﻿{"id":2092379,"date":"2026-05-21T22:32:19","date_gmt":"2026-05-21T14:32:19","guid":{"rendered":"http:\/\/www.nlpir.org\/wordpress\/?p=2092379"},"modified":"2026-05-21T22:32:19","modified_gmt":"2026-05-21T14:32:19","slug":"kdd-2026%ef%bc%9a%e4%bb%8e%e6%89%93%e5%88%86%e5%88%b0%e7%ac%a6%e5%8f%b7%e5%8c%96%e5%a4%b1%e8%b4%a5%e8%af%8a%e6%96%ad%ef%bc%8c%e8%ae%a9%e5%a4%a7%e6%a8%a1%e5%9e%8b%e6%8e%a8%e7%90%86%e5%8f%af%e8%af%8a","status":"publish","type":"post","link":"http:\/\/www.nlpir.org\/wordpress\/2026\/05\/21\/kdd-2026%ef%bc%9a%e4%bb%8e%e6%89%93%e5%88%86%e5%88%b0%e7%ac%a6%e5%8f%b7%e5%8c%96%e5%a4%b1%e8%b4%a5%e8%af%8a%e6%96%ad%ef%bc%8c%e8%ae%a9%e5%a4%a7%e6%a8%a1%e5%9e%8b%e6%8e%a8%e7%90%86%e5%8f%af%e8%af%8a\/","title":{"rendered":"KDD 2026\uff1a\u4ece\u6253\u5206\u5230\u7b26\u53f7\u5316\u5931\u8d25\u8bca\u65ad\uff0c\u8ba9\u5927\u6a21\u578b\u63a8\u7406&#8221;\u53ef\u8bca\u65ad\u3001\u53ef\u89e3\u91ca&#8221;"},"content":{"rendered":"\n<p>\u5317\u4eac\u7406\u5de5\u5927\u5b66\u3001\u4e2d\u5173\u6751\u5b66\u9662\u3001\u4e2d\u56fd\u79d1\u5b66\u9662\u81ea\u52a8\u5316\u7814\u7a76\u6240<\/p>\n\n\n\n<p class=\"has-text-align-center\"><img decoding=\"async\" style=\"width: 650px;\" src=\"https:\/\/mmbiz.qpic.cn\/sz_mmbiz_jpg\/yozlYZeLGPeOklH98bTiaPsRLicjr1opRhicr1JkCXVCtxD9pabDT4s2q6sauIaBq12VJXGxRW5mnfSZLSE0LNIzRHuSVNUO6fA2qwMuZYdSuE\/640?wx_fmt=jpeg&amp;from=appmsg&amp;watermark=1&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1#imgIndex=1\" alt=\"\u56fe\u7247\"><\/p>\n\n\n\n<p><strong>KDD\u4f1a\u8bae\u7b80\u4ecb<\/strong><\/p>\n\n\n\n<p>KDD\uff08ACM SIGKDD Conference on Knowledge Discovery and Data Mining\uff09\u662f\u6570\u636e\u6316\u6398\u4e0e\u77e5\u8bc6\u53d1\u73b0\u9886\u57df\u7684\u56fd\u9645\u9876\u7ea7\u5b66\u672f\u4f1a\u8bae\uff0c\u7531 ACM \u65d7\u4e0b\u7684 SIGKDD \u4e13\u59d4\u4f1a\u4e3b\u529e\uff0c\u81ea 1999 \u5e74\u8d77\u6bcf\u5e74\u4e3e\u529e\u4e00\u5c4a\uff0c\u88ab\u4e2d\u56fd\u8ba1\u7b97\u673a\u5b66\u4f1a\uff08CCF\uff09\u5217\u4e3a A \u7c7b\u63a8\u8350\u4f1a\u8bae\u3002<\/p>\n\n\n\n<p>\u4f5c\u4e3a\u6570\u636e\u79d1\u5b66\u4e0e\u5927\u6570\u636e\u751f\u6001\u7684\u98ce\u5411\u6807\uff0cKDD \u4e13\u6ce8\u4e8e\u201c\u4ece\u6d77\u91cf\u6570\u636e\u4e2d\u53d1\u73b0\u6709\u4ef7\u503c\u77e5\u8bc6\u201d\u7684\u5168\u6d41\u7a0b\u7814\u7a76\uff0c\u8bae\u9898\u8986\u76d6\u6570\u636e\u6316\u6398\u7b97\u6cd5\u3001\u673a\u5668\u5b66\u4e60\u3001\u6df1\u5ea6\u5b66\u4e60\u3001\u56fe\u795e\u7ecf\u7f51\u7edc\u3001\u63a8\u8350\u7cfb\u7edf\u53ca\u5f02\u5e38\u68c0\u6d4b\u7b49\u524d\u6cbf\u65b9\u5411\uff0c\u540c\u65f6\u4e5f\u9ad8\u5ea6\u91cd\u89c6\u5728\u751f\u7269\u4fe1\u606f\u3001\u91d1\u878d\u98ce\u63a7\u53ca\u793e\u4f1a\u7f51\u7edc\u7b49\u9886\u57df\u7684\u843d\u5730\u5e94\u7528\u3002<\/p>\n\n\n\n<p>\u8be5\u4f1a\u8bae\u5b66\u672f\u95e8\u69db\u6781\u9ad8\uff0c\u4e3b\u4f1a\u8bba\u6587\u5f55\u7528\u7387\u5e38\u5e74\u7ef4\u6301\u5728 15%~20% \u4e4b\u95f4\uff0c\u4e0e ICML\u3001NeurIPS \u5e76\u5217\u4e3a\u4eba\u5de5\u667a\u80fd\u4e0e\u6570\u636e\u79d1\u5b66\u9886\u57df\u7684\u4e09\u5927\u9876\u5c16\u4f1a\u8bae\u3002<\/p>\n\n\n\n<p><strong>\u8bba\u6587\u5b97\u65e8<\/strong><\/p>\n\n\n\n<p>\u5927\u6a21\u578b (LLM) \u5df2\u7ecf\u80fd\u505a\u51fa\u50cf\u6a21\u50cf\u6837\u7684\u591a\u6b65\u63a8\u7406\u3002\u4f46\u8d8a\u6765\u8d8a\u591a\u7814\u7a76\u53d1\u73b0\u4e00\u4e2a\u5c34\u5c2c\u73b0\u8c61\uff1a<strong>\u7b54\u6848\u5bf9\u4e86\uff0c\u601d\u7ef4\u94fe\u5374\u672a\u5fc5\u6210\u7acb<\/strong>\u2014\u2014\u4e2d\u95f4\u6b65\u9aa4\u8df3\u8dc3\u3001\u9690\u85cf\u5047\u8bbe\u3001\u5ffd\u7565\u8fb9\u754c\u6761\u4ef6\uff0c\u7ed3\u679c&#8221;\u78b0\u5de7\u5bf9&#8221;\u3002\u5728\u533b\u7597\u3001\u91d1\u878d\u3001\u79d1\u7814\u52a9\u7406\u8fd9\u79cd\u9ad8\u98ce\u9669\u573a\u666f\u91cc\uff0c&#8221;\u5076\u5c14\u7b54\u5bf9&#8221;\u548c&#8221;\u8fc7\u7a0b\u53ef\u4fe1&#8221;\u5b8c\u5168\u662f\u4e24\u4ef6\u4e8b\u3002<\/p>\n\n\n\n<p>\u73b0\u6709\u9a8c\u8bc1\u624b\u6bb5\u90fd\u96be\u4ee5\u56de\u7b54&#8221;\u4e3a\u4ec0\u4e48\u9519&#8221;\uff1aAnswer Matching \u53ea\u7ed9 1 bit\uff1bLLM-as-Judge \u4e3b\u89c2\u96be\u5ba1\u8ba1\uff1bReward \/ PRM \u662f\u4e00\u4e2a\u6807\u91cf\uff1b\u5df2\u6709\u7684\u795e\u7ecf\u7b26\u53f7\u65b9\u6cd5\u53c8\u5206\u4e0d\u6e05&#8221;<strong>\u7ffb\u8bd1\u5931\u8d25<\/strong>&#8220;\u548c&#8221;<strong>\u63a8\u7406\u7f3a\u9677<\/strong>&#8220;\u3002<\/p>\n\n\n\n<p>\u6765\u81ea&nbsp;<strong>\u5317\u4eac\u7406\u5de5\u5927\u5b66 NLPIR \u5b9e\u9a8c\u5ba4<\/strong>&nbsp;\u4e0e&nbsp;<strong>\u4e2d\u5173\u6751\u5b66\u9662 \/ \u4e2d\u56fd\u79d1\u5b66\u9662\u81ea\u52a8\u5316\u7814\u7a76\u6240<\/strong>&nbsp;\u7684\u8054\u5408\u7814\u7a76\u56e2\u961f\uff0c\u5728\u8bba\u6587&nbsp;<strong>\u300aSymDiag: Explainable Diagnosis for LLM Reasoning via Neuro-Symbolic Verification\u300b\uff08KDD 2026\uff09<\/strong>\u4e2d\u63d0\u51fa\u4e86\u4e00\u79cd\u5168\u65b0\u7684\u8303\u5f0f\uff1a<strong>\u628a\u63a8\u7406\u9a8c\u8bc1\u91cd\u65b0\u5b9a\u4e49\u4e3a&#8221;\u53ef\u89e3\u91ca\u7684\u5931\u8d25\u8bca\u65ad&#8221;<\/strong>\u2014\u2014\u4e0d\u4ec5\u6307\u51fa&#8221;\u54ea\u4e00\u6b65\u9519&#8221;\uff0c\u66f4\u7ed9\u51fa\u53ef\u88ab\u7b26\u53f7\u6c42\u89e3\u5668\u72ec\u7acb\u6821\u9a8c\u7684\u8bc1\u636e\uff0c\u5e76\u636e\u6b64\u9a71\u52a8\u8fed\u4ee3\u4fee\u590d\u3002<\/p>\n\n\n\n<p><strong>\u4ece&#8221;\u6253\u5206&#8221;\u5230&#8221;\u8bca\u65ad&#8221;<\/strong><\/p>\n\n\n\n<p>\u8fd9\u9879\u5de5\u4f5c\u7684\u51fa\u53d1\u70b9\u5f88\u6734\u7d20\uff1a\u591a\u8f6e\u4e8b\u5b9e\u6838\u67e5\u65f6\uff0c<strong>\u6bcf\u4e00\u6b65\u8bc1\u636e\u5e76\u4e0d\u662f\u7b80\u5355\u5730\u5806\u5728\u4e00\u8d77<\/strong>\uff0c\u800c\u662f\u6709<strong>\u5c42\u6b21\u3001\u4f9d\u8d56\u3001\u7ea6\u675f<\/strong>\u5173\u7cfb\u3002\u5149\u770b\u6700\u7ec8\u7b54\u6848\u3001\u6216\u8005\u8bf7\u53e6\u4e00\u4e2a LLM \u6765\u6253\u5206\uff0c\u90fd\u770b\u4e0d\u89c1\u8fd9\u4e9b\u7ed3\u6784\u3002<\/p>\n\n\n\n<p class=\"has-text-align-center\"><img decoding=\"async\" src=\"https:\/\/mmbiz.qpic.cn\/sz_mmbiz_jpg\/yozlYZeLGPfamBQl78d0NP8olr3wLoxZhkHTu37gkVTbGKBY9zxpNgzbfrokicI9vDvUkyaTtF5FbuuUibticSs74dNhGHd3vPUl1zzYedLU9c\/640?wx_fmt=jpeg&amp;from=appmsg&amp;watermark=1&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1#imgIndex=5\" alt=\"\u56fe\u7247\" style=\"width: 650px;\"><\/p>\n\n\n\n<p>\u8bba\u6587\u501f\u9274\u795e\u7ecf\u7b26\u53f7\u65b9\u6cd5\u7684\u601d\u8def\uff0c\u628a CoT \u7f16\u8bd1\u6210\u7edf\u4e00\u7b7e\u540d\u4e0b\u7684<strong>\u7b26\u53f7\u7a0b\u5e8f<\/strong>\uff0c\u518d\u4ea4\u7ed9 SWI-Prolog \u8fd9\u6837\u7684\u6c42\u89e3\u5668\u53bb\u505a<strong>\u6b65\u7ea7 SAT \/ \u8574\u542b\u6821\u9a8c<\/strong>\u3002<\/p>\n\n\n\n<p>\u8fd9\u610f\u5473\u7740\u2014\u2014\u5f53\u4f60\u8bf4&#8221;\u8fd9\u4e00\u6b65\u9519\u4e86&#8221;\uff0c<strong>\u4f60\u80fd\u62ff\u51fa\u4e00\u4efd\u53cd\u4f8b \/ Unsat \u6838\u5fc3 \/ \u7f3a\u5931\u524d\u63d0<\/strong>\uff0c\u800c\u4e0d\u662f\u6a21\u7cca\u7684&#8221;\u6211\u89c9\u5f97\u4e0d\u592a\u5bf9&#8221;\u3002<\/p>\n\n\n\n<p><strong>\u5173\u952e\u6311\u6218\uff1a\u662f\u771f\u9519\uff0c\u8fd8\u662f\u7ffb\u8bd1\u566a\u58f0\uff1f<\/strong><\/p>\n\n\n\n<p>\u628a\u81ea\u7136\u8bed\u8a00\u7ffb\u6210\u5f62\u5f0f\u5316\u8bed\u8a00\u518d\u6c42\u89e3\uff0c\u8fd9\u6761\u8def\u5176\u5b9e<strong>\u4e0d\u65b0<\/strong>\u3002Logic-LM\u3001SymbCoT\u3001Aristotle \u90fd\u5c1d\u8bd5\u8fc7\u3002\u4f46<strong>\u6240\u6709\u8fd9\u6761\u8def\u4e0a\u7684\u65b9\u6cd5\u90fd\u88ab\u540c\u4e00\u4e2a\u95ee\u9898\u56f0\u6270<\/strong>\uff1a\u5f53\u6c42\u89e3\u5668\u62a5&#8221;\u8fdd\u89c4&#8221;\u65f6\uff0c\u6211\u4eec\u5230\u5e95\u662f\u6293\u4f4f\u4e86\u4e00\u4e2a\u771f\u5b9e\u63a8\u7406\u7f3a\u9677\uff1f\u8fd8\u662f\u53ea\u662f<strong>\u7ffb\u8bd1\u8fc7\u7a0b\u4ea7\u751f\u7684\u566a\u58f0<\/strong>\uff1f<\/p>\n\n\n\n<p>\u5f53\u524d\u7684\u6838\u5fc3\u96be\u70b9\u662f\u5982\u679c\u4e0d\u80fd\u533a\u5206\u4e8c\u8005\uff0c\u6240\u6709\u53cd\u9988\u90fd\u4f1a\u88ab\u6c61\u67d3\u2014\u2014\u9a8c\u8bc1\u5668\u8d8a\u4e25\u683c\uff0c\u53cd\u800c\u9519\u5f97\u8d8a\u79bb\u8c31\u3002\u4e00\u4e2a\u597d\u7684\u8bca\u65ad\u7cfb\u7edf\uff0c\u5fc5\u987b\u5728<strong>\u7ffb\u8bd1\u9519\u8bef (TranslationError)&nbsp;<\/strong>\u4e0e&nbsp;<strong>\u63a8\u7406\u9519\u8bef (ReasoningError)&nbsp;<\/strong>\u4e4b\u95f4\u753b\u51fa\u4e00\u6761\u53ef\u9760\u7684\u754c\u3002<\/p>\n\n\n\n<p>SymDiag \u7684\u7834\u5c40\u70b9\u662f<strong>\u81ea\u5ba1\u8ba1\u673a\u5236 (Self-Auditor)<\/strong>\uff1a\u5bf9\u540c\u4e00\u6bb5 CoT \u540c\u65f6\u751f\u6210\u4e24\u4efd\u72ec\u7acb\u7684\u7b26\u53f7\u7f16\u7801\u2014\u2014<strong>Formal Translation&nbsp;<\/strong>\u8d70\u6807\u51c6\u7ffb\u8bd1\uff0c<strong>Critical Restatement<\/strong>&nbsp;\u505a\u66f4\u4e25\u683c\u7684\u590d\u8ff0\u518d\u7f16\u8bd1\u3002\u4e24\u6761\u8def\u5f84\u76f8\u4e92&#8221;\u4f5c\u8bc1&#8221;\uff0c\u80fd\u66b4\u9732\u539f\u6587\u91cc\u88ab\u63a9\u76d6\u7684\u9690\u85cf\u5047\u8bbe\u3002<\/p>\n\n\n\n<p>\u5982\u679c\u67d0\u4e2a&#8221;\u8fdd\u89c4&#8221;\u5728\u6700\u5c0f\u5316\u3001\u6587\u672c\u4e00\u81f4\u7684\u6539\u5199\u4e0b\u5c31\u6d88\u5931\u4e86\u2014\u2014\u90a3\u5b83\u662f<strong>\u7ffb\u8bd1\u9519\u8bef<\/strong>\uff1b\u5426\u5219\u624d\u4f1a\u88ab\u8ba4\u5b9a\u4e3a<strong>\u771f\u6b63\u7684\u63a8\u7406\u9519\u8bef<\/strong>\u3002\u53ea\u6709\u901a\u8fc7\u5ba1\u8ba1\u7684\u72b6\u6001\u624d\u4f1a\u8fdb\u5165\u4e0b\u4e00\u9636\u6bb5\u7684\u6b65\u7ea7\u9a8c\u8bc1\u3002<\/p>\n\n\n\n<p><strong>SymDiag \u6846\u67b6\uff1a\u56db\u4ef6\u5957\u6784\u6210\u7684\u8bca\u65ad\u95ed\u73af<\/strong><\/p>\n\n\n\n<p class=\"has-text-align-center\"><img decoding=\"async\" src=\"https:\/\/mmbiz.qpic.cn\/sz_mmbiz_jpg\/yozlYZeLGPdLianop7uw3os5Zlq9Lc47Qc0Ojyn0OEtSUTblI0OiaN6WkVZnG3SsqlpYAcX6gDV2H4Eg8MefAcpeOwy6pjy7dcPtRgI7LGEdQ\/640?wx_fmt=jpeg&amp;from=appmsg&amp;watermark=1&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1#imgIndex=8\" alt=\"\u56fe\u7247\" style=\"width: 650px;\"><\/p>\n\n\n\n<p><strong>\u2460 \u53cc\u5206\u652f\u795e\u7ecf\u7b26\u53f7\u751f\u6210\u5668<\/strong>\uff1a\u540c\u4e00\u6761 CoT \u88ab\u7f16\u8bd1\u4e3a\u4e24\u4efd\u72ec\u7acb\u7684\u5f62\u5f0f\u5316\u7a0b\u5e8f\uff0c<strong>\u4e24\u6761\u8def\u5f84\u76f8\u4e92\u6821\u9a8c<\/strong>\uff0c\u964d\u4f4e\u5355\u70b9\u7ffb\u8bd1\u504f\u5dee\u3002<\/p>\n\n\n\n<p><strong>\u2461 Self-Auditor\uff08\u6838\u5fc3\u521b\u65b0\uff09<\/strong>\uff1a\u7528\u8de8\u5206\u652f\u4e00\u81f4\u6027\u68c0\u67e5 + \u8f7b\u91cf\u7b26\u53f7\u6b63\u5219\u5316\u6d4b\u8bd5\uff0c\u628a\u6bcf\u4e00\u5904&#8221;\u8fdd\u89c4&#8221;\u660e\u786e\u5f52\u56e0\u5230<strong>TranslationError \u6216 ReasoningError<\/strong>\u3002<\/p>\n\n\n\n<p><strong>\u2462 \u6b65\u7ea7\u7b26\u53f7\u9a8c\u8bc1<\/strong>\uff1a\u5bf9\u6bcf\u4e00\u6b65\u72b6\u6001 Si = {Pi, Ii, Ci}\uff08\u524d\u63d0 \/ \u63a8\u65ad \/ \u7ea6\u675f\uff09\u6267\u884c SAT \u4e0e\u5c40\u90e8\u8574\u542b\u68c0\u67e5\uff0c\u8f93\u51fa\u53ef\u6821\u9a8c\u8bc1\u636e\u3002<\/p>\n\n\n\n<p><strong>\u2463 \u8bca\u65ad\u5f15\u5bfc\u7684\u4fee\u590d<\/strong>\uff1a\u6839\u636e\u5931\u8d25\u7684\u5f71\u54cd\u8303\u56f4 (scope) \u9009\u62e9&#8221;\u5c40\u90e8\u8865\u4e01&#8221;\u6216&#8221;\u5168\u5c40\u91cd\u5199&#8221;\uff0c\u5f62\u6210\u8bca\u65ad \u2192 \u4fee\u590d \u2192 \u518d\u8bca\u65ad\u7684\u95ed\u73af\u3002<\/p>\n\n\n\n<p>\u4e00\u53e5\u8bdd\u603b\u7ed3\uff1aSymDiag\u00a0<strong>\u4e0d\u662f\u66f4\u597d\u7684 verifier\uff0c\u4e5f\u4e0d\u662f\u66f4\u597d\u7684 reward<\/strong>\uff0c\u800c\u662f\u4e00\u4e2a\u53ef\u89e3\u91ca\u7684\u795e\u7ecf\u7b26\u53f7\u8bca\u65ad\u7cfb\u7edf\u2014\u2014\u5b83\u5b9a\u4f4d\u9519\u8bef\u3001\u5f52\u56e0\u539f\u56e0\uff0c\u5e76\u4ee5\u53ef\u88ab\u72ec\u7acb\u6821\u9a8c\u7684\u8bc1\u636e\u9a71\u52a8\u4fee\u590d\u3002<\/p>\n\n\n\n<p><strong>\u5b9e\u9a8c\uff1a\u8de8\u56db\u5927\u9886\u57df\u7684\u5168\u9762\u9a8c\u8bc1<\/strong><\/p>\n\n\n\n<p>\u7814\u7a76\u56e2\u961f\u6784\u5efa\u4e86\u4e00\u4e2a\u8986\u76d6<strong>\u6570\u5b66 \/ \u903b\u8f91 \/ \u79d1\u5b66 \/ \u901a\u7528\u63a8\u7406<\/strong>\u7684\u7edf\u4e00\u8bca\u65ad\u6570\u636e\u96c6\uff0c\u81ea\u52a8\u6784\u5efa\u6837\u672c\u8fbe<strong>\u00a0437,792 \u6761<\/strong>\uff0c\u5e76\u4eba\u5de5\u5ba1\u6838\u51fa<strong>\u00a0240 \u6761<\/strong>\u9ad8\u8d28\u91cf\u9ec4\u91d1\u96c6\u3002<\/p>\n\n\n\n<p class=\"has-text-align-center\"><img decoding=\"async\" src=\"https:\/\/mmbiz.qpic.cn\/mmbiz_jpg\/yozlYZeLGPfD5XTZ92PYTjv5FO26667MZ0owKgRBq37LqrFdbNJs8amYar3ocAhSQapn2z7oq5tKzdovRAQ21WZXJqvL8qy2icHblBDB0TWA\/640?wx_fmt=jpeg&amp;from=appmsg&amp;watermark=1&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1#imgIndex=10\" alt=\"\u56fe\u7247\" style=\"width: 650px;\"><\/p>\n\n\n\n<p><strong>\u4e0d\u5fe0\u5b9e\u63a8\u7406\u68c0\u6d4b\uff1a<\/strong>SymDiag \u5728\u6240\u6709<strong>\u00a08 \u4e2a benchmark\u00a0<\/strong>\u4e0a\u5747\u53d6\u5f97 F1 \u6700\u4f18\uff0c\u6574\u4f53 F1 \u8fbe\u5230<strong>\u00a070.7<\/strong>\u3002\u5728 AR-LSAT \/ LogiDed \/ MMLU \u8fd9\u79cd&#8221;\u7b54\u6848\u5bf9\u3001\u8fc7\u7a0b\u9519&#8221;\u9ad8\u5360\u6bd4\u7684\u6570\u636e\u96c6\u4e0a\u4f18\u52bf\u6700\u5927\u2014\u2014\u57fa\u7ebf\u5728\u8fd9\u91cc\u51e0\u4e4e\u5b8c\u5168\u5931\u6548\u3002<\/p>\n\n\n\n<p class=\"has-text-align-center\"><img decoding=\"async\" style=\"width: 650px;\" src=\"https:\/\/mmbiz.qpic.cn\/mmbiz_jpg\/yozlYZeLGPcQUDywzPxfgCrJ4KzibgwCUQvNPtZGv79FruBYGO8bDNxFdBw4aicrnDoIRqpQ9ggJa1uKKTS4eTthYmiaxZJPFib6DQoTLokcGRU\/640?wx_fmt=jpeg&amp;from=appmsg&amp;watermark=1&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1#imgIndex=11\" alt=\"\u56fe\u7247\"><\/p>\n\n\n\n<p><strong>\u591a\u8f6e\u4fee\u590d\uff1a<\/strong>Answer Matching \u51e0\u4e4e\u65e0\u6539\u5584\uff1bReward \/ LogicReward \u7684\u6807\u91cf\u4fe1\u53f7\u5608\u6742\uff1bLLM-as-Judge \u65e9\u671f\u7565\u6709\u63d0\u5347\u4f46\u5f88\u5feb\u9971\u548c\u3002<strong>\u53ea\u6709 SymDiag \u5728\u6bcf\u4e00\u8f6e\u90fd\u9886\u5148\u6240\u6709\u57fa\u7ebf<\/strong>\u2014\u2014\u5e76\u4e14\u589e\u957f\u66f4\u9661\u3001\u4e0d\u9971\u548c\u3002<\/p>\n\n\n\n<p class=\"has-text-align-center\"><img decoding=\"async\" src=\"https:\/\/mmbiz.qpic.cn\/mmbiz_jpg\/yozlYZeLGPeyvH2txF4bGuYg1Oggde6NusJxzatk7sjGxhVhQEwPygMv93CFic3f0u9sMN5S7hcwCw9MglZCMibC1mWLlLF6Svricw1wTxV468\/640?wx_fmt=jpeg&amp;from=appmsg&amp;watermark=1&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1#imgIndex=12\" alt=\"\u56fe\u7247\" style=\"width: 650px;\"><\/p>\n\n\n\n<p><strong>\u9519\u8bef\u5f52\u56e0\uff1a\u6a21\u578b\u8d8a\u5927\uff0c&#8221;\u5e7b\u89c9\u89c4\u5219&#8221;\u8d8a\u4e25\u91cd<\/strong><\/p>\n\n\n\n<p>\u501f\u52a9 SymDiag \u7684\u6b65\u7ea7\u8bca\u65ad\uff0c\u7814\u7a76\u56e2\u961f\u7b2c\u4e00\u6b21\u7cfb\u7edf\u523b\u753b\u4e86\u4e0d\u540c\u89c4\u6a21\u6a21\u578b\u7684&#8221;\u5931\u8d25\u753b\u50cf&#8221;\uff1a<\/p>\n\n\n\n<p class=\"has-text-align-center\"><img decoding=\"async\" style=\"width: 650px;\" src=\"https:\/\/mmbiz.qpic.cn\/sz_mmbiz_jpg\/yozlYZeLGPdKacL7sk5CwRO6VAg3ZMRZ58PwSsf4ZDj2P2MVibYLPHg6RsyQ1ibfvBrPMaibkWfia6HGYICJNYiaiaHWMooCzuqOISycltpiajTRy4\/640?wx_fmt=jpeg&amp;from=appmsg&amp;watermark=1&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1#imgIndex=14\" alt=\"\u56fe\u7247\"><\/p>\n\n\n\n<p><strong>\u5c0f\u6a21\u578b (Llama-3.2-1B \/ Qwen3-1.7B)<\/strong>\uff1a\u7b97\u672f\u9519\u8bef 35% \/ 30%\u3001\u524d\u63d0\u7f3a\u5931 30% \/ 28%\u2014\u2014\u7b26\u53f7\u64cd\u63a7\u4e0e\u524d\u63d0\u8ffd\u8e2a\u4e0d\u7a33\u3002<\/p>\n\n\n\n<p><strong>\u4e2d\u7b49\u6a21\u578b (Qwen3-8B)<\/strong>\uff1a\u4f4e\u5c42\u9519\u8bef\u4e0b\u964d\uff0c\u975e\u8574\u542b\u63a8\u7406 22%\u3001\u7ea6\u675f\u5ffd\u7565 25% \u4e0a\u5347\u3002<\/p>\n\n\n\n<p><strong>\u5927\u6a21\u578b (GPTOSS-20B)<\/strong>\uff1a\u7b97\u672f\u53ea\u5269 5%\uff0c\u4f46<strong>\u89c4\u5219\u8bef\u7528\u6fc0\u589e\u5230 32%\u3001\u7c7b\u578b\u9519\u914d 20%<\/strong>\u2014\u2014\u66f4\u5f3a\u7684\u62bd\u8c61\u80fd\u529b\u53cd\u800c\u5e26\u6765<strong>&#8220;\u8fc7\u5ea6\u6cdb\u5316&#8221;<\/strong>\u98ce\u9669\u3002<\/p>\n\n\n\n<p>\u8fd9\u4e00\u53d1\u73b0\u610f\u5473\u7740\uff1a<strong>\u63a8\u7406\u76d1\u7763\u5e94\u5f53\u6a21\u578b\u89c4\u6a21\u611f\u77e5<\/strong>\u2014\u2014\u5c0f\u6a21\u578b\u9700\u8981\u7ea6\u675f\u5f3a\u5316\u4e0e\u524d\u63d0\u8865\u5168\uff0c\u5927\u6a21\u578b\u5219\u9700\u8981\u9632\u8303\u89c4\u5219\u5e7b\u89c9\u3002\u540c\u4e00\u79cd reward\uff0c\u5bf9\u5927\u5c0f\u6a21\u578b\u5b8c\u5168\u662f\u4e24\u79cd\u836f\u3002<\/p>\n\n\n\n<p><strong>\u5173\u952e\u6570\u5b57<\/strong><\/p>\n\n\n\n<p class=\"has-text-align-center\"><img decoding=\"async\" src=\"https:\/\/mmbiz.qpic.cn\/mmbiz_jpg\/yozlYZeLGPdVCuCR89SnHNEowWAmYSh7VDukNqzlGWnaVr7ibKMLicTnQE3RdfS6sShLics8aj2ewlF8HjIFViao7M6Z5fpAicjfwCurgyLmnyUY\/640?wx_fmt=jpeg&amp;from=appmsg&amp;watermark=1&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1#imgIndex=16\" alt=\"\u56fe\u7247\" style=\"width: 650px;\"><\/p>\n\n\n\n<p>\u7ecf\u8fc7 3 \u8f6e Self-Auditor \u53cd\u9988\uff0c<strong>\u7ffb\u8bd1\u9519\u8bef\u4ece 20.1% \u964d\u5230\u63a5\u8fd1 0<\/strong>\uff0c\u6267\u884c\u5931\u8d25\u4ece 5.5% \u964d\u5230 1% \u4ee5\u4e0b\u3002\u8bc1\u636e\u8868\u660e\uff1a\u663e\u5f0f SAT \/ \u8574\u542b\u68c0\u67e5\u662f\u8bca\u65ad\u51c6\u786e\u6027\u7684\u4e3b\u9a71\u52a8\uff1bSelf-Auditor \u5219\u662f\u8bca\u65ad\u53ef\u9760\u6027\u7684\u5173\u952e\u5f00\u5173\u3002<\/p>\n\n\n\n<p><strong>\u610f\u4e49<\/strong><\/p>\n\n\n\n<p>\u8fd9\u7bc7\u8bba\u6587\u7684\u4ef7\u503c\u4e0d\u4ec5\u5728\u4e8e&#8221;\u63d0\u5347\u4e86\u591a\u8df3\u4e8b\u5b9e\u6838\u67e5\u7684\u51c6\u786e\u7387&#8221;\uff0c\u66f4\u5728\u4e8e<strong>\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u8303\u5f0f<\/strong>\uff1a<\/p>\n\n\n\n<p>\u7528<strong>\u7ed3\u6784\u5316\u7684\u5931\u8d25\u8bca\u65ad<\/strong>\u53d6\u4ee3<strong>\u7b3c\u7edf\u7684\u6253\u5206<\/strong>\uff0c\u7ea6\u675f\u548c\u4f18\u5316\u5927\u6a21\u578b\u7684\u591a\u8df3\u63a8\u7406\u8fc7\u7a0b\u3002\u8fc7\u53bb\u6211\u4eec\u7ecf\u5e38\u7528 Chain-of-Thought \u6765\u89c2\u5bdf\u6a21\u578b&#8221;\u600e\u4e48\u60f3&#8221;\uff0c\u4f46 CoT \u672c\u8eab\u5e76\u4e0d\u4fdd\u8bc1\u903b\u8f91\u5173\u7cfb\u771f\u6b63\u5b58\u5728\uff0c\u4e5f\u4e0d\u4fdd\u8bc1\u6bcf\u4e00\u6b65\u90fd\u6709\u8bc1\u636e\u652f\u6491\u3002<\/p>\n\n\n\n<p>SymDiag \u7ed9\u51fa\u4e86\u8fdb\u4e00\u6b65\u8981\u6c42\u6a21\u578b\u660e\u786e\u8bf4\u660e\uff1a<\/p>\n\n\n\n<ul>\n<li><strong>\u54ea\u4e9b\u4fe1\u606f<\/strong>\u6765\u81ea\u8bc1\u636e\uff1f<\/li>\n\n\n\n<li><strong>\u54ea\u4e9b\u7ed3\u8bba<\/strong>\u662f\u63a8\u5bfc\u51fa\u6765\u7684\uff1f<\/li>\n\n\n\n<li><strong>\u6bcf\u4e00\u6b65<\/strong>\u63a8\u5bfc\u4f9d\u8d56\u54ea\u4e9b\u524d\u7f6e\u4fe1\u606f\uff1f<\/li>\n\n\n\n<li><strong>\u6700\u7ec8\u7ed3\u8bba<\/strong>\u5982\u4f55\u7531\u8fd9\u4e9b\u7ed3\u6784\u5316\u8bc1\u636e\u652f\u6491\uff1f<\/li>\n<\/ul>\n\n\n\n<p>\u8fd9\u4f7f\u5f97\u63a8\u7406\u8fc7\u7a0b\u66f4\u5177\u53ef\u89e3\u91ca\u6027\uff0c\u4e5f\u66f4\u5bb9\u6613\u68c0\u67e5\u4e0e\u8c03\u8bd5\u3002\u5bf9\u4e8e\u9700\u8981\u53ef\u9760\u6027\u7684\u573a\u666f\uff0c\u6bd4\u5982\u4e8b\u5b9e\u6838\u67e5\u3001\u533b\u5b66\u8bc1\u636e\u7efc\u8ff0\u3001\u79d1\u7814\u95ee\u7b54\u548c\u590d\u6742\u4fe1\u606f\u68c0\u7d22\uff0c\u8fd9\u7c7b\u5bf9\u7ed3\u6784\u56e0\u679c\u6a21\u578b\u7684\u63a8\u7406\u65b9\u5f0f\u53ef\u80fd\u5177\u6709\u91cd\u8981\u610f\u4e49\u3002<\/p>\n\n\n\n<p>SymDiag&nbsp;<strong>\u63d0\u4f9b\u7684\u7ffb\u8bd1\/\u63a8\u7406\u9519\u8bef\u89e3\u8026\u673a\u5236<\/strong>\uff0c\u4e5f\u4e3a\u7c7b\u4f3c\u5de5\u4f5c\u63d0\u4f9b\u4e86\u4e00\u4e2a\u65b0\u601d\u8def\u2014\u2014\u53ef\u9760\u7684\u6b65\u7ea7\u7b26\u53f7\u63a8\u7406\u4e0d\u5728\u4e8e\u628a\u7ffb\u8bd1\u505a\u5f97\u66f4&#8221;\u82b1\u54e8&#8221;\uff0c\u800c\u5728\u4e8e\u5728\u51fa\u9519\u7684\u65f6\u5019\uff0c\u80fd\u8ba9&#8221;\u7ffb\u8bd1&#8221;\u548c&#8221;\u903b\u8f91&#8221;\u4e24\u6761\u817f\u540c\u65f6\u88ab\u68c0\u67e5\u3001\u4fee\u590d\u3002<\/p>\n\n\n\n<p>\u5b83\u63d0\u9192\u6211\u4eec\uff1a\u672a\u6765\u7684\u5927\u6a21\u578b\u63a8\u7406\u80fd\u529b\uff0c\u4e0d\u5e94\u8be5\u53ea\u770b\u6700\u7ec8\u7b54\u6848\u662f\u5426\u751f\u6210\u957f\u957f\u7684\u89e3\u91ca\uff0c\u800c\u5e94\u8be5\u770b\u662f\u5426\u80fd\u591f\u6784\u5efa\u66f4\u53ef\u9760\u3001\u66f4\u53ef\u9a8c\u8bc1\u3001\u66f4\u6709\u4f9d\u636e\u7684<strong>\u63a8\u7406\u7ed3\u6784<\/strong>\u3002<\/p>\n\n\n\n<p>\u4ece&#8221;<strong>\u8bf4\u5f97\u50cf\u6709\u9053\u7406<\/strong>&#8220;\u5230&#8221;<strong>\u6bcf\u4e00\u6b65\u90fd\u6709\u8bc1\u636e\u548c\u7ed3\u6784\u652f\u6491<\/strong>&#8220;\uff0c\u8fd9\u6216\u8bb8\u662f\u5927\u6a21\u578b\u53ef\u9760\u63a8\u7406\u8d70\u5411\u771f\u5b9e\u90e8\u7f72\u7684\u5173\u952e\u3002<\/p>\n\n\n\n<p><strong>\u8bba\u6587\u6807\u9898\uff1a<\/strong><\/p>\n\n\n\n<p><strong>SymDiag: Explainable Diagnosis for LLM Reasoning via Neuro-Symbolic Verification<\/strong><\/p>\n\n\n\n<p><strong>\u5f55\u7528\u4f1a\u8bae\uff1a<\/strong><\/p>\n\n\n\n<p>ACM SIGKDD 2026 (KDD &#8217;26) \u00b7 Research Track<\/p>\n\n\n\n<p><strong>\u5173\u952e\u8bcd\uff1a<\/strong><\/p>\n\n\n\n<p>LLM \u00b7 Diagnosis \u00b7 Symbolic \u00b7 Faithful Reasoning \u00b7 Explainability<\/p>\n\n\n\n<p><strong>\u8bba\u6587\u94fe\u63a5\uff1a<\/strong><\/p>\n\n\n\n<p>\u6682\u672a\u516c\u5f00\uff08camera-ready \u540e\u5c06\u63d0\u4f9b\uff09<\/p>\n\n\n\n<p><strong>\u4ee3\u7801 \/ \u6570\u636e\uff1a<\/strong><\/p>\n\n\n\n<p>camera-ready \u540e\u5f00\u6e90<\/p>\n\n\n\n<p><strong>\u7b2c\u4e00\u4f5c\u8005\uff1a\u5d14\u6587\u8000<\/strong><\/p>\n\n\n\n<p class=\"has-text-align-center\"><img decoding=\"async\" style=\"width: 650px;\" src=\"https:\/\/mmbiz.qpic.cn\/mmbiz_jpg\/yozlYZeLGPdO4cOzT5GpH7YPIy20FYj64kumbMHP9vSwNZvG5eNsRhQW23HUPlY3gZtHrEhQBKEibUByXUP0B8IJpo6eCSEwbqQqw5RYOfV4\/640?wx_fmt=jpeg&amp;from=appmsg&amp;watermark=1&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1#imgIndex=18\" alt=\"\u56fe\u7247\"><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5317\u4eac\u7406\u5de5\u5927\u5b66\u3001\u4e2d\u5173\u6751\u5b66\u9662\u3001\u4e2d\u56fd\u79d1\u5b66\u9662\u81ea\u52a8\u5316\u7814\u7a76\u6240 KDD\u4f1a\u8bae\u7b80\u4ecb KDD\uff08ACM &hellip; <a href=\"http:\/\/www.nlpir.org\/wordpress\/2026\/05\/21\/kdd-2026%ef%bc%9a%e4%bb%8e%e6%89%93%e5%88%86%e5%88%b0%e7%ac%a6%e5%8f%b7%e5%8c%96%e5%a4%b1%e8%b4%a5%e8%af%8a%e6%96%ad%ef%bc%8c%e8%ae%a9%e5%a4%a7%e6%a8%a1%e5%9e%8b%e6%8e%a8%e7%90%86%e5%8f%af%e8%af%8a\/\">\u7ee7\u7eed\u9605\u8bfb <span class=\"meta-nav\">&rarr;<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[31,38],"tags":[],"_links":{"self":[{"href":"http:\/\/www.nlpir.org\/wordpress\/wp-json\/wp\/v2\/posts\/2092379"}],"collection":[{"href":"http:\/\/www.nlpir.org\/wordpress\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.nlpir.org\/wordpress\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.nlpir.org\/wordpress\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/www.nlpir.org\/wordpress\/wp-json\/wp\/v2\/comments?post=2092379"}],"version-history":[{"count":1,"href":"http:\/\/www.nlpir.org\/wordpress\/wp-json\/wp\/v2\/posts\/2092379\/revisions"}],"predecessor-version":[{"id":2092409,"href":"http:\/\/www.nlpir.org\/wordpress\/wp-json\/wp\/v2\/posts\/2092379\/revisions\/2092409"}],"wp:attachment":[{"href":"http:\/\/www.nlpir.org\/wordpress\/wp-json\/wp\/v2\/media?parent=2092379"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.nlpir.org\/wordpress\/wp-json\/wp\/v2\/categories?post=2092379"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.nlpir.org\/wordpress\/wp-json\/wp\/v2\/tags?post=2092379"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}