{"id":120038,"date":"2026-10-30T23:24:31","date_gmt":"2026-10-30T15:24:31","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/120038.html"},"modified":"2026-10-30T23:24:31","modified_gmt":"2026-10-30T15:24:31","slug":"26-%e4%b8%8d%e4%bc%9a%e8%bf%98%e6%9c%89%e4%ba%ba%e6%8b%bf%e3%80%8c%e6%84%9f%e8%a7%89%e5%8f%98%e5%a5%bd%e4%ba%86%e3%80%8d%e5%bd%93%e6%94%b9%e8%bf%9b%e5%90%a7%ef%bc%9f%e5%9f%ba%e7%ba%bf%e4%b8%8d","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/120038.html","title":{"rendered":"26-\u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427\uff1f\u57fa\u7ebf\u4e0d\u5728\uff0c\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db"},"content":{"rendered":"

\u4e0a\u4e00\u7bc7：\u4e0d\u4f1a\u8fd8\u6709\u4eba\u7ea2\u706f\u53ea\u4f1a\u91cd\u8dd1\u548c\u7529\u9505\u5427？\u5047\u7eff\u81ea\u52a8\u4fee\u662f\u81ea\u6740 | \u7ea2\u706f\u5582\u56de\u95ed\u73af<\/p>\n

\u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427？\u57fa\u7ebf\u4e0d\u5728，\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db<\/h2>\n

\u7cfb\u5217：\u300a\u4ece Rule \u5230\u843d\u5730：AI \u5de5\u7a0b\u5316\u5b9e\u8df5\u300b\u7b2c 26 \u7bc7 \u6807\u7b7e：\u8bc4\u6d4b \u00b7 \u56de\u5f52\u57fa\u7ebf \u00b7 \u5b8f F1 \u00b7 Recall@K \u00b7 MRR \u4e13\u680f：\u4ece\u5c0f\u767d\u5230AI\u843d\u5730\u5927\u5e08<\/p>\n

\u5e8f\u8a00<\/h3>\n

\u6d4b\u8bd5\u5f27\u8d70\u5b8c\u4e09\u7ad9：\u7b2c 23 \u7bc7\u628a\u9a8c\u6536\u5199\u8fdb YAML，\u7b2c 24 \u7bc7\u5f52\u6863\u5931\u8d25\u73b0\u573a，\u7b2c 25 \u7bc7\u628a\u7ea2\u706f\u5582\u56de\u8bb0\u5fc6\u5e93\u3002\u8fd9\u4e00\u7bc7\u56de\u7b54\u4e00\u4e2a\u66f4\u5201\u94bb\u7684\u95ee\u9898：\u4f60\u600e\u4e48\u8bc1\u660e\u8fd9\u6b21\u6539\u52a8\u8ba9\u7cfb\u7edf\u53d8\u597d\u4e86，\u800c\u4e0d\u662f\u53ea\u662f\u6ca1\u53d8\u574f？\u6d4b\u8bd5\u65ad\u8a00\u7ba1\u300c\u5bf9\u300d，\u6307\u6807\u57fa\u7ebf\u7ba1\u300c\u597d\u300d\u2014\u2014\u4e24\u7801\u4e8b\u3002\u8bfb\u5b8c\u4f60\u8981\u80fd\u7ed9\u81ea\u5df1\u7cfb\u7edf\u91cc\u4efb\u4f55\u4e00\u4e2a\u300c\u7b97\u6cd5\u5473\u300d\u7ec4\u4ef6（\u5206\u7c7b\u89c4\u5219\u3001\u68c0\u7d22\u6392\u5e8f\u3001\u53ec\u56de\u7b56\u7565）\u7acb\u8d77\u4e00\u5957\u80fd\u8fdb CI \u7684\u8bc4\u6d4b\u95e8\u69db\u3002<\/p>\n

\u8bf4\u53e5\u81ea\u5632\u7684：\u6211\u5e72\u8fc7\u6700\u6f02\u4eae\u7684\u4e00\u6b21\u300c\u4f18\u5316\u300d，\u662f\u6539\u4e86\u68c0\u7d22\u7684\u540c\u4e49\u8bcd\u8868，\u8dd1\u5b8c\u81ea\u5df1\u624b\u5199\u7684\u4e09\u6761 case，\u5168\u4e2d，\u5174\u51b2\u51b2\u5408\u4e86\u7801\u3002\u4e24\u5468\u540e\u624d\u53d1\u73b0\u4e00\u4e2a\u51b7\u95e8\u610f\u56fe\u7c7b\u522b\u7684\u53ec\u56de\u6389\u4e86\u4e00\u534a\u2014\u2014\u90a3\u7c7b case \u6211\u4e00\u6761\u90fd\u6ca1\u5199\u3002\u611f\u89c9\u826f\u597d\u4e0d\u662f\u8bc1\u636e，\u662f\u76f2\u533a\u3002<\/p>\n

\u8bb0\u4f4f\u4e00\u53e5\u8bdd：\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u4e0d\u662f\u6307\u6807；\u57fa\u7ebf\u4e0d\u5728，\u6539\u8fdb\u4e0d\u7b97\u2014\u2014\u6389\u51fa 2pp \u5bb9\u5fcd\u5e26\u5c31\u5426\u51b3，\u60f3\u964d\u57fa\u7ebf\u5148\u8d70 PR\u3002\u4e0b\u9762\u8fd9\u4efd eval_gate.py \u7eaf\u6807\u51c6\u5e93\u3001\u590d\u5236\u5373\u8dd1，\u540c\u5c4f\u8e29\u4e0b\u4e09\u4e2a\u5239\u8f66：\u5b8f F1 \u62e6\u300c\u5c0f\u7c7b\u88ab\u5403\u6389\u300d\u3001\u5206\u5c42\u95e8\u69db\u62e6\u300c\u5e73\u5747\u6570\u9a97\u672f\u300d\u3001\u57fa\u7ebf diff \u62e6\u300c\u8ba9\u6d4b\u8bd5\u53d8\u7eff\u7684\u6377\u5f84\u300d\u3002<\/p>\n

\u6982\u5ff5<\/h3>\n

\u4e09\u4e2a\u7ef4\u5ea6\u3001\u4e09\u7ec4\u6307\u6807，\u5206\u5de5\u8981\u660e\u786e：<\/p>\n

\u610f\u56fe\u5206\u7c7b\u770b Top-1 \u51c6\u786e\u7387\u4e0e\u5b8f F1\u3002\u4e3a\u4ec0\u4e48\u4e0d\u7528 accuracy \u5f53\u95e8\u7981？\u610f\u56fe\u96c6\u5929\u7136\u4e0d\u5747\u8861\u2014\u2014\u300c\u6587\u6863\u95ee\u7b54\u300d\u7c7b\u8f93\u5165\u4e94\u82b1\u516b\u95e8\u5360\u5927\u5934，\u5c0f\u7c7b\u4e00\u4e2a\u6570\u636e\u96c6\u91cc\u53ea\u6709\u5341\u51e0\u6761\u3002accuracy \u4f1a\u88ab\u5927\u7c7b\u4e3b\u5bfc：\u5927\u7c7b\u5168\u5bf9，\u5c0f\u7c7b\u5168\u519b\u8986\u6ca1，\u6570\u5b57\u7167\u6837\u597d\u770b\u3002\u5b8f F1 \u5bf9\u6bcf\u4e2a\u7c7b\u522b\u5206\u522b\u7b97 F1 \u518d\u65e0\u6743\u91cd\u5e73\u5747，\u5c0f\u7c7b\u9519\u4e86\u76f4\u63a5\u628a\u5e73\u5747\u503c\u62fd\u4e0b\u6765\u3002\u6df7\u6dc6\u77e9\u9635\u662f\u6392\u969c\u5165\u53e3：Top-1 \u6389\u4e86，\u5148\u770b\u662f\u8c01\u548c\u8c01\u4e92\u76f8\u8ba4\u9519\u3002<\/p>\n

\u68c0\u7d22\u770b Recall@K \u4e0e MRR\u3002Recall@K = \u524d K \u4e2a\u7ed3\u679c\u547d\u4e2d\u7684\u76f8\u5173\u6587\u6863\u6570 \/ \u5168\u90e8\u76f8\u5173\u6587\u6863\u6570，\u8861\u91cf\u300c\u635e\u5230\u6ca1\u6709\u300d\u2014\u2014RAG \u573a\u666f\u91cc\u6f0f\u6389\u7684\u6587\u6863\u540e\u9762 rerank \u6551\u4e0d\u56de\u6765，\u6240\u4ee5\u5b83\u662f\u4f18\u5148\u4f18\u5316\u9879\u3002MRR = \u5404 query\u300c\u7b2c\u4e00\u6761\u6b63\u786e\u7ed3\u679c\u7684\u6392\u540d\u5012\u6570\u300d\u7684\u5e73\u5747，\u8861\u91cf\u300c\u635e\u5230\u4e86\u6392\u7b2c\u51e0\u300d\u2014\u2014\u9002\u5408\u300c\u53ea\u8981\u4e00\u6761\u6b63\u786e\u7b54\u6848\u300d\u7684\u4efb\u52a1，\u6bd4\u5982\u628a\u610f\u56fe\u8def\u7531\u5230\u90a3\u6761\u5bf9\u7684\u8bb0\u5fc6\u3002\u76f8\u5173\u6587\u6863\u82e5\u5206\u7ea7，\u8be5\u6362 nDCG，\u4f46\u522b\u4e3a\u7528\u4e0d\u4e0a\u7684\u5f62\u6001\u4e0a\u91cd\u6307\u6807\u3002<\/p>\n

\u57fa\u7ebf\u662f\u7248\u672c\u5316\u7684 JSON（\u6216 dict），\u5143\u6570\u636e\u5199\u660e\u6bcf\u4e2a\u6570\u5b57\u7684\u51fa\u5904：\u63d0\u5347\u57fa\u7ebf\u9700 PR \u8bf4\u660e\u4f9d\u636e，\u7981\u6b62\u6084\u964d\u3002\u5f53\u524d\u6307\u6807\u4f4e\u4e8e\u57fa\u7ebf 2 \u4e2a\u767e\u5206\u70b9（2pp）\u5373 fail\u30022pp \u662f\u523b\u610f\u7559\u7684\u5bb9\u5dee\u2014\u2014\u6570\u636e\u96c6\u6709\u968f\u673a\u6210\u5206，\u4e0d\u7ed9\u5bb9\u5dee\u4f1a\u628a flaky test \u53d8\u6210\u65e5\u5e38；\u4f46 2pp \u4e5f\u8db3\u591f\u5c0f，\u771f\u5b9e\u7684\u7b97\u6cd5\u52a3\u5316\u62e6\u5f97\u4f4f\u3002<\/p>\n

\u53ef\u4ee5\u7406\u89e3\u4e3a：\u4f53\u68c0\u770b\u7684\u662f\u300c\u548c\u53bb\u5e74\u4ee3\u6bd4\u300d，\u4e0d\u662f\u300c\u4eca\u5929\u611f\u89c9\u300d\u3002\u57fa\u7ebf\u5c31\u662f\u90a3\u5f20\u5b58\u6863\u7684\u62a5\u544a；\u6ca1\u6709\u5b58\u6863，\u4f60\u62ff\u4ec0\u4e48\u8bc1\u660e\u7626\u4e86\u4e24\u65a4？<\/p>\n

\u95ee\u9898\u5728\u54ea<\/h3>\n

\u6ca1\u6709\u56de\u5f52\u95e8\u69db\u7684\u56e2\u961f，\u6539\u52a8\u7684\u9ed8\u8ba4\u59ff\u6001\u662f\u300c\u6211\u89c9\u5f97\u66f4\u597d\u5c31\u5408\u300d\u3002\u5355\u6b21\u90fd\u5408\u7406，\u6bcf\u5b63\u5ea6\u6389 1pp \u6ca1\u6709\u4efb\u4f55\u4e00\u6b21\u770b\u8d77\u6765\u4e25\u91cd，\u4e00\u5e74\u540e\u56de\u5934\u770b\u6307\u6807\u5df2\u7ecf\u70c2\u4e86\u2014\u2014\u6e29\u6c34\u716e\u9752\u86d9\u5f0f\u7684\u300c\u987a\u624b\u4f18\u5316\u300d\u3002\u8fd9\u4e0e A\/B \u6d4b\u8bd5\u601d\u60f3\u540c\u6e90：\u7ebf\u4e0a\u5b9e\u9a8c\u62ff\u5bf9\u7167\u7ec4\u62e6\u4f4f\u300c\u611f\u89c9\u826f\u597d\u300d\u7684\u4ea7\u54c1\u6539\u52a8，\u7248\u672c\u5316\u57fa\u7ebf\u62ff\u56fa\u5b9a\u6570\u636e\u96c6\u62e6\u4f4f\u300c\u611f\u89c9\u826f\u597d\u300d\u7684\u7b97\u6cd5\u6539\u52a8\u3002<\/p>\n

\u8fd8\u6709\u4e24\u79cd\u66f4\u9690\u853d\u7684\u81ea\u6b3a，\u672c\u6587 demo \u5404\u6f14\u4e00\u6b21：\u5176\u4e00，\u603b\u91cf\u6307\u6807\u597d\u770b\u3001\u67d0\u4e2a\u5206\u5c42\u5168\u70c2\u2014\u2014\u5e73\u5747\u6570\u9a97\u672f；\u89e3\u6cd5\u662f\u5206\u5c42\u95e8\u69db（\u6700\u5dee project_type \u4e0d\u5f97\u4f4e\u4e8e\u7ebf）\u3002\u5176\u4e8c，\u6d4b\u8bd5\u7ea2\u4e86\u4e0d\u53bb\u4fee\u6539\u52a8，\u53bb\u6539\u57fa\u7ebf\u2014\u2014\u628a 0.85 \u6539\u6210 0.80 \u5c31\u300c\u7eff\u300d\u4e86；\u89e3\u6cd5\u662f\u628a\u57fa\u7ebf diff \u672c\u8eab\u7eb3\u5165\u95e8\u7981，\u964d\u57fa\u7ebf\u89c6\u540c\u56de\u5f52，\u5fc5\u987b\u8d70\u663e\u5f0f\u5ba1\u6279\u3002<\/p>\n

\u6700\u540e\u63d0\u9192\u4e00\u53e5\u5de5\u7a0b\u8d26：\u8bc4\u6d4b\u4e0d\u662f\u514d\u8d39\u7684\u3002\u5efa\u5e93\u3001\u704c\u6570\u636e\u3001\u8dd1\u4eff\u771f\u90fd\u82b1\u65f6\u95f4，\u5168\u585e\u8fdb\u6bcf\u6b21 PR \u4f1a\u62d6\u57ae\u5f00\u53d1\u8282\u594f\u3002\u901a\u884c\u89e3\u6cd5\u662f\u5206\u7ea7\u2014\u2014PR \u5feb\u901f\u95e8\u69db\u8dd1\u786e\u5b9a\u6027\u8f7b\u8bc4\u6d4b，nightly \u515c\u4f4f\u91cd\u7684\u5168\u91cf（\u542b\u4eff\u771f\u3001\u542b\u771f API \u7684\u5bf9\u7167\u8bc4\u6d4b）\u3002\u4ee3\u4ef7\u662f nightly \u624d\u80fd\u53d1\u73b0\u7684\u95ee\u9898\u6700\u591a\u6ede\u540e\u4e00\u5929，\u81ea\u5df1\u6743\u8861\u7740\u4e70\u3002<\/p>\n

\u4e3e\u4f8b<\/h3>\n

\u573a\u666f：\u4e00\u6b21\u300c\u610f\u56fe\u89c4\u5219\u5fae\u8c03\u300d\u5408\u7801\u524d\u540e，\u8dd1 tests\/eval\/ \u91cc\u4e24\u5957\u56fa\u5b9a\u6570\u636e\u96c6：32 \u6761\u610f\u56fe\u6807\u6ce8\u5bf9\u300110 \u6761\u5206\u9879\u76ee\u7c7b\u578b\u7684\u68c0\u7d22 query\u3002\u57fa\u7ebf intent_rule: top1 0.85 \/ macro_f1 0.85，retrieval: recall@5 0.80 \/ mrr 0.60，\u5206\u5c42\u4e0b\u9650 0.70\u3002<\/p>\n

\u6b65\u9aa4\u603b\u89c8（\u4e0b\u9762\u8fd9\u6bb5\u4ee3\u7801\u5728\u5e72\u4ec0\u4e48，\u4e00\u53e3\u6c14\u8bb2\u6e05）：<\/p>\n

  • macro_f1(pairs)：\u4ece (\u9884\u6d4b, \u771f\u5b9e) \u5bf9\u5217\u8868\u7b97 Top-1\u3001\u5b8f F1 \u4e0e\u5404\u7c7b F1\u2014\u2014\u6bcf\u7c7b\u5206\u522b\u7b97 P\/R \u518d\u5408\u6210 F1，\u6700\u540e\u65e0\u6743\u91cd\u5e73\u5747\u3002<\/li>\n
  • check()：\u5f53\u524d\u503c\u4f4e\u4e8e\u57fa\u7ebf\u8d85\u8fc7 REGRESSION_TOLERANCE = 0.02 \u5373\u6253 FAIL\u2014\u2014\u5bb9\u5fcd\u5e26\u4e4b\u5916\u65e0\u5546\u91cf\u3002<\/li>\n
  • \u68c0\u7d22\u4fa7 recall_at_5() \u4e0e rr() \u7b97\u603b\u91cf，\u518d\u6309 project_type \u5207\u5206\u5c42：\u4efb\u4f55\u4e00\u7ec4\u4f4e\u4e8e 0.70，\u603b\u91cf\u518d\u597d\u770b\u4e5f\u89c6\u540c\u56de\u5f52\u3002<\/li>\n
  • \u57fa\u7ebf diff \u68c0\u67e5：\u62ff\u300cPR \u91cc\u7684\u57fa\u7ebf\u300d\u5bf9\u300c\u7248\u672c\u5316\u57fa\u7ebf\u300d，\u53d1\u73b0\u4e0b\u8c03\u4e14 pr_approval \u91cc\u6ca1\u6709\u5ba1\u6279\u8bb0\u5f55 \u2192 \u76f4\u63a5 reject\u3002<\/li>\n
  • \u6c47\u603b\u5bf9\u7167：\u4e09\u4e2a\u5239\u8f66\u5404\u62e6\u54ea\u7c7b\u81ea\u6b3a\u3002<\/li>\n

    \u5b8c\u6574\u53ef\u8dd1：eval_gate.py<\/h4>\n

    # eval_gate.py \u2014\u2014 \u7b2c26\u7bc7\u5b8c\u6574\u6848\u4f8b（\u865a\u6784\u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db，\u4ec5\u4f9d\u8d56\u6807\u51c6\u5e93）<\/span>
    \n# \u7eaf\u6807\u51c6\u5e93，\u590d\u5236\u5373\u8dd1\u3002\u7528\u6cd5：python eval_gate.py<\/span>
    \nimport<\/span> copy<\/p>\n

    REGRESSION_TOLERANCE =<\/span> 0.02<\/span> # \u56de\u5f52\u95e8\u69db 2pp：\u4f4e\u4e8e\u57fa\u7ebf\u8d85 2 \u4e2a\u767e\u5206\u70b9\u5373\u5426\u51b3<\/span>
    \nMIN_STRATUM_RECALL =<\/span> 0.70<\/span> # \u5206\u5c42\u95e8\u69db：\u6700\u5dee project_type \u7684 Recall@5 \u4e0b\u9650<\/span><\/p>\n

    # \u57fa\u7ebf\u662f\u7248\u672c\u5316 JSON（\u8fd9\u91cc\u7528 dict \u793a\u610f），\u5143\u6570\u636e\u5199\u660e\u51fa\u5904\u4e0e\u7eaa\u5f8b<\/span>
    \nBASELINES =<\/span> {<\/span>
    \n "_meta"<\/span>:<\/span> "\u57fa\u7ebf\u6570\u503c\u7248\u672c\u5316：\u63d0\u5347\u57fa\u7ebf\u9700 PR \u8bf4\u660e\u4f9d\u636e，\u7981\u6b62\u6084\u964d\u3002\u56de\u5f52\u95e8\u69db=\u4f4e\u4e8e\u57fa\u7ebf 2pp \u5373\u5426\u51b3\u3002"<\/span>,<\/span>
    \n "intent_rule"<\/span>:<\/span> {<\/span>"top1"<\/span>:<\/span> 0.85<\/span>,<\/span> "macro_f1"<\/span>:<\/span> 0.85<\/span>}<\/span>,<\/span>
    \n "retrieval"<\/span>:<\/span> {<\/span>"recall@5"<\/span>:<\/span> 0.80<\/span>,<\/span> "mrr"<\/span>:<\/span> 0.60<\/span>}<\/span>,<\/span>
    \n}<\/span><\/p>\n

    def<\/span> macro_f1<\/span>(<\/span>pairs)<\/span>:<\/span>
    \n """(\u9884\u6d4b, \u771f\u5b9e) \u5217\u8868 -> (top1, \u5b8fF1, \u5404\u7c7b F1)\u3002\u5b8f F1 = \u6bcf\u7c7b\u5404\u7b97 F1 \u518d\u65e0\u6743\u91cd\u5e73\u5747\u3002"""<\/span>
    \n labels =<\/span> sorted<\/span>(<\/span>{<\/span>g for<\/span> _,<\/span> g in<\/span> pairs}<\/span> |<\/span> {<\/span>p for<\/span> p,<\/span> _ in<\/span> pairs}<\/span>)<\/span>
    \n f1s =<\/span> {<\/span>}<\/span>
    \n for<\/span> lb in<\/span> labels:<\/span>
    \n tp =<\/span> sum<\/span>(<\/span>1<\/span> for<\/span> p,<\/span> g in<\/span> pairs if<\/span> p ==<\/span> lb and<\/span> g ==<\/span> lb)<\/span>
    \n fp =<\/span> sum<\/span>(<\/span>1<\/span> for<\/span> p,<\/span> g in<\/span> pairs if<\/span> p ==<\/span> lb and<\/span> g !=<\/span> lb)<\/span>
    \n fn =<\/span> sum<\/span>(<\/span>1<\/span> for<\/span> p,<\/span> g in<\/span> pairs if<\/span> p !=<\/span> lb and<\/span> g ==<\/span> lb)<\/span>
    \n prec =<\/span> tp \/<\/span> (<\/span>tp +<\/span> fp)<\/span> if<\/span> tp +<\/span> fp else<\/span> 0.0<\/span>
    \n rec =<\/span> tp \/<\/span> (<\/span>tp +<\/span> fn)<\/span> if<\/span> tp +<\/span> fn else<\/span> 0.0<\/span>
    \n f1s[<\/span>lb]<\/span> =<\/span> 2<\/span> *<\/span> prec *<\/span> rec \/<\/span> (<\/span>prec +<\/span> rec)<\/span> if<\/span> prec +<\/span> rec else<\/span> 0.0<\/span>
    \n top1 =<\/span> sum<\/span>(<\/span>1<\/span> for<\/span> p,<\/span> g in<\/span> pairs if<\/span> p ==<\/span> g)<\/span> \/<\/span> len<\/span>(<\/span>pairs)<\/span>
    \n return<\/span> top1,<\/span> sum<\/span>(<\/span>f1s.<\/span>values(<\/span>)<\/span>)<\/span> \/<\/span> len<\/span>(<\/span>f1s)<\/span>,<\/span> f1s<\/p>\n

    def<\/span> check<\/span>(<\/span>name,<\/span> cur,<\/span> base)<\/span>:<\/span>
    \n if<\/span> base –<\/span> cur ><\/span> REGRESSION_TOLERANCE:<\/span>
    \n print<\/span>(<\/span>f"[FAIL] <\/span>{<\/span>name}<\/span><\/span>: <\/span>{<\/span>cur:<\/span>.2f<\/span>}<\/span><\/span> < \u57fa\u7ebf <\/span>{<\/span>base:<\/span>.2f<\/span>}<\/span><\/span> – 2pp（\u5dee <\/span>{<\/span>base –<\/span> cur:<\/span>.2f<\/span>}<\/span><\/span>）\u2014\u2014\u6df7\u6dc6\u77e9\u9635\u89c1 reports\/"<\/span><\/span>)<\/span>
    \n return<\/span> False<\/span>
    \n print<\/span>(<\/span>f"[PASS] <\/span>{<\/span>name}<\/span><\/span>: <\/span>{<\/span>cur:<\/span>.2f<\/span>}<\/span><\/span>（\u57fa\u7ebf <\/span>{<\/span>base:<\/span>.2f<\/span>}<\/span><\/span>）"<\/span><\/span>)<\/span>
    \n return<\/span> True<\/span><\/p>\n

    def<\/span> pairs<\/span>(<\/span>counts)<\/span>:<\/span>
    \n return<\/span> [<\/span>(<\/span>p,<\/span> g)<\/span> for<\/span> (<\/span>p,<\/span> g)<\/span>,<\/span> n in<\/span> counts.<\/span>items(<\/span>)<\/span> for<\/span> _ in<\/span> range<\/span>(<\/span>n)<\/span>]<\/span><\/p>\n

    # \u2014\u2014 \u610f\u56fe\u5206\u7c7b：\u6539\u52a8\u524d 32 \u6761\u3001\u6539\u52a8\u540e\u53ea\u9519\u5f97\u66f4\u300c\u504f\u5c0f\u7c7b\u300d\u2014\u2014<\/span>
    \nV1 =<\/span> pairs(<\/span>{<\/span>(<\/span>"query"<\/span>,<\/span> "query"<\/span>)<\/span>:<\/span> 14<\/span>,<\/span> (<\/span>"refactor"<\/span>,<\/span> "refactor"<\/span>)<\/span>:<\/span> 10<\/span>,<\/span> (<\/span>"explain"<\/span>,<\/span> "explain"<\/span>)<\/span>:<\/span> 4<\/span>,<\/span>
    \n (<\/span>"query"<\/span>,<\/span> "refactor"<\/span>)<\/span>:<\/span> 1<\/span>,<\/span> (<\/span>"refactor"<\/span>,<\/span> "query"<\/span>)<\/span>:<\/span> 1<\/span>,<\/span>
    \n (<\/span>"explain"<\/span>,<\/span> "refactor"<\/span>)<\/span>:<\/span> 1<\/span>,<\/span> (<\/span>"refactor"<\/span>,<\/span> "explain"<\/span>)<\/span>:<\/span> 1<\/span>}<\/span>)<\/span>
    \nV2 =<\/span> pairs(<\/span>{<\/span>(<\/span>"query"<\/span>,<\/span> "query"<\/span>)<\/span>:<\/span> 14<\/span>,<\/span> (<\/span>"refactor"<\/span>,<\/span> "refactor"<\/span>)<\/span>:<\/span> 10<\/span>,<\/span> (<\/span>"explain"<\/span>,<\/span> "explain"<\/span>)<\/span>:<\/span> 3<\/span>,<\/span>
    \n (<\/span>"query"<\/span>,<\/span> "refactor"<\/span>)<\/span>:<\/span> 1<\/span>,<\/span> (<\/span>"refactor"<\/span>,<\/span> "query"<\/span>)<\/span>:<\/span> 1<\/span>,<\/span>
    \n (<\/span>"explain"<\/span>,<\/span> "refactor"<\/span>)<\/span>:<\/span> 1<\/span>,<\/span> (<\/span>"refactor"<\/span>,<\/span> "explain"<\/span>)<\/span>:<\/span> 1<\/span>,<\/span>
    \n (<\/span>"query"<\/span>,<\/span> "explain"<\/span>)<\/span>:<\/span> 1<\/span>}<\/span>)<\/span> # \u610f\u56fe\u89c4\u5219\u5fae\u8c03：\u4e00\u6761 explain \u88ab query \u5403\u6389<\/span><\/p>\n

    print<\/span>(<\/span>"==== \u610f\u56fe\u5206\u7c7b：accuracy \u4f1a\u9a97\u4eba，\u5b8f F1 \u4e0d\u4f1a ===="<\/span>)<\/span>
    \nfor<\/span> tag,<\/span> data in<\/span> [<\/span>(<\/span>"\u6539\u52a8\u524d "<\/span>,<\/span> V1)<\/span>,<\/span> (<\/span>"\u89c4\u5219\u5fae\u8c03\u540e"<\/span>,<\/span> V2)<\/span>]<\/span>:<\/span>
    \n top1,<\/span> mf,<\/span> per =<\/span> macro_f1(<\/span>data)<\/span>
    \n detail =<\/span> " "<\/span>.<\/span>join(<\/span>f"<\/span>{<\/span>k}<\/span><\/span> <\/span>{<\/span>v:<\/span>.2f<\/span>}<\/span><\/span>"<\/span><\/span> for<\/span> k,<\/span> v in<\/span> per.<\/span>items(<\/span>)<\/span>)<\/span>
    \n print<\/span>(<\/span>f"<\/span>{<\/span>tag}<\/span><\/span>: top1 <\/span>{<\/span>top1:<\/span>.2f<\/span>}<\/span><\/span> | macro_f1 <\/span>{<\/span>mf:<\/span>.2f<\/span>}<\/span><\/span> | \u5206\u7c7b F1: <\/span>{<\/span>detail}<\/span><\/span>"<\/span><\/span>)<\/span>
    \nprint<\/span>(<\/span>)<\/span>
    \nt1,<\/span> m1,<\/span> _ =<\/span> macro_f1(<\/span>V1)<\/span>
    \nt2,<\/span> m2,<\/span> _ =<\/span> macro_f1(<\/span>V2)<\/span>
    \ncheck(<\/span>"\u57fa\u7ebf\u590d\u6838 intent_rule.top1"<\/span>,<\/span> t1,<\/span> BASELINES[<\/span>"intent_rule"<\/span>]<\/span>[<\/span>"top1"<\/span>]<\/span>)<\/span>
    \ncheck(<\/span>"\u57fa\u7ebf\u590d\u6838 intent_rule.macro_f1"<\/span>,<\/span> m1,<\/span> BASELINES[<\/span>"intent_rule"<\/span>]<\/span>[<\/span>"macro_f1"<\/span>]<\/span>)<\/span>
    \ncheck(<\/span>"\u91cd\u8bc4 intent_rule.top1"<\/span>,<\/span> t2,<\/span> BASELINES[<\/span>"intent_rule"<\/span>]<\/span>[<\/span>"top1"<\/span>]<\/span>)<\/span>
    \ncheck(<\/span>"\u91cd\u8bc4 intent_rule.macro_f1"<\/span>,<\/span> m2,<\/span> BASELINES[<\/span>"intent_rule"<\/span>]<\/span>[<\/span>"macro_f1"<\/span>]<\/span>)<\/span><\/p>\n

    # \u2014\u2014 \u68c0\u7d22：Recall@5 \u7ba1\u300c\u635e\u5230\u6ca1\u300d，MRR \u7ba1\u300c\u635e\u5230\u6392\u7b2c\u51e0\u300d；\u8fd8\u6309 project_type \u5206\u5c42 \u2014\u2014<\/span>
    \nQUERIES =<\/span> [<\/span> # (project_type, \u76f8\u5173\u6587\u6863\u96c6, \u6392\u5e8f\u540e\u7684\u68c0\u7d22\u7ed3\u679c)<\/span>
    \n (<\/span>"web"<\/span>,<\/span> {<\/span>"d1"<\/span>}<\/span>,<\/span> [<\/span>"d1"<\/span>,<\/span> "x1"<\/span>,<\/span> "x2"<\/span>]<\/span>)<\/span>,<\/span>
    \n (<\/span>"web"<\/span>,<\/span> {<\/span>"d2"<\/span>}<\/span>,<\/span> [<\/span>"x1"<\/span>,<\/span> "d2"<\/span>,<\/span> "x2"<\/span>]<\/span>)<\/span>,<\/span>
    \n (<\/span>"web"<\/span>,<\/span> {<\/span>"d3"<\/span>,<\/span> "d4"<\/span>}<\/span>,<\/span> [<\/span>"d3"<\/span>,<\/span> "x1"<\/span>,<\/span> "x2"<\/span>,<\/span> "d4"<\/span>,<\/span> "x3"<\/span>]<\/span>)<\/span>,<\/span>
    \n (<\/span>"web"<\/span>,<\/span> {<\/span>"d5"<\/span>}<\/span>,<\/span> [<\/span>"x1"<\/span>,<\/span> "x2"<\/span>,<\/span> "x3"<\/span>,<\/span> "d5"<\/span>,<\/span> "x4"<\/span>]<\/span>)<\/span>,<\/span>
    \n (<\/span>"data"<\/span>,<\/span> {<\/span>"d6"<\/span>}<\/span>,<\/span> [<\/span>"x1"<\/span>,<\/span> "x2"<\/span>,<\/span> "d6"<\/span>]<\/span>)<\/span>,<\/span>
    \n (<\/span>"data"<\/span>,<\/span> {<\/span>"d7"<\/span>,<\/span> "d8"<\/span>}<\/span>,<\/span> [<\/span>"d7"<\/span>,<\/span> "x1"<\/span>,<\/span> "x2"<\/span>,<\/span> "x3"<\/span>,<\/span> "x4"<\/span>,<\/span> "x5"<\/span>,<\/span> "x6"<\/span>,<\/span> "d8"<\/span>]<\/span>)<\/span>,<\/span>
    \n (<\/span>"data"<\/span>,<\/span> {<\/span>"d9"<\/span>}<\/span>,<\/span> [<\/span>"d9"<\/span>,<\/span> "x1"<\/span>]<\/span>)<\/span>,<\/span>
    \n (<\/span>"iot"<\/span>,<\/span> {<\/span>"d10"<\/span>,<\/span> "d11"<\/span>}<\/span>,<\/span> [<\/span>"x1"<\/span>,<\/span> "d10"<\/span>,<\/span> "x2"<\/span>]<\/span>)<\/span>,<\/span>
    \n (<\/span>"iot"<\/span>,<\/span> {<\/span>"d12"<\/span>}<\/span>,<\/span> [<\/span>"d12"<\/span>]<\/span>)<\/span>,<\/span>
    \n (<\/span>"iot"<\/span>,<\/span> {<\/span>"d13"<\/span>,<\/span> "d14"<\/span>,<\/span> "d15"<\/span>,<\/span> "d16"<\/span>}<\/span>,<\/span> [<\/span>"x1"<\/span>,<\/span> "x2"<\/span>,<\/span> "x3"<\/span>,<\/span> "x4"<\/span>,<\/span> "d13"<\/span>,<\/span> "x5"<\/span>]<\/span>)<\/span>,<\/span>
    \n]<\/span><\/p>\n

    def<\/span> recall_at_5<\/span>(<\/span>rel,<\/span> ranked,<\/span> k=<\/span>5<\/span>)<\/span>:<\/span>
    \n return<\/span> len<\/span>(<\/span>rel &<\/span> set<\/span>(<\/span>ranked[<\/span>:<\/span>k]<\/span>)<\/span>)<\/span> \/<\/span> len<\/span>(<\/span>rel)<\/span><\/p>\n

    def<\/span> rr<\/span>(<\/span>rel,<\/span> ranked)<\/span>:<\/span>
    \n for<\/span> i,<\/span> doc in<\/span> enumerate<\/span>(<\/span>ranked,<\/span> 1<\/span>)<\/span>:<\/span>
    \n if<\/span> doc in<\/span> rel:<\/span>
    \n return<\/span> 1<\/span> \/<\/span> i
    \n return<\/span> 0.0<\/span><\/p>\n

    print<\/span>(<\/span>)<\/span>
    \nprint<\/span>(<\/span>"==== \u68c0\u7d22：\u603b\u91cf\u4e0e\u5206\u5c42，\u4e00\u4e2a\u90fd\u4e0d\u80fd\u5c11\u770b ===="<\/span>)<\/span>
    \nrecs =<\/span> [<\/span>recall_at_5(<\/span>rel,<\/span> ranked)<\/span> for<\/span> _,<\/span> rel,<\/span> ranked in<\/span> QUERIES]<\/span>
    \nrrs =<\/span> [<\/span>rr(<\/span>rel,<\/span> ranked)<\/span> for<\/span> _,<\/span> rel,<\/span> ranked in<\/span> QUERIES]<\/span>
    \noverall_rec,<\/span> overall_mrr =<\/span> sum<\/span>(<\/span>recs)<\/span> \/<\/span> len<\/span>(<\/span>recs)<\/span>,<\/span> sum<\/span>(<\/span>rrs)<\/span> \/<\/span> len<\/span>(<\/span>rrs)<\/span>
    \nb =<\/span> BASELINES[<\/span>"retrieval"<\/span>]<\/span>
    \nok =<\/span> check(<\/span>"retrieval.recall@5"<\/span>,<\/span> overall_rec,<\/span> b[<\/span>"recall@5"<\/span>]<\/span>)<\/span> and<\/span> check(<\/span>"retrieval.mrr"<\/span>,<\/span> overall_mrr,<\/span> b[<\/span>"mrr"<\/span>]<\/span>)<\/span>
    \nif<\/span> ok:<\/span>
    \n print<\/span>(<\/span>"\u603b\u91cf\u8fbe\u6807\u2014\u2014\u5148\u522b\u6025\u7740\u5e86\u795d，\u5207\u4e00\u5200\u5206\u5c42："<\/span>)<\/span>
    \ntypes =<\/span> sorted<\/span>(<\/span>{<\/span>pt for<\/span> pt,<\/span> _,<\/span> _ in<\/span> QUERIES}<\/span>)<\/span>
    \nline =<\/span> " | "<\/span>.<\/span>join(<\/span>
    \n f"<\/span>{<\/span>pt}<\/span><\/span> <\/span>{<\/span>sum<\/span>(<\/span>r for<\/span> (<\/span>p,<\/span> _,<\/span> _)<\/span>,<\/span> r in<\/span> zip<\/span>(<\/span>QUERIES,<\/span> recs)<\/span> if<\/span> p ==<\/span> pt)<\/span> \/<\/span> sum<\/span>(<\/span>1<\/span> for<\/span> q in<\/span> QUERIES if<\/span> q[<\/span>0<\/span>]<\/span> ==<\/span> pt)<\/span>:<\/span>.2f<\/span>}<\/span><\/span>"<\/span><\/span>
    \n f"(<\/span>{<\/span>sum<\/span>(<\/span>1<\/span> for<\/span> q in<\/span> QUERIES if<\/span> q[<\/span>0<\/span>]<\/span> ==<\/span> pt)<\/span>}<\/span><\/span>\u6761)"<\/span><\/span> for<\/span> pt in<\/span> types)<\/span>
    \nprint<\/span>(<\/span>f"\u6309 project_type：<\/span>{<\/span>line}<\/span><\/span>"<\/span><\/span>)<\/span>
    \nfor<\/span> pt in<\/span> types:<\/span>
    \n sub =<\/span> [<\/span>r for<\/span> (<\/span>p,<\/span> _,<\/span> _)<\/span>,<\/span> r in<\/span> zip<\/span>(<\/span>QUERIES,<\/span> recs)<\/span> if<\/span> p ==<\/span> pt]<\/span>
    \n avg =<\/span> sum<\/span>(<\/span>sub)<\/span> \/<\/span> len<\/span>(<\/span>sub)<\/span>
    \n if<\/span> avg <<\/span> MIN_STRATUM_RECALL:<\/span>
    \n print<\/span>(<\/span>f"[FAIL] \u5206\u5c42\u95e8\u69db：<\/span>{<\/span>pt}<\/span><\/span> \u7ec4 recall@5 <\/span>{<\/span>avg:<\/span>.2f<\/span>}<\/span><\/span> < <\/span>{<\/span>MIN_STRATUM_RECALL:<\/span>.2f<\/span>}<\/span><\/span>\u2014\u2014\u603b\u91cf\u8fbe\u6807\u4f46\u67d0\u7c7b\u5168\u70c2，\u89c6\u540c\u56de\u5f52"<\/span><\/span>)<\/span><\/p>\n

    # \u2014\u2014 \u5077\u61d2 diff：\u6539\u57fa\u7ebf\u8ba9\u6d4b\u8bd5\u53d8\u7eff？\u2014\u2014<\/span>
    \nprint<\/span>(<\/span>)<\/span>
    \nprint<\/span>(<\/span>'==== \u5077\u61d2 diff：\u628a\u57fa\u7ebf 0.85 \u6539\u5230 0.80 \u8ba9 top1 "\u53d8\u7eff"？ ===='<\/span>)<\/span>
    \npr =<\/span> copy.<\/span>deepcopy(<\/span>BASELINES)<\/span>
    \npr[<\/span>"intent_rule"<\/span>]<\/span>[<\/span>"top1"<\/span>]<\/span> =<\/span> 0.80<\/span>
    \npr_approval =<\/span> {<\/span>}<\/span> # \u672c\u6b21 PR \u63cf\u8ff0\u91cc\u6ca1\u6709\u5f15\u7528\u4efb\u4f55\u57fa\u7ebf\u53d8\u66f4\u4f9d\u636e<\/span>
    \nfor<\/span> comp,<\/span> ms in<\/span> pr.<\/span>items(<\/span>)<\/span>:<\/span>
    \n if<\/span> not<\/span> isinstance<\/span>(<\/span>ms,<\/span> dict<\/span>)<\/span>:<\/span>
    \n continue<\/span>
    \n for<\/span> m,<\/span> v in<\/span> ms.<\/span>items(<\/span>)<\/span>:<\/span>
    \n if<\/span> v <<\/span> BASELINES[<\/span>comp]<\/span>[<\/span>m]<\/span> and<\/span> f"<\/span>{<\/span>comp}<\/span><\/span>.<\/span>{<\/span>m}<\/span><\/span>"<\/span><\/span> not<\/span> in<\/span> pr_approval:<\/span>
    \n print<\/span>(<\/span>f"[reject] <\/span>{<\/span>comp}<\/span><\/span>.<\/span>{<\/span>m}<\/span><\/span> \u57fa\u7ebf\u88ab\u4e0b\u8c03 <\/span>{<\/span>BASELINES[<\/span>comp]<\/span>[<\/span>m]<\/span>:<\/span>.2f<\/span>}<\/span><\/span> -> <\/span>{<\/span>v:<\/span>.2f<\/span>}<\/span><\/span>，PR \u65e0\u5ba1\u6279\u8bb0\u5f55"<\/span><\/span>)<\/span>
    \nprint<\/span>(<\/span>"\u964d\u57fa\u7ebf\u89c6\u540c\u56de\u5f52：\u6bcf\u4e00\u6b21\u9000\u8ba9\u90fd\u5fc5\u987b\u53d8\u6210\u663e\u5f0f\u51b3\u7b56，\u800c\u4e0d\u662f diff \u91cc\u987a\u624b\u7684\u4e24\u884c\u6570\u5b57\u3002"<\/span>)<\/span><\/p>\n

    print<\/span>(<\/span>)<\/span>
    \nprint<\/span>(<\/span>"==== \u6c47\u603b：\u540c\u5c4f\u591a\u5239\u8f66，\u5404\u62e6\u4e00\u7c7b\u81ea\u6b3a ===="<\/span>)<\/span>
    \nprint<\/span>(<\/span>"macro_f1 \u62e6\u300c\u5c0f\u7c7b\u88ab\u5403\u6389\u300d | \u5206\u5c42\u95e8\u69db\u62e6\u300c\u5e73\u5747\u6570\u9a97\u672f\u300d | \u57fa\u7ebf diff \u62e6\u300c\u8ba9\u6d4b\u8bd5\u53d8\u7eff\u7684\u6377\u5f84\u300d"<\/span>)<\/span>
    \nprint<\/span>(<\/span>"\u4ee3\u7801\u6d4b\u8bd5\u8bc1\u660e\u300c\u6ca1\u5f04\u574f\u300d，\u8bc4\u6d4b\u95e8\u69db\u5ea6\u91cf\u300c\u53d8\u597d\u4e86\u300d\u3002\u7f3a\u4e00\u4e2a，\u7ea2\u7eff\u706f\u90fd\u53ea\u662f\u88c5\u9970\u3002"<\/span>)<\/span><\/p>\n

    \u8dd1\u8d77\u6765\u957f\u8fd9\u6837<\/h4>\n

    python eval_gate.py，\u771f\u5b9e\u8f93\u51fa\u539f\u6837\u8d34\u4e0a：<\/p>\n

    ==== \u610f\u56fe\u5206\u7c7b：accuracy \u4f1a\u9a97\u4eba，\u5b8f F1 \u4e0d\u4f1a ====
    \n\u6539\u52a8\u524d : top1 0.88 | macro_f1 0.86 | \u5206\u7c7b F1: explain 0.80 query 0.93 refactor 0.83
    \n\u89c4\u5219\u5fae\u8c03\u540e: top1 0.84 | macro_f1 0.80 | \u5206\u7c7b F1: explain 0.67 query 0.90 refactor 0.83<\/p>\n

    [PASS] \u57fa\u7ebf\u590d\u6838 intent_rule.top1: 0.88（\u57fa\u7ebf 0.85）
    \n[PASS] \u57fa\u7ebf\u590d\u6838 intent_rule.macro_f1: 0.86（\u57fa\u7ebf 0.85）
    \n[PASS] \u91cd\u8bc4 intent_rule.top1: 0.84（\u57fa\u7ebf 0.85）
    \n[FAIL] \u91cd\u8bc4 intent_rule.macro_f1: 0.80 < \u57fa\u7ebf 0.85 – 2pp（\u5dee 0.05）\u2014\u2014\u6df7\u6dc6\u77e9\u9635\u89c1 reports\/<\/p>\n

    ==== \u68c0\u7d22：\u603b\u91cf\u4e0e\u5206\u5c42，\u4e00\u4e2a\u90fd\u4e0d\u80fd\u5c11\u770b ====
    \n[PASS] retrieval.recall@5: 0.82（\u57fa\u7ebf 0.80）
    \n[PASS] retrieval.mrr: 0.68（\u57fa\u7ebf 0.60）
    \n\u603b\u91cf\u8fbe\u6807\u2014\u2014\u5148\u522b\u6025\u7740\u5e86\u795d，\u5207\u4e00\u5200\u5206\u5c42：
    \n\u6309 project_type：data 0.83(3\u6761) | iot 0.58(3\u6761) | web 1.00(4\u6761)
    \n[FAIL] \u5206\u5c42\u95e8\u69db：iot \u7ec4 recall@5 0.58 < 0.70\u2014\u2014\u603b\u91cf\u8fbe\u6807\u4f46\u67d0\u7c7b\u5168\u70c2，\u89c6\u540c\u56de\u5f52<\/p>\n

    ==== \u5077\u61d2 diff：\u628a\u57fa\u7ebf 0.85 \u6539\u5230 0.80 \u8ba9 top1 "\u53d8\u7eff"？ ====
    \n[reject] intent_rule.top1 \u57fa\u7ebf\u88ab\u4e0b\u8c03 0.85 -> 0.80，PR \u65e0\u5ba1\u6279\u8bb0\u5f55
    \n\u964d\u57fa\u7ebf\u89c6\u540c\u56de\u5f52：\u6bcf\u4e00\u6b21\u9000\u8ba9\u90fd\u5fc5\u987b\u53d8\u6210\u663e\u5f0f\u51b3\u7b56，\u800c\u4e0d\u662f diff \u91cc\u987a\u624b\u7684\u4e24\u884c\u6570\u5b57\u3002<\/p>\n

    ==== \u6c47\u603b：\u540c\u5c4f\u591a\u5239\u8f66，\u5404\u62e6\u4e00\u7c7b\u81ea\u6b3a ====
    \nmacro_f1 \u62e6\u300c\u5c0f\u7c7b\u88ab\u5403\u6389\u300d | \u5206\u5c42\u95e8\u69db\u62e6\u300c\u5e73\u5747\u6570\u9a97\u672f\u300d | \u57fa\u7ebf diff \u62e6\u300c\u8ba9\u6d4b\u8bd5\u53d8\u7eff\u7684\u6377\u5f84\u300d
    \n\u4ee3\u7801\u6d4b\u8bd5\u8bc1\u660e\u300c\u6ca1\u5f04\u574f\u300d，\u8bc4\u6d4b\u95e8\u69db\u5ea6\u91cf\u300c\u53d8\u597d\u4e86\u300d\u3002\u7f3a\u4e00\u4e2a，\u7ea2\u7eff\u706f\u90fd\u53ea\u662f\u88c5\u9970\u3002<\/p>\n

    \u8f93\u51fa\u4f60\u770b\u4ec0\u4e48<\/h4>\n
  • \u5934\u4e24\u884c\u662f\u5168\u7bc7\u6700\u91cd\u8981\u7684\u4e00\u8bfe：\u540c\u4e00\u6b21\u6539\u52a8，top1 \u4ece 0.88 \u6389\u5230 0.84\u300c\u770b\u7740\u8fd8\u884c\u300d，\u5b8f F1 \u4ece 0.86 \u6389\u5230 0.80 \u76f4\u63a5\u7ea2\u3002\u770b\u5206\u7c7b F1 \u90a3\u4e00\u5217：explain 0.80 \u2192 0.67\u2014\u2014\u5c0f\u7c7b\u88ab query \u5403\u6389\u4e86\u4e00\u6761\u6837\u672c，\u5927\u7c7b\u51e0\u4e4e\u4e0d\u52a8\u3002\u5982\u679c\u53ea\u76ef accuracy，\u8fd9\u6b21\u52a3\u5316\u4f1a\u65e0\u58f0\u5408\u5165；\u5b8f F1 \u5bf9\u5c0f\u7c7b\u4e00\u89c6\u540c\u4ec1，\u62fd\u4e0d\u4f4f\u7684\u5c31\u662f\u62fd\u4e0d\u4f4f\u3002<\/li>\n
  • \u57fa\u7ebf\u590d\u6838\u90a3\u4e24\u884c\u4e0d\u662f\u5e9f\u8bdd：\u5148\u7528\u6539\u52a8\u524d\u7684\u6570\u636e\u9a8c\u8bc1\u300c\u57fa\u7ebf\u672c\u8eab\u7ad9\u5f97\u4f4f\u300d（0.88\/0.86 \u90fd\u8fc7），\u518d\u5bf9\u6539\u52a8\u540e\u91cd\u8bc4\u3002\u8df3\u8fc7\u590d\u6838\u76f4\u63a5\u91cd\u8bc4，\u4f60\u53ef\u80fd\u62ff\u4e00\u5957\u672c\u6765\u5c31\u8fbe\u4e0d\u4e86\u7684\u57fa\u7ebf\u5426\u51b3\u6240\u6709\u4eba\u3002<\/li>\n
  • FAIL \u884c\u672b\u5c3e\u7684\u300c\u6df7\u6dc6\u77e9\u9635\u89c1 reports\/\u300d：\u62a5\u544a\u5f52\u6863\u5230\u5e26\u65f6\u95f4\u6233\u7684 JSON，\u5931\u8d25\u65ad\u8a00\u76f4\u63a5\u6307\u5411\u73b0\u573a\u2014\u2014\u6392\u969c\u4e0d\u7528\u91cd\u8dd1\u3002\u8fd9\u548c\u7b2c 24 \u7bc7\u5931\u8d25\u5f52\u6863\u662f\u540c\u4e00\u4e2a\u54f2\u5b66：\u62a5\u9519\u4fe1\u606f\u662f\u7ed9\u7528\u6237\u770b\u7684\u5bfc\u822a，\u4e0d\u662f\u7ed9\u7528\u6237\u770b\u7684\u5899\u3002<\/li>\n
  • \u68c0\u7d22\u6bb5\u5148\u7eff\u540e\u7ea2：\u603b\u91cf recall@5 0.82\u3001MRR 0.68 \u90fd\u8fc7\u7ebf，\u4ee3\u7801\u4e5f\u4f53\u8d34\u5730\u63d0\u9192\u4f60\u300c\u5148\u522b\u6025\u7740\u5e86\u795d\u300d\u2014\u2014\u5207\u5206\u5c42\u540e iot 0.58 \u73b0\u51fa\u539f\u5f62\u3002\u4e09\u6761 iot query \u91cc\u4e24\u6761\u628a\u76f8\u5173\u6587\u6863\u6392\u5230 5 \u540d\u5f00\u5916，\u5168\u5c40\u5341\u6761\u4e00\u5e73\u5747，\u4ec0\u4e48\u90fd\u770b\u4e0d\u51fa\u6765\u3002\u5206\u5c42\u95e8\u69db\u6bd4\u5168\u5c40\u95e8\u69db\u7d27\u4e00\u622a，\u5c31\u662f\u9632\u5e73\u5747\u6570\u9a97\u672f\u7684\u3002<\/li>\n
  • \u6700\u540e\u4e00\u6bb5\u5239\u8f66\u9632\u7684\u662f\u4eba：\u6539\u52a8\u6ca1\u8fc7\u95e8\u69db，\u6709\u4eba\u52a8\u4e86\u57fa\u7ebf\u7684\u8111\u7b4b\u2014\u2014top1 0.85 \u2192 0.80 \u8fd9\u79cd diff \u4e00\u773c\u5047，\u4f46\u300c\u987a\u624b\u5fae\u8c03\u57fa\u7ebf 0.01\u300d\u6df7\u5728\u5927 PR \u91cc\u5c31\u5f88\u96be\u8bf4\u4e86\u3002\u6240\u4ee5\u964d\u57fa\u7ebf\u4e0d\u770b\u5e45\u5ea6\u3001\u770b\u5ba1\u6279：pr_approval \u91cc\u6ca1\u6709\u8bb0\u5f55，\u4e00\u5f8b reject\u3002\u95e8\u69db\u9632\u7684\u4ece\u6765\u4e0d\u662f\u6076\u610f，\u662f\u6bcf\u5468\u4e09\u4e0b\u5348\u4e94\u70b9\u534a\u7684\u75b2\u60eb\u3002<\/li>\n

    \u8bfb\u5230\u8fd9\u91cc，\u4e0d\u59a8\u7acb\u523b\u505a\u4e00\u4e2a\u4e94\u5206\u949f\u52a8\u4f5c：\u6253\u5f00\u4f60\u4eec\u4ed3\u5e93，\u627e\u5230\u4e0e\u672c\u7bc7\u5bf9\u5e94\u7684\u90a3\u4e2a\u76ee\u5f55（\u6ca1\u6709\u5c31\u65b0\u5efa），\u628a\u6587\u4e2d\u6700\u5c0f\u793a\u610f\u843d\u6210\u771f\u5b9e\u6587\u4ef6\u540d\u3002\u6587\u4ef6\u5728，\u8ba8\u8bba\u624d\u6709\u5750\u6807；\u6587\u4ef6\u4e0d\u5728，\u4e0b\u5468\u8fd8\u4f1a\u4ece\u300c\u6211\u4eec\u89c9\u5f97\u300d\u5f00\u59cb\u3002\u7b2c 26 \u7bc7\u7684\u9a8c\u6536\u4e0d\u662f\u300c\u8bfb\u5b8c\u300d，\u662f\u300c\u4ed3\u5e93\u91cc\u591a\u4e86\u4e00\u4efd\u53ef\u6307\u51fa\u7684\u4e1c\u897f\u300d\u3002<\/p>\n

    \u4ece\u6700\u5c0f\u53ef\u7528\u7248\u8d77\u6b65：tests\/eval\/baselines.json \u5148\u53ea\u9501\u4e00\u4e2a\u4f60\u6700\u6709\u628a\u63e1\u7684\u6307\u6807（\u6bd4\u5982\u610f\u56fe\u5206\u7c7b macro_f1），datasets\/ \u653e\u4e00\u4efd 30 \u6761\u4ee5\u5185\u3001\u4eba\u5de5\u6807\u6ce8\u5f97\u8d77\u7684 jsonl，\u95e8\u69db\u811a\u672c\u8fdb CI\u3002\u522b\u4e00\u4e0a\u6765\u94fa\u56db\u4e2a\u7ef4\u5ea6\u2014\u2014\u57fa\u7ebf\u7684\u7b2c\u4e00\u654c\u4e0d\u662f\u8bbe\u9519，\u662f\u8bbe\u5b8c\u6ca1\u4eba\u7ef4\u62a4\u3002<\/p>\n

    \u81ea\u68c0<\/h3>\n

    # \u8bc4\u6d4b\u95e8\u69db\u81ea\u68c0<\/p>\n

    – [ ] \u300c\u6539\u8fdb\u300d\u6709\u6ca1\u6709\u6539\u52a8\u524d\u9501\u5b9a\u7684\u7248\u672c\u5316\u57fa\u7ebf？
    \n– [ ] \u91cd\u8bc4\u4e4b\u524d，\u57fa\u7ebf\u672c\u8eab\u5148\u590d\u6838\u8fc7\u5417？
    \n– [ ] \u95e8\u7981\u6307\u6807\u5bf9\u5c0f\u7c7b\u516c\u5e73\u5417（\u5b8f F1 \/ \u9010\u7c7b）？
    \n– [ ] \u5206\u5c42\u81f3\u5c11\u5207\u8fc7\u4e00\u5200\u2014\u2014\u603b\u91cf\u8fbe\u6807\u67d0\u7c7b\u5168\u70c2\u7b97\u56de\u5f52\u5417？
    \n– [ ] \u964d\u57fa\u7ebf\u5728 CI \u91cc\u4f1a\u89e6\u53d1\u7ea2\u706f（\u987b\u8d70 PR \u5ba1\u6279）\u5417？<\/p>\n

    \u5408\u4e0a\u672c\u6587\u81ea\u95ee\u56db\u4e2a\u3002\u4e00，\u4f60\u7684\u300c\u6539\u8fdb\u300d\u6709\u6539\u52a8\u524d\u9501\u5b9a\u7684\u57fa\u7ebf\u5417，\u8fd8\u662f\u6bcf\u6b21\u90fd\u73b0\u573a\u627e\u4e00\u7ec4\u597d\u770b\u7684\u6570\u636e？\u4e8c，\u4f60\u7684\u95e8\u7981\u6307\u6807\u5bf9\u5c0f\u7c7b\u522b\u516c\u5e73\u5417\u2014\u2014accuracy \u4e3b\u5bfc\u7684\u62a5\u8868\u91cc，\u6700\u5c0f\u7684\u90a3\u4e2a\u7c7b\u591a\u4e45\u6ca1\u88ab\u5355\u72ec\u770b\u8fc7\u4e86？\u4e09，\u5206\u5c42\u5207\u8fc7\u4e00\u5200\u5417？\u6309\u9879\u76ee\u7c7b\u578b\u3001\u6309\u610f\u56fe\u3001\u6309\u5ba2\u6237\u89c4\u6a21，\u4efb\u9009\u4e00\u5200\u3002\u56db，\u6539\u57fa\u7ebf\u8fd9\u4e2a\u52a8\u4f5c，\u5728\u4f60\u7684 CI \u91cc\u4f1a\u89e6\u53d1\u7ea2\u706f\u5417？\u56db\u95ee\u6709\u4e00\u4e2a\u7b54\u4e0d\u4e0a，\u4f60\u7684\u300c\u53d8\u597d\u4e86\u300d\u5c31\u8fd8\u662f\u611f\u89c9\u3002<\/p>\n

    \u601d\u8003<\/h3>\n

    \u5c0f\u7ed3\u4e09\u53e5\u3002\u6307\u6807\u9009\u62e9\u8ddf\u7740\u4efb\u52a1\u5f62\u6001\u8d70：\u53ea\u8981\u4e00\u6761\u6b63\u786e\u7b54\u6848\u5c31\u7528 MRR，\u7c7b\u522b\u4e0d\u5747\u8861\u5c31\u7528\u5b8f F1，\u9632\u5e73\u5747\u6570\u9a97\u672f\u5c31\u52a0\u5206\u5c42\u7edf\u8ba1\u3002\u57fa\u7ebf\u5fc5\u987b\u7248\u672c\u5316\u4e14\u53ea\u5347\u4e0d\u964d，\u964d\u8981\u8d70\u663e\u5f0f\u6d41\u7a0b\u3002\u8bc4\u6d4b\u5206\u7ea7（PR \u8f7b\u95e8\u69db + nightly \u5168\u91cf）\u662f\u7eaa\u5f8b\u548c\u901f\u5ea6\u4e4b\u95f4\u7684\u901a\u7528\u5e73\u8861\u89e3\u3002<\/p>\n

    \u8fd9\u5957\u4f53\u7cfb\u771f\u6b63\u7684\u4ea7\u51fa\u4e0d\u662f\u90a3\u51e0\u4e2a\u6570\u5b57，\u662f\u5f00\u53d1\u59ff\u6001\u7684\u8f6c\u53d8：\u52a8\u624b\u6539\u68c0\u7d22\u4e4b\u524d\u5148\u60f3\u300c\u54ea\u4e2a\u6307\u6807\u8be5\u52a8\u3001\u54ea\u4e2a\u4e0d\u80fd\u52a8\u300d，\u6539\u5b8c\u770b\u65ad\u8a00\u800c\u4e0d\u662f\u770b\u611f\u89c9\u3002\u5f53\u300c\u4e0d\u80fd\u6084\u6084\u53d8\u5dee\u300d\u4ece\u56e2\u961f\u9ed8\u5951\u53d8\u6210\u673a\u5668\u6267\u884c\u7684\u4e8b\u5b9e，\u4f60\u624d\u914d\u8c08\u6301\u7eed\u6539\u8fdb\u3002<\/p>\n

    \u53ef\u4ee5\u7406\u89e3\u4e3a：\u7b2c 25 \u7bc7\u7684\u7ea2\u706f\u5582\u56de\u4fdd\u8bc1\u4e86\u300c\u9519\u4e00\u6b21\u5b66\u4e00\u6b21\u300d，\u672c\u7bc7\u7684\u95e8\u69db\u4fdd\u8bc1\u4e86\u300c\u5b66\u5b8c\u771f\u53d8\u597d\u300d\u3002\u4ee3\u7801\u6d4b\u8bd5\u4fdd\u8bc1\u5bf9，\u8bc4\u6d4b\u95e8\u69db\u5ea6\u91cf\u597d\u2014\u2014\u5bf9\u800c\u4e0d\u597d，\u662f\u50f5\u5316；\u597d\u800c\u4e0d\u5bf9，\u662f\u7384\u5b66\u3002\u7ec8\u7ae0\u8981\u628a\u6574\u6761\u8def\u6536\u62e2\u6210\u4e00\u5f20\u6e05\u5355：\u52fe\u5b8c\u5b83，\u624d\u7b97\u843d\u5730\u3002<\/p>\n

    \n

    \u4e0b\u7bc7\u9884\u544a：\u300a\u4e0d\u4f1a\u8fd8\u6709\u4eba\u5b66\u5b8c\u8fd8\u89c9\u5f97\u300c\u611f\u89c9\u7528\u4e0a\u4e86 AI\u300d\u5c31\u7b97\u843d\u5730\u5427？\u6e05\u5355\u6253\u5206\u4e0d\u8ba4\u611f\u89c9 | \u5de5\u7a0b\u5316\u7ec8\u7ae0\u300b\u2014\u2014\u7b2c 27 \u7bc7\u628a 26 \u7ad9\u7684\u9489\u5b50\u6536\u62e2\u6210\u53ef\u5b63\u5ea6\u91cd\u52fe\u7684\u6e05\u5355，\u7ea2\u7ebf\u6f0f\u4e00\u6761，\u603b\u5206\u6e05\u96f6\u3002<\/p>\n

    \u89c9\u5f97\u6709\u7528\u7684\u8bdd，\u6b22\u8fce\u70b9\u8d5e\u3001\u7559\u8a00\u6652\u4e00\u6652\u4f60\u88ab\u300c\u5e73\u5747\u6570\u9a97\u672f\u300d\u5751\u8fc7\u7684\u73b0\u573a；\u6709\u300c\u57fa\u7ebf\u88ab\u540c\u4e8b\u987a\u624b\u6539\u5c0f\u300d\u6545\u4e8b\u7684，\u66f4\u6b22\u8fce\u5c55\u5f00\u8bb2\u8bb2\u3002<\/p>\n

    \u4e13\u680f：\u4ece\u5c0f\u767d\u5230AI\u843d\u5730\u5927\u5e08｜\u300a\u4ece Rule \u5230\u843d\u5730：AI \u5de5\u7a0b\u5316\u5b9e\u8df5\u300b<\/p>\n","protected":false},"excerpt":{"rendered":"

    \u4e0a\u4e00\u7bc7：\u4e0d\u4f1a\u8fd8\u6709\u4eba\u7ea2\u706f\u53ea\u4f1a\u91cd\u8dd1\u548c\u7529\u9505\u5427？\u5047\u7eff\u81ea\u52a8\u4fee\u662f\u81ea\u6740 | \u7ea2\u706f\u5582\u56de\u95ed\u73af
    \n\u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427？\u57fa\u7ebf\u4e0d\u5728，\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db \u7cfb\u5217：\u300a\u4ece Rule \u5230\u843d\u5730：AI \u5de5\u7a0b\u5316\u5b9e\u8df5\u300b\u7b2c 26 \u7bc7 \u6807\u7b7e：\u8bc4\u6d4b \u00b7 \u56de\u5f52\u57fa\u7ebf \u00b7 \u5b8f F1 \u00b7 RecallK \u00b7 MRR \u4e13\u680f：\u4ece\u5c0f\u767d\u5230AI\u843d\u5730\u5927\u5e08 \u5e8f\u8a00
    \n\u6d4b\u8bd5\u5f27\u8d70\u5b8c\u4e09\u7ad9：\u7b2c 23 <\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[50,3684,62],"topic":[],"class_list":["post-120038","post","type-post","status-publish","format-standard","hentry","category-server","tag-50","tag-3684","tag-62"],"yoast_head":"\n26-\u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427\uff1f\u57fa\u7ebf\u4e0d\u5728\uff0c\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/120038.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"26-\u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427\uff1f\u57fa\u7ebf\u4e0d\u5728\uff0c\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u4e0a\u4e00\u7bc7：\u4e0d\u4f1a\u8fd8\u6709\u4eba\u7ea2\u706f\u53ea\u4f1a\u91cd\u8dd1\u548c\u7529\u9505\u5427？\u5047\u7eff\u81ea\u52a8\u4fee\u662f\u81ea\u6740 | \u7ea2\u706f\u5582\u56de\u95ed\u73af \u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427？\u57fa\u7ebf\u4e0d\u5728，\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db \u7cfb\u5217：\u300a\u4ece Rule \u5230\u843d\u5730：AI \u5de5\u7a0b\u5316\u5b9e\u8df5\u300b\u7b2c 26 \u7bc7 \u6807\u7b7e：\u8bc4\u6d4b \u00b7 \u56de\u5f52\u57fa\u7ebf \u00b7 \u5b8f F1 \u00b7 RecallK \u00b7 MRR \u4e13\u680f：\u4ece\u5c0f\u767d\u5230AI\u843d\u5730\u5927\u5e08 \u5e8f\u8a00 \u6d4b\u8bd5\u5f27\u8d70\u5b8c\u4e09\u7ad9：\u7b2c 23\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/120038.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-30T15:24:31+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"7 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/120038.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/120038.html\",\"name\":\"26-\u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427\uff1f\u57fa\u7ebf\u4e0d\u5728\uff0c\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-10-30T15:24:31+00:00\",\"dateModified\":\"2026-10-30T15:24:31+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/120038.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/120038.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/120038.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"26-\u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427\uff1f\u57fa\u7ebf\u4e0d\u5728\uff0c\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"26-\u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427\uff1f\u57fa\u7ebf\u4e0d\u5728\uff0c\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/120038.html","og_locale":"zh_CN","og_type":"article","og_title":"26-\u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427\uff1f\u57fa\u7ebf\u4e0d\u5728\uff0c\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u4e0a\u4e00\u7bc7：\u4e0d\u4f1a\u8fd8\u6709\u4eba\u7ea2\u706f\u53ea\u4f1a\u91cd\u8dd1\u548c\u7529\u9505\u5427？\u5047\u7eff\u81ea\u52a8\u4fee\u662f\u81ea\u6740 | \u7ea2\u706f\u5582\u56de\u95ed\u73af \u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427？\u57fa\u7ebf\u4e0d\u5728，\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db \u7cfb\u5217：\u300a\u4ece Rule \u5230\u843d\u5730：AI \u5de5\u7a0b\u5316\u5b9e\u8df5\u300b\u7b2c 26 \u7bc7 \u6807\u7b7e：\u8bc4\u6d4b \u00b7 \u56de\u5f52\u57fa\u7ebf \u00b7 \u5b8f F1 \u00b7 RecallK \u00b7 MRR \u4e13\u680f：\u4ece\u5c0f\u767d\u5230AI\u843d\u5730\u5927\u5e08 \u5e8f\u8a00 \u6d4b\u8bd5\u5f27\u8d70\u5b8c\u4e09\u7ad9：\u7b2c 23","og_url":"https:\/\/www.wsisp.com\/helps\/120038.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-10-30T15:24:31+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"7 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/120038.html","url":"https:\/\/www.wsisp.com\/helps\/120038.html","name":"26-\u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427\uff1f\u57fa\u7ebf\u4e0d\u5728\uff0c\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-10-30T15:24:31+00:00","dateModified":"2026-10-30T15:24:31+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/120038.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/120038.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/120038.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"26-\u4e0d\u4f1a\u8fd8\u6709\u4eba\u62ff\u300c\u611f\u89c9\u53d8\u597d\u4e86\u300d\u5f53\u6539\u8fdb\u5427\uff1f\u57fa\u7ebf\u4e0d\u5728\uff0c\u6539\u8fdb\u4e0d\u7b97 | \u6548\u679c\u8bc4\u6d4b\u4e0e\u56de\u5f52\u95e8\u69db"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/120038","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=120038"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/120038\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=120038"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=120038"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=120038"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=120038"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}