph to Abstract Syntax Tree Framework for Text-to-SQL<\/em><\/p>\n

\u94fe\u63a5\uff1ahttps:\/\/rhythmcao.github.io\/publication\/2023-hg2ast\/2023-hg2ast.pdf<\/a><\/p>\n

\u7528\u4e86\u5982\u4e0b\u6307\u6807\uff1a<\/p>\n

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

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

\u7ed3\u8bba<\/strong><\/h2>\n

EX \u7528\u5f97\u6700\u591a\uff0c\u8bba\u6587\u548c\u6392\u884c\u699c\u91cc\u6700\u7ecf\u5e38\u89c1\u5230\uff1bEM \u548c test suite accuracy \u6b21\u4e4b\uff0cCM \u548c VES \u5728\u7279\u6b8a\u573a\u666f\u4e0b\u4f7f\u7528\u3002\u5176\u4ed6\u63a2\u7d22\u6027\u6307\u6807\uff0c\u6682\u65f6\u6ca1\u89c1\u5230\u5728\u6392\u884c\u699c\u6216\u8005\u5176\u4ed6\u8bba\u6587\u91cc\u4f7f\u7528\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

\u8fd9\u662f Text2SQL \u7cfb\u5217\u7b2c\u56db\u7bc7\uff0c\u4e4b\u524d\u5206\u522b\u5199\u4e86\u6570\u636e\u96c6\u3001\u5f53\u4e0b SOTA \u7684 LLM-based \u7684\u5b9e\u73b0\u65b9\u5f0f\u53ca\u8036\u9c81\u5927\u5b66 2024 \u5e74\u6570\u636e\u96c6 \u2026 …<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ngg_post_thumbnail":0},"categories":[288],"tags":[],"_links":{"self":[{"href":"\/index.php?rest_route=\/wp\/v2\/posts\/9488"}],"collection":[{"href":"\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=9488"}],"version-history":[{"count":1,"href":"\/index.php?rest_route=\/wp\/v2\/posts\/9488\/revisions"}],"predecessor-version":[{"id":9489,"href":"\/index.php?rest_route=\/wp\/v2\/posts\/9488\/revisions\/9489"}],"wp:attachment":[{"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=9488"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=9488"},{"taxonomy":"post_tag","embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=9488"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}