{"id":10006803,"date":"2022-11-04T10:36:09","date_gmt":"2022-11-04T02:36:09","guid":{"rendered":"https:\/\/ityet.com\/?p=10006803"},"modified":"2022-11-04T10:36:09","modified_gmt":"2022-11-04T02:36:09","slug":"easynlp%e5%8f%91%e5%b8%83%e8%9e%8d%e5%90%88%e8%af%ad%e8%a8%80%e5%ad%a6%e5%92%8c%e4%ba%8b%e5%ae%9e%e7%9f%a5%e8%af%86%e7%9a%84%e4%b8%ad%e6%96%87%e9%a2%84%e8%ae%ad%e7%bb%83%e6%a8%a1%e5%9e%8bckbert","status":"publish","type":"post","link":"https:\/\/itmax.cn\/?p=10006803","title":{"rendered":"EasyNLP\u53d1\u5e03\u878d\u5408\u8bed\u8a00\u5b66\u548c\u4e8b\u5b9e\u77e5\u8bc6\u7684\u4e2d\u6587\u9884\u8bad\u7ec3\u6a21\u578bCKBERT"},"content":{"rendered":"

\u7b80\u4ecb\uff1a\u00a0\u672c\u2f42\u7b80\u8981\u4ecb\u7ecdCKBERT\u7684\u6280\u672f\u89e3\u8bfb\uff0c\u4ee5\u53ca\u5982\u4f55\u5728EasyNLP\u6846\u67b6\u3001HuggingFace Models\u548c\u963f\u91cc\u4e91\u673a\u5668\u5b66\u4e60\u5e73\u53f0PAI\u4e0a\u4f7f\u2f64CKBERT\u6a21\u578b\u3002\u5bfc\u8bfb\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u5728NLP\u7684\u5404\u4e2a\u5e94\u7528\u4e2d\u90fd\u6709\u53ca\u5176\u5e7f\u6cdb\u7684\u5e94\u7528\uff1b\u7136\u800c\uff0c\u7ecf\u5178\u7684\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\uff08\u4f8b\u5982BERT\uff09\u7f3a\u4e4f\u5bf9\u77e5\u8bc6\u7684\u7406\u89e3\uff0c\u4f8b\u5982\u77e5\u8bc6\u56fe\u8c31\u4e2d\u7684\u5173\u7cfb\u4e09\u5143\u7ec4\u3002\u77e5\u8bc6\u589e\u5f3a\u9884\u8bad\u7ec3\u6a21\u578b\u4f7f\u7528\u5916\u90e8\u77e5\u8bc6\uff08\u77e5\u8bc6\u56fe\u8c31\uff0c\u5b57\u5178\u548c\u6587\u672c\u7b49\uff09\u6216\u8005\u53e5\u5b50\u5185\u90e8\u7684\u8bed\u8a00\u5b66\u77e5\u8bc6\u8fdb\u884c\u589e\u5f3a\u3002\u6211\u4eec\u53d1\u73b0\uff0c\u77e5\u8bc6\u6ce8\u5165\u7684\u8fc7\u7a0b\u90fd\u4f34\u968f\u7740\u5f88\u5927\u89c4\u6a21\u7684\u77e5\u8bc6\u53c2\u6570\uff0c\u4e0b\u6e38\u4efb\u52a1fine-tune\u7684\u65f6\u5019\u4ecd\u7136\u9700\u8981\u5916\u90e8\u6570\u636e\u7684\u652f\u6491\u624d\u80fd\u8fbe\u5230\u6bd4\u8f83\u597d\u7684\u6548\u679c\uff0c\u4ece\u800c\u65e0\u6cd5\u5728\u4e91\u73af\u5883\u4e2d\u5f88\u597d\u7684\u63d0\u4f9b\u7ed9\u7528\u6237\u8fdb\u884c\u4f7f\u7528\u3002CKBERT\uff08Chinese Knowledge-enhanced BERT\uff09\u662fEasyNLP\u56e2\u961f\u81ea\u7814\u7684\u4e2d\u6587\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u7ed3\u5408\u4e86\u4e24\u79cd\u77e5\u8bc6\u7c7b\u578b\uff08\u5916\u90e8\u77e5\u8bc6\u56fe\u8c31\uff0c\u5185\u90e8\u8bed\u8a00\u5b66\u77e5\u8bc6\uff09\u5bf9\u6a21\u578b\u8fdb\u884c\u77e5\u8bc6\u6ce8\u5165\uff0c\u540c\u65f6\u4f7f\u5f97\u77e5\u8bc6\u6ce8\u5165\u7684\u65b9\u5f0f\u65b9\u4fbf\u6a21\u578b\u53ef\u6269\u5c55\u3002\u6211\u4eec\u7684\u5b9e\u9a8c\u9a8c\u8bc1\u4e5f\u8868\u660eCKBERT\u7684\u6a21\u578b\u7cbe\u5ea6\u8d85\u8d8a\u4e86\u591a\u79cd\u7ecf\u5178\u4e2d\u6587\u6a21\u578b\u3002\u5728\u672c\u6b21\u7684\u6846\u67b6\u5347\u7ea7\u4e2d\uff0c\u6211\u4eec\u5c06\u591a\u79cd\u89c4\u6a21\u7684CKBERT\u6a21\u578b\u8d21\u732e\u7ed9\u5f00\u6e90\u793e\u533a\uff0c\u5e76\u4e14\u8fd9\u4e9bCKBERT\u6a21\u578b\u4e0eHuggingFace Models\u5b8c\u5168\u517c\u5bb9\u3002\u6b64\u5916\uff0c\u7528\u6237\u4e5f\u53ef\u4ee5\u5728\u963f\u91cc\u4e91\u673a\u5668\u5b66\u4e60\u5e73\u53f0PAI\u4e0a\u65b9\u4fbf\u5730\u5229\u7528\u4e91\u8d44\u6e90\u4f7f\u7528CKBERT\u6a21\u578b\u3002\u00a0EasyNLP\uff08https:\/\/github.com\/alibaba\/Ea…<\/a>\uff09\u662f\u963f\u2fa5\u4e91\u673a\u5668\u5b66\u4e60PAI \u56e2\u961f\u57fa\u4e8e PyTorch \u5f00\u53d1\u7684\u6613\u2f64\u4e14\u4e30\u5bcc\u7684\u4e2d\u2f42NLP\u7b97\u6cd5\u6846\u67b6\uff0c\u2f40\u6301\u5e38\u2f64\u7684\u4e2d\u2f42\u9884\u8bad\u7ec3\u6a21\u578b\u548c\u2f24\u6a21\u578b\u843d\u5730\u6280\u672f\uff0c\u5e76\u4e14\u63d0\u4f9b\u4e86\u4ece\u8bad\u7ec3\u5230\u90e8\u7f72\u7684\u2f00\u7ad9\u5f0f NLP \u5f00\u53d1\u4f53\u9a8c\u3002EasyNLP \u63d0\u4f9b\u4e86\u7b80\u6d01\u7684\u63a5\u2f1d\u4f9b\u2f64\u6237\u5f00\u53d1 NLP \u6a21\u578b\uff0c\u5305\u62ecNLP\u5e94\u2f64 AppZoo \u548c\u9884\u8bad\u7ec3 ModelZoo\uff0c\u540c\u65f6\u63d0\u4f9b\u6280\u672f\u5e2e\u52a9\u2f64\u6237\u2fbc\u6548\u7684\u843d\u5730\u8d85\u2f24\u9884\u8bad\u7ec3\u6a21\u578b\u5230\u4e1a\u52a1\u3002\u7531\u4e8e\u8de8\u6a21\u6001\u7406\u89e3\u9700\u6c42\u7684\u4e0d\u65ad\u589e\u52a0\uff0cEasyNLP\u4e5f\u2f40\u6301\u5404\u79cd\u8de8\u6a21\u6001\u6a21\u578b\uff0c\u7279\u522b\u662f\u4e2d\u2f42\u9886\u57df\u7684\u8de8\u6a21\u6001\u6a21\u578b\uff0c\u63a8\u5411\u5f00\u6e90\u793e\u533a\uff0c\u5e0c\u671b\u80fd\u591f\u670d\u52a1\u66f4\u591a\u7684 NLP \u548c\u591a\u6a21\u6001\u7b97\u6cd5\u5f00\u53d1\u8005\u548c\u7814 \u7a76\u8005\uff0c\u4e5f\u5e0c\u671b\u548c\u793e\u533a\u2f00\u8d77\u63a8\u52a8 NLP \/\u591a\u6a21\u6001\u6280\u672f\u7684\u53d1\u5c55\u548c\u6a21\u578b\u843d\u5730\u3002\u00a0\u672c\u2f42\u7b80\u8981\u4ecb\u7ecdCKBERT\u7684\u6280\u672f\u89e3\u8bfb\uff0c\u4ee5\u53ca\u5982\u4f55\u5728EasyNLP\u6846\u67b6\u3001HuggingFace Models\u548c\u963f\u91cc\u4e91\u673a\u5668\u5b66\u4e60\u5e73\u53f0PAI\u4e0a\u4f7f\u2f64CKBERT\u6a21\u578b\u3002\u4e2d\u6587\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u6982\u89c8\u5728\u8fd9\u4e00\u8282\u4e2d\uff0c\u6211\u4eec\u9996\u5148\u7b80\u8981\u56de\u987e\u7ecf\u5178\u7684\u4e2d\u6587\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u3002\u76ee\u524d\u4e2d\u6587\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u4e3b\u8981\u5305\u62ec\u4e86\u4e24\u79cd\u7c7b\u578b\uff1a\u901a\u7528\u9886\u57df\u7684\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\uff0c\u4e3b\u8981\u5305\u62ec\u4e86BERT\u3001MacBERT\u548cPERT\u7b49\u6a21\u578b\uff1b\u77e5\u8bc6\u589e\u5f3a\u7684\u4e2d\u6587\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u4e3b\u8981\u5305\u62ec\u4e86ERNIE-baidu\uff0cLattice-BERT\uff0cK-BERT\u548cERNIE-THU\u7b49\u6a21\u578b\u3002\u901a\u7528\u9886\u57df\u7684\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578bBERT\u76f4\u63a5\u4f7f\u7528Google\u53d1\u5e03\u7684\u57fa\u4e8e\u4e2d\u6587\u7ef4\u57fa\u6587\u672c\u8bed\u6599\u8fdb\u884c\u8bad\u7ec3\u7684\u6a21\u578b\u3002MacBERT\u662fBERT\u7684\u6539\u8fdb\u7248\u672c\uff0c\u5f15\u5165\u4e86\u7ea0\u9519\u578b\u63a9\u7801\u8bed\u8a00\u6a21\u578b\uff08MLM as correction\uff0cMac\uff09\u9884\u8bad\u7ec3\u4efb\u52a1\uff0c\u7f13\u89e3\u4e86\u201c\u9884\u8bad\u7ec3-\u4e0b\u6e38\u4efb\u52a1\u201d\u4e0d\u4e00\u81f4\u7684\u95ee\u9898\u3002\u5728\u63a9\u7801\u8bed\u8a00\u6a21\u578b\uff08MLM\uff09\u4e2d\uff0c\u5f15\u5165\u4e86[MASK]\u6807\u8bb0\u8fdb\u884c\u63a9\u7801\uff0c\u4f46[MASK]\u6807\u8bb0\u5e76\u4e0d\u4f1a\u51fa\u73b0\u5728\u4e0b\u6e38\u4efb\u52a1\u4e2d\u3002\u5728MacBERT\u4e2d\uff0c\u4f7f\u7528\u76f8\u4f3c\u8bcd\u6765\u53d6\u4ee3[MASK]\u6807\u8bb0\u3002\u76f8\u4f3c\u8bcd\u901a\u8fc7Synonyms toolkit \u5de5\u5177\u83b7\u53d6\uff0c\u7b97\u6cd5\u57fa\u4e8eword2vec\u76f8\u4f3c\u5ea6\u8ba1\u7b97\u3002\u540c\u65f6\uff0cMacBERT\u4e5f\u5f15\u5165\u4e86Whole Word Masking\u548cN-gram Masking\u6280\u672f\u3002\u5f53\u8981\u5bf9N-gram\u8fdb\u884c\u63a9\u7801\u65f6\uff0c\u4f1a\u5bf9N-gram\u91cc\u7684\u6bcf\u4e2a\u8bcd\u5206\u522b\u67e5\u627e\u76f8\u4f3c\u8bcd\uff1b\u5f53\u6ca1\u6709\u76f8\u4f3c\u8bcd\u53ef\u66ff\u6362\u65f6\uff0c\u5c06\u4f7f\u7528\u968f\u673a\u8bcd\u8fdb\u884c\u66ff\u6362\u3002\u7531\u4e8e\u4e00\u5b9a\u7a0b\u5ea6\u7684\u4e71\u5e8f\u6587\u672c\u4e0d\u5f71\u54cd\u8bed\u4e49\u7406\u89e3\uff0cPBERT\u4ece\u4e71\u5e8f\u6587\u672c\u4e2d\u5b66\u4e60\u8bed\u4e49\u77e5\u8bc6\u3002\u5b83\u5bf9\u539f\u59cb\u8f93\u5165\u6587\u672c\u8fdb\u884c\u4e00\u5b9a\u7684\u8bcd\u5e8f\u8c03\u6362\uff0c\u4ece\u800c\u5f62\u6210\u4e71\u5e8f\u6587\u672c\uff08\u56e0\u6b64\u4e0d\u4f1a\u5f15\u5165\u989d\u5916\u7684[MASK]\u6807\u8bb0\uff09\uff0c\u5176\u5b66\u4e60\u76ee\u6807\u662f\u9884\u6d4b\u539fToken\u6240\u5728\u7684\u4f4d\u7f6e\u3002\u77e5\u8bc6\u589e\u5f3a\u7684\u4e2d\u6587\u9884\u8bad\u7ec3\u6a21\u578bBERT\u5728\u9884\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u4f7f\u7528\u7684\u6570\u636e\u4ec5\u662f\u5bf9\u5355\u4e2a\u5b57\u7b26\u8fdb\u884c\u5c4f\u853d\uff0c\u4f8b\u5982\u4e0b\u56fe\u6240\u793a\uff0c\u8bad\u7ec3BERT\u65f6\u901a\u8fc7\u201c\u54c8\u201d\u4e0e\u201c\u6ee8\u201d\u7684\u5c40\u90e8\u5171\u73b0\u5224\u65ad\u51fa\u201c\u5c14\u201d\u5b57\uff0c\u4f46\u662f\u6a21\u578b\u5176\u5b9e\u5e76\u6ca1\u6709\u5b66\u4e60\u5230\u4e0e\u201c\u54c8\u5c14\u6ee8\u201d\u76f8\u5173\u7684\u77e5\u8bc6\uff0c\u5373\u53ea\u662f\u5b66\u4e60\u5230\u201c\u54c8\u5c14\u6ee8\u201d\u8fd9\u4e2a\u8bcd\uff0c\u4f46\u662f\u5e76\u4e0d\u77e5\u9053\u201c\u54c8\u5c14\u6ee8\u201d\u6240\u4ee3\u8868\u7684\u542b\u4e49\u3002ERNIE-Baidu\u5728\u9884\u8bad\u7ec3\u65f6\u4f7f\u7528\u7684\u6570\u636e\u662f\u5bf9\u6574\u4e2a\u8bcd\u8fdb\u884c\u5c4f\u853d\uff0c\u4ece\u800c\u5b66\u4e60\u8bcd\u4e0e\u5b9e\u4f53\u7684\u8868\u8fbe\uff0c\u4f8b\u5982\u5c4f\u853d\u201c\u54c8\u5c14\u6ee8\u201d\u4e0e\u201c\u51b0\u96ea\u201d\u8fd9\u6837\u7684\u8bcd\uff0c\u4f7f\u6a21\u578b\u80fd\u591f\u5efa\u6a21\u51fa\u201c\u54c8\u5c14\u6ee8\u201d\u4e0e\u201c\u9ed1\u9f99\u6c5f\u201d\u7684\u5173\u7cfb\uff0c\u5b66\u5230\u201c\u54c8\u5c14\u6ee8\u201d\u662f\u201c\u9ed1\u9f99\u6c5f\u201d\u7684\u7701\u4f1a\u4ee5\u53ca\u201c\u54c8\u5c14\u6ee8\u201d\u662f\u4e2a\u51b0\u96ea\u57ce\u5e02\u8fd9\u6837\u7684\u542b\u4e49\u3002
\"\u56fe\u7247\"
\u4e0eERNIE-Baidu\u7c7b\u4f3c\uff0cLattice-BERT\u5229\u7528Word-Lattice\u7ed3\u6784\u6574\u5408\u8bcd\u7ea7\u522b\u4fe1\u606f\u3002\u5177\u4f53\u6765\u8bf4\uff0cLattice-BERT\u8bbe\u8ba1\u4e86\u4e00\u4e2aLattice\u4f4d\u7f6e\u6ce8\u610f\u673a\u5236\uff0c\u6765\u8868\u8fbe\u8bcd\u7ea7\u522b\u7684\u4fe1\u606f\uff0c\u540c\u65f6\u63d0\u51fa\u4e86Masked Segment Prediction\u7684\u9884\u6d4b\u4efb\u52a1\uff0c\u4ee5\u63a8\u52a8\u6a21\u578b\u5b66\u4e60\u6765\u81ea\u4e30\u5bcc\u4f46\u5197\u4f59\u7684\u5185\u5728Lattice\u4fe1\u606f\u3002
\"\u56fe\u7247\"
\u9664\u4e86\u8bed\u8a00\u5b66\u77e5\u8bc6\uff0c\u66f4\u591a\u7684\u5de5\u4f5c\u5229\u7528\u77e5\u8bc6\u56fe\u8c31\u4e2d\u7684\u4e8b\u5b9e\u6027\u77e5\u8bc6\u4e30\u5bcc\u4e2d\u6587\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u8868\u5f81\u3002\u5176\u4e2d\uff0cK-BERT\u63d0\u51fa\u4e86\u9762\u5411\u77e5\u8bc6\u56fe\u8c31\u7684\u77e5\u8bc6\u589e\u5f3a\u8bed\u8a00\u6a21\u578b\uff0c\u5c06\u4e09\u5143\u7ec4\u4f5c\u4e3a\u9886\u57df\u77e5\u8bc6\u6ce8\u5165\u5230\u53e5\u5b50\u4e2d\u3002\u7136\u800c\uff0c\u8fc7\u591a\u7684\u77e5\u8bc6\u878d\u5165\u4f1a\u5bfc\u81f4\u77e5\u8bc6\u566a\u97f3\uff0c\u4f7f\u53e5\u5b50\u504f\u79bb\u5176\u6b63\u786e\u7684\u542b\u4e49\u3002\u4e3a\u4e86\u514b\u670d\u77e5\u8bc6\u566a\u97f3, K-BERT\u5f15\u5165\u4e86Soft-position\u548cVisibel Matrix\u6765\u9650\u5236\u77e5\u8bc6\u7684\u5f71\u54cd\u3002\u7531\u4e8eK-BERT\u80fd\u591f\u4ece\u9884\u8bad\u7ec3\u7684BERT\u4e2d\u52a0\u8f7d\u6a21\u578b\u53c2\u6570\uff0c\u56e0\u6b64\u901a\u8fc7\u914d\u5907KG\uff0c\u53ef\u4ee5\u5f88\u5bb9\u6613\u5730\u5c06\u9886\u57df\u77e5\u8bc6\u6ce8\u5165\u5230\u6a21\u578b\u4e2d\uff0c\u800c\u4e0d\u9700\u8981\u5bf9\u6a21\u578b\u8fdb\u884c\u9884\u8bad\u7ec3\u3002EasyNLP\u6846\u67b6\u4e5f\u96c6\u6210\u4e86K-BERT\u7684\u6a21\u578b\u548c\u529f\u80fd\uff08\u770b\u8fd9\u91cc\uff09\u3002
\"\u56fe\u7247\"
ERNIE-THU\u662f\u4e00\u79cd\u878d\u5165\u77e5\u8bc6Embedding\u7684\u9884\u8bad\u7ec3\u6a21\u578b\u3002\u5b83\u9996\u5148\u4f7f\u7528TAGME\u63d0\u53d6\u6587\u672c\u4e2d\u7684\u5b9e\u4f53\uff0c\u5e76\u5c06\u8fd9\u4e9b\u5b9e\u4f53\u94fe\u6307\u5230KG\u4e2d\u7684\u5bf9\u5e94\u5b9e\u4f53\u5bf9\u8c61\uff0c\u7136\u540e\u83b7\u5f97\u8fd9\u4e9b\u5b9e\u4f53\u5bf9\u8c61\u5bf9\u5e94\u7684Embedding\u3002\u5b9e\u4f53\u5bf9\u8c61\u7684Embedding\u7531\u77e5\u8bc6\u8868\u793a\u65b9\u6cd5\uff08\u4f8b\u5982TransE\uff09\u8bad\u7ec3\u5f97\u5230\u3002\u6b64\u5916\uff0cERNIE-THU\u5728BERT\u6a21\u578b\u7684\u57fa\u7840\u4e0a\u8fdb\u884c\u6539\u8fdb\uff0c\u9664\u4e86MLM\u3001NSP\u4efb\u52a1\u5916\uff0c\u91cd\u65b0\u6dfb\u52a0\u4e86\u4e00\u4e2a\u548cKG\u76f8\u5173\u7684\u9884\u8bad\u7ec3\u76ee\u6807\uff1aMask\u6389Token\u548cEntity (\u5b9e\u4f53) \u7684\u5bf9\u9f50\u5173\u7cfb\uff0c\u5e76\u8981\u6c42\u6a21\u578b\u4ece\u56fe\u8c31\u7684\u5b9e\u4f53\u4e2d\u9009\u62e9\u5408\u9002\u7684Entity\u5b8c\u6210\u5bf9\u9f50\u3002
\"\u56fe\u7247\"
\u81ea\u7814CKBERT\u6a21\u578b\u6280\u672f\u8be6\u89e3\u7531\u4e8e\u5f53\u524d\u7684\u77e5\u8bc6\u589e\u5f3a\u9884\u8bad\u7ec3\u6a21\u578b\u5927\u90fd\u4f7f\u7528\u5916\u90e8\u77e5\u8bc6\uff08\u77e5\u8bc6\u56fe\u8c31\uff0c\u5b57\u5178\u548c\u6587\u672c\u7b49\uff09\u6216\u8005\u53e5\u5b50\u5185\u90e8\u7684\u8bed\u8a00\u5b66\u77e5\u8bc6\u8fdb\u884c\u589e\u5f3a\uff0c\u540c\u65f6\u77e5\u8bc6\u6ce8\u5165\u7684\u8fc7\u7a0b\u90fd\u4f34\u968f\u7740\u5f88\u5927\u89c4\u6a21\u7684\u77e5\u8bc6\u53c2\u6570\uff0c\u4e0b\u6e38\u4efb\u52a1fine-tune\u7684\u65f6\u5019\u4ecd\u7136\u9700\u8981\u5916\u90e8\u6570\u636e\u7684\u652f\u6491\u624d\u80fd\u8fbe\u5230\u6bd4\u8f83\u597d\u7684\u6548\u679c\uff0c\u4ece\u800c\u65e0\u6cd5\u5728\u4e91\u73af\u5883\u4e2d\u5f88\u597d\u7684\u63d0\u4f9b\u7ed9\u7528\u6237\u8fdb\u884c\u4f7f\u7528\u3002CKBERT\uff08Chinese Knowledge-enhanced BERT\uff09\u662fEasyNLP\u56e2\u961f\u81ea\u7814\u7684\u4e2d\u6587\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u7ed3\u5408\u4e86\u4e24\u79cd\u77e5\u8bc6\u7c7b\u578b\uff08\u5916\u90e8\u77e5\u8bc6\u56fe\u8c31\uff0c\u5185\u90e8\u8bed\u8a00\u5b66\u77e5\u8bc6\uff09\u5bf9\u6a21\u578b\u8fdb\u884c\u77e5\u8bc6\u6ce8\u5165\uff0c\u540c\u65f6\u4f7f\u5f97\u77e5\u8bc6\u6ce8\u5165\u7684\u65b9\u5f0f\u65b9\u4fbf\u6a21\u578b\u53ef\u6269\u5c55\u3002\u9488\u5bf9\u5b9e\u9645\u7684\u4e1a\u52a1\u9700\u6c42\uff0c\u6211\u4eec\u63d0\u4f9b\u4e86\u4e09\u79cd\u4e0d\u540c\u89c4\u6a21\u53c2\u6570\u91cf\u7684\u6a21\u578b\uff0c\u8be6\u7ec6\u914d\u7f6e\u5982\u4e0b\u6240\u793a\uff1a\u6a21\u578b\u914d\u7f6ealibaba-pai\/pai-ckbert-base-zhalibaba-pai\/pai-ckbert-large-zhalibaba-pai\/pai-ckbert-huge-zh\u53c2\u6570\u91cf\uff08Parameters\uff09151M428M1.3B\u5c42\u6570\uff08Number of Layers\uff09122424\u6ce8\u610f\u529b\u5934\u6570\uff08Attention Heads\uff0912168\u9690\u5411\u91cf\u7ef4\u5ea6\uff08Hidden Size\uff0976810242048\u6587\u672c\u957f\u5ea6\uff08Text Length\uff09128128128FFN \u5c42\u7ef4\u5ea6307240968192\u00a0CKBERT\u7684\u6a21\u578b\u67b6\u6784\u5982\u4e0b\u56fe\u6240\u793a\uff1a
\"\u56fe\u7247\"
\u4e3a\u4e86\u65b9\u4fbf\u6a21\u578b\u8fdb\u884c\u6269\u5c55\u53c2\u6570\uff0c\u6a21\u578b\u53ea\u5728\u6570\u636e\u8f93\u5165\u5c42\u9762\u548c\u9884\u8bad\u7ec3\u4efb\u52a1\u5c42\u9762\u8fdb\u884c\u4e86\u6539\u52a8\uff0c\u6ca1\u6709\u5bf9\u6a21\u578b\u67b6\u6784\u8fdb\u884c\u6539\u52a8\u3002\u56e0\u6b64\uff0cCKBERT\u7684\u6a21\u578b\u7ed3\u6784\u4e0e\u793e\u533a\u7248\u7684BERT\u6a21\u578b\u5bf9\u9f50\u3002\u5728\u6570\u636e\u8f93\u5165\u5c42\uff0c\u4e00\u5171\u8981\u5904\u7406\u4e24\u90e8\u5206\u7684\u77e5\u8bc6\uff0c\u5916\u90e8\u56fe\u8c31\u4e09\u5143\u7ec4\u548c\u53e5\u5b50\u7ea7\u5185\u90e8\u7684\u8bed\u8a00\u5b66\u77e5\u8bc6\u3002\u9488\u5bf9\u8bed\u8a00\u5b66\u77e5\u8bc6\uff0c\u6211\u4eec\u4f7f\u7528\u4e86\u54c8\u5de5\u5927LTP\u5e73\u53f0\u8fdb\u884c\u53e5\u5b50\u6570\u636e\u7684\u5904\u7406\uff0c\u8fdb\u884c\u8bed\u4e49\u89d2\u8272\u6807\u6ce8\u548c\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u7b49\uff0c\u7136\u540e\u6839\u636e\u89c4\u5219\uff0c\u5c06\u8bc6\u522b\u7ed3\u679c\u4e2d\u91cd\u8981\u7684\u6210\u5206\u8fdb\u884c\u6807\u6ce8\u3002\u9488\u5bf9\u5916\u90e8\u4e09\u5143\u7ec4\u77e5\u8bc6\u662f\u6839\u636e\u53e5\u5b50\u4e2d\u51fa\u73b0\u7684\u5b9e\u4f53\u6784\u9020\u5b9e\u4f53\u7684\u6b63\u8d1f\u4e09\u5143\u7ec4\u6837\u672c\uff0c\u6b63\u6837\u672c\u662f\u6839\u636e\u56fe\u8c31\u4e2d1-hop \u5b9e\u4f53\u8fdb\u884c\u7684\u91c7\u6837\uff0c\u8d1f\u6837\u672c\u662f\u6839\u636e\u56fe\u8c31\u4e2dmulti-hop\u8fdb\u884c\u7684\u91c7\u6837\uff0c\u4f46\u8d1f\u6837\u672c\u7684\u91c7\u6837\u8fc7\u7a0b\u53ea\u80fd\u5728\u89c4\u5b9a\u7684\u591a\u8df3\u8303\u56f4\u5185\uff0c\u800c\u4e0d\u80fd\u5728\u56fe\u8c31\u4e2d\u8ddd\u79bb\u592a\u8fdc\u3002\u00a0CKBERT\u91c7\u7528\u4e24\u79cd\u9884\u8bad\u7ec3\u4efb\u52a1\u8fdb\u884c\u6a21\u578b\u7684\u9884\u8bad\u7ec3\uff0c\u8bed\u8a00\u5b66\u611f\u77e5\u7684\u63a9\u7801\u8bed\u8a00\u6a21\u578b\u548c\u591a\u8df3\u77e5\u8bc6\u5bf9\u6bd4\u5b66\u4e60\uff1a\u8bed\u8a00\u5b66\u611f\u77e5\u7684\u63a9\u7801\u8bed\u8a00\u6a21\u578b\uff08Linguistic-aware MLM\uff09\uff1a\u5728\u8bed\u4e49\u4f9d\u5b58\u5173\u7cfb\u4e2d\u7684\u4e3b\u4f53\u89d2\u8272\uff08\u65bd\u4e8b\u8005AGT\u548c\u5f53\u4e8b\u8005EXP \uff09\u90e8\u5206\u7528[MASK]\u8fdb\u884c\u906e\u63a9\uff0c\u540c\u65f6\u5728\u8bcd\u7684\u524d\u540e\u90fd\u52a0\u4e0aSDP\uff0c\u9644\u52a0\u4e0a\u8bcd\u6c47\u7684\u8fb9\u754c\u4fe1\u606f\u3002\u5728\u4f9d\u5b58\u53e5\u6cd5\u5173\u7cfb\u4e2d\uff0c\u5c06\u4e3b\u8c13\u51b0\u5173\u7cfb\uff0c\u5b9a\u4e2d\u5173\u7cfb\uff0c\u5e76\u5217\u5173\u7cfb\u7b49\u6309\u7167\u4e0a\u8ff0mask\u673a\u5236\u8fdb\u884c\u5904\u7406\u4e3aDEP\u3002\u6574\u4f53\u8fdb\u884c\u9884\u8bad\u7ec3\u7684token\u6570\u91cf\u662f\u6574\u53e5\u8bdd\u768415%\uff0c\u5176\u4e2d40%\u8fdb\u884c\u968f\u673aMASK\uff0c30%\u548c30%\u5206\u914d\u5230\u8bed\u4e49\u4f9d\u5b58\u5173\u7cfb\u548c\u4f9d\u5b58\u53e5\u6cd5\u5173\u7cfb\u8bcd\u6c47\u4e0a\u6765\u3002\u635f\u5931\u51fd\u6570\u5982\u4e0b\uff1a
\"\u56fe\u7247\"
\u591a\u8df3\u77e5\u8bc6\u5bf9\u6bd4\u5b66\u4e60\uff1a\u5c06\u4e0a\u8ff0\u6784\u9020\u7684\u6b63\u8d1f\u6837\u672c\u6570\u636e\u9488\u5bf9\u8be5\u6ce8\u5165\u7684\u5b9e\u4f53\u8fdb\u884c\u5904\u7406\uff0c\u6bcf\u4e00\u4e2a\u53e5\u4e2d\u5b9e\u4f53\u6784\u90201\u4e2a\u6b63\u6837\u672c\uff0c4\u4e2a\u8d1f\u6837\u672c\uff0c\u901a\u8fc7\u6807\u51c6\u7684infoNCE\u635f\u5931\u4efb\u52a1\u8fdb\u884c\u5916\u90e8\u77e5\u8bc6\u7684\u5b66\u4e60\u3002\u635f\u5931\u51fd\u6570\u5982\u4e0b\uff1a
\"\u56fe\u7247\"
\u5176\u4e2d\uff0c
\"\u56fe\u7247\"
\u662f\u9884\u8bad\u7ec3\u6a21\u578b\u4ea7\u751f\u7684\u4e0a\u4e0b\u6587\u5b9e\u4f53\u8868\u793a\uff0c
\"\u56fe\u7247\"
\u8868\u793a\u6b63\u6837\u672c\u7684\u4e09\u5143\u7ec4\u8868\u793a\u7ed3\u679c\uff0c
\"\u56fe\u7247\"
\u8868\u793a\u8d1f\u6837\u672c\u7684\u4e09\u5143\u7ec4\u8868\u793a\u7ed3\u679c\u3002CKBERT\u6a21\u578b\u7684\u5b9e\u73b0\u5728EasyNLP\u6846\u67b6\u4e2d\uff0c\u6211\u4eec\u7684\u6a21\u578b\u5b9e\u73b0\u5206\u4e3a\u4e09\u4e2a\u90e8\u5206\uff1a\u6570\u636e\u9884\u5904\u7406\uff0c\u6a21\u578b\u7ed3\u6784\u5fae\u8c03\u548c\u635f\u5931\u51fd\u6570\u7684\u8bbe\u8ba1\u3002\u9996\u5148\uff0c\u5728\u6570\u636e\u9884\u5904\u7406\u73af\u8282\uff0c\u4e3b\u8981\u7531\u4ee5\u4e0b\u4e24\u4e2a\u6b65\u9aa4\u7ec4\u6210\uff1a1.NER\u5b9e\u4f53\u53ca\u8bed\u4e49\u5173\u7cfb\u7684\u63d0\u53d6\uff1b2.\u77e5\u8bc6\u56fe\u8c31\u7684\u4fe1\u606f\u6ce8\u5165\u3002\u5173\u4e8eNER\u5b9e\u4f53\u53ca\u8bed\u4e49\u4fe1\u606f\u7684\u63d0\u53d6\uff0c\u4e3b\u8981\u91c7\u7528LTP\uff08Language Technology Platform\uff09\u5bf9\u539f\u59cb\u53e5\u5b50\u8fdb\u884c\u5206\u8bcd\u548c\u53e5\u6cd5\u5206\u6790\uff0c\u8be5\u90e8\u5206\u7684\u6838\u5fc3\u4ee3\u7801\u5982\u4e0b\u6240\u793a\uff1adef ltp_process(ltp: LTP,<\/p>\n

            data: List[Dict[str, Union[str, List[Union[int, str]]]]]):\n"""use ltp to process the data\nArgs:\n    Dict ([str, str]): data\n    example:\n        {'text':['\u6211\u53eb\u6c64\u59c6\u53bb\u62ff\u4f1e\u3002'],...}\nReturns:\n    Dict[str, str]: result\n"""\nnew_data = list(map(lambda x:x['text'][0].replace(" ", ""), data))\nseg, hiddens = ltp.seg(new_data)\nresult = {}\nresult['seg'] = seg\nresult['ner'] = ltp.ner(hiddens)\nresult['dep'] = ltp.dep(hiddens)\nresult['sdp'] = ltp.sdp(hiddens)\nfor index in range(len(data)):\n    data[index]['text'][0] = data[index]['text'][0].replace(" ", "")\n    data[index]['seg'] = result['seg'][index]\n    data[index]['ner'] = result['ner'][index]\n    data[index]['dep'] = result['dep'][index]\n    data[index]['sdp'] = result['sdp'][index]\u8be5\u90e8\u5206\u5b8c\u6210\u4e4b\u540e\u9700\u8981\u57fa\u4e8e\u539f\u59cb\u53e5\u5b50\u4e2d\u7684\u8bed\u4e49\u4f9d\u5b58\u5173\u7cfb\u5bf9\u76f8\u5e94\u7684\u8bcd\u8fdb\u884c\u6574\u4f53\u7684mask\uff0c\u8be5\u90e8\u5206\u7684mask\u7b56\u7565\u53c2\u8003BERT\u7684mask\u7b56\u7565\u7684\u8bbe\u8ba1\uff0c\u7ed9\u4e0d\u540c\u7c7b\u578b\u7684\u5173\u7cfb\u5206\u914d\u7279\u5b9a\u7684\u6982\u7387\uff0c\u5e76\u57fa\u4e8e\u8be5\u6982\u7387\u5bf9\u4e0d\u540c\u7c7b\u578b\u5173\u7cfb\u8fdb\u884cmask\uff0c\u8be5\u90e8\u5206\u7684\u6838\u5fc3\u4ee3\u7801\u5982\u4e0b\uff1adef dep_sdp_mask(left_numbers: List[int], \n                 data_new: List[List[Union[int, str]]], \n                 markers_: List[List[int]], \n                 selected_numbers_: set, \n                 number_: int,\n                 marker_attrs: Dict[str, List[int]]) -> int:\n    """ mask the `mask_labels` for sdp and dep and record the maskers for each mask item\n    Args:\n        left_numbers (List[int]): the options that have not been used\n        data_new (List[List[Union[int, str]]]): preprocessed data for original dep and sdp\n        markers_ (List[List[int]]): a list that is uesd to save the maskers for each mask item\n        selected_numbers_ (set): a set that is used to save the selected options\n        number_ (int): the number of mask labels\n        marker_attrs Dict[str, List[int]]: marker attributes\n    Returns:\n        int: 0 mean no mask, the others mean the number of masked ids\n    """\n    np.random.shuffle(left_numbers)\n    for item_ in left_numbers:\n        target_item = data_new[item_]\n        seg_ids = np.array(target_item[:2]) - 1\n        delete_ids = np.where(seg_ids < 1)[0]\n        seg_ids = np.delete(seg_ids, delete_ids)\n        temp_ids = seg2id(seg_ids)\n        ids = []\n        for item in temp_ids:\n            ids += item.copy()\n        if check_ids(ids):\n            length_ = len(ids)\n            if number_ > length_:\n                for id_ in ids:\n                    mask_labels[id_] = 1\n                if target_item[2] in marker_attrs:\n                    detail_info.append([\n                        target_item,\n                        [seg_data[target_item[0] - 1],seg_data[target_item[1] - 1]],\n                    ])\n                    if len(temp_ids) == 1:\n                        markers_.append([temp_ids[0][0], temp_ids[0][-1]])\n                    elif len(temp_ids) == 2:\n                        for i in marker_attrs[target_item[2]]:\n                            markers_.append([temp_ids[i][0], temp_ids[i][-1]])\n                selected_numbers_.add(item_)\n                return length_\n            else:\n                return 0\n    return 0\u5728\u5b8c\u6210\u5bf9\u539f\u59cb\u53e5\u5b50\u7684\u9884\u5904\u7406\u4e4b\u540e\uff0c\u5728\u6a21\u578b\u7684dataloader\u91cc\u9700\u8981\u5bf9\u6570\u636e\u8fdb\u884c\u77e5\u8bc6\u6ce8\u5165\uff0c\u7531\u4e8e\u6a21\u578b\u4e2d\u5f15\u5165\u4e86\u5bf9\u6bd4\u5b66\u4e60\uff0c\u56e0\u6b64\u8be5\u90e8\u5206\u9700\u8981\u5728\u6570\u636e\u8f6c\u6362\u9636\u6bb5\u540c\u65f6\u751f\u6210positive\u548cnegative\u7684\u6837\u672c\u6570\u636e\u3002\u5b9e\u73b0\u8fd9\u4e00\u8fc7\u7a0b\u7684\u6838\u5fc3\u4ee3\u7801\u5982\u4e0b\uff1adef get_positive_and_negative_examples(\nself,\nner_data: str,\nnegative_level: int = 3) -> Union[bool, Dict[str, List[str]]]:\n"""get the positive examples and negative examples for the ner data\nArgs:\n    ner_data (str): the ner entity\n    negative_level (int, optional): the deepth of the relationship. Defaults to 3.\nReturns:\n    Union[bool, Dict[str, List[str]]]: if the `ner_data` not in `konwledge`, return False, otherwise, return the positive and negative examples\n"""\nknowledge: Dict[str, Dict[str, str]] = self.Knowledge_G\ncommon_used = set()\ndef get_data(key: str, \n            data: Dict[str, str], \n            results: List[str], \n            deep: int, \n            insert_flag: bool = False):\n    """get the negative examples recursively\n    Args:\n        key (str): the ner\n        data (Dict[str, str]): the related data about `key`\n        results (List[str]): a list used to save the negative examples\n        deep (int): the recursive number\n        insert_flag (bool, optional): whether insert data to `results`. Defaults to False.\n    """\n    nonlocal knowledge\n    common_used.add(key)\n    if deep == 0:\n        return\n    else:\n        for key_item in data:\n            if data[key_item] not in common_used and insert_flag == True:\n                results.append(data[key_item])\n            if data[key_item] in knowledge and data[key_item] not in common_used:\n                get_data(data[key_item], knowledge[data[key_item]], results, deep - 1, True)\n\nall_examples = {\n    'ner': ner_data,\n    'positive_examples': [],\n    'negative_examples': []\n}\nif ner_data in knowledge:\n    tp_data = knowledge[ner_data]\n    negative_examples = []\n    if '\u63cf\u8ff0' in tp_data:\n        positive_example = tp_data['\u63cf\u8ff0']\n    else:\n        keys = list(tp_data.keys())\n        choice = np.random.choice([_ for _ in range(len(keys))], 1)[0]\n        positive_example = tp_data[keys[choice]]\n    # # the description usually contains the ner entity, if not, concate the `ner_data` and the positive example\n    if ner_data in positive_example:\n        all_examples['positive_examples'].append(positive_example)\n    else:\n        all_examples['positive_examples'].append(ner_data + positive_example)\n    get_data(ner_data, tp_data, negative_examples, negative_level)\n    # concate the ner entity and each negative example\n    negative_examples = list(map(lambda x: ner_data + x if ner_data not in x else x, negative_examples))\n    all_examples['negative_examples'] = negative_examples\n    return all_examples\nreturn False\u5728\u5b8c\u6210\u77e5\u8bc6\u6ce8\u5165\u4e4b\u540e\uff0c\u6a21\u578b\u7684\u6570\u636e\u9884\u5904\u7406\u73af\u8282\u5c31\u5b9e\u73b0\u4e86\u3002\u7d27\u63a5\u7740\uff0c\u7531\u4e8e\u77e5\u8bc6\u6ce8\u5165\u9700\u8981\u989d\u5916\u6dfb\u52a0\u7279\u6b8a\u7684Token\uff0c\u56e0\u6b64\uff0c\u5728\u6a21\u578b\u7684Embedding\u5c42\u9700\u8981\u91cd\u65b0\u8c03\u6574\u5927\u5c0f\uff0c\u8be5\u90e8\u5206\u7684\u5b9e\u73b0\u4ee3\u7801\u5982\u4e0b\uff1amodel.backbone.resize_token_embeddings(len(train_dataset.tokenizer))<\/code><\/pre>\n
model.config.vocab_size = len(train_dataset.tokenizer)\u5728\u5bf9\u6a21\u578b\u7ed3\u6784\u8fdb\u884c\u5fae\u8c03\u4e4b\u540e\uff0c\u6700\u540e\u5c31\u662f\u4fee\u6539\u539f\u59cb\u7684loss\u51fd\u6570\uff0c\u7531\u4e8e\u5f15\u5165\u4e86\u5bf9\u6bd4\u5b66\u4e60\uff0c\u8fd9\u91cc\u9700\u8981\u5728\u539f\u6765loss\u7684\u57fa\u7840\u4e4b\u4e0a\u65b0\u52a0\u4e00\u4e2a\u5bf9\u6bd4\u5b66\u4e60\u7684loss\uff08CKBert\u91c7\u7528SimCLS\u4f5c\u4e3a\u5bf9\u6bd4\u5b66\u4e60\u7684loss\u51fd\u6570\uff09\uff0c\u8be5\u90e8\u5206\u7684\u6838\u5fc3\u4ee3\u7801\u5b9e\u73b0\u5982\u4e0b\uff1adef compute_simcse(self, original_outputs: torch.Tensor,<\/p>\n
               forward_outputs: torch.Tensor) -> float:\n    original_hidden_states = original_outputs['hidden_states'].unsqueeze(-2)\n    loss = nn.CrossEntropyLoss()\n    forward_outputs = torch.mean(forward_outputs, dim=-2)\n    cos_result = self.CosSim(original_hidden_states, forward_outputs)\n    cos_result_size = cos_result.size()\n    cos_result = cos_result.view(-1, cos_result_size[-1])\n    labels = torch.zeros(cos_result.size(0), device=original_outputs['hidden_states'].device).long()\n    loss_ = loss(cos_result, labels)\n    return loss_CKBERT\u52a0\u901f\u9884\u8bad\u7ec3\u7531\u4e8eCKBERT\u7684\u9884\u8bad\u7ec3\u9700\u8981\u8017\u8d39\u5927\u91cf\u65f6\u95f4\u548c\u8ba1\u7b97\u8d44\u6e90\uff0c\u6211\u4eec\u6709\u5fc5\u987b\u5bf9CKBERT\u7684\u9884\u8bad\u7ec3\u8fdb\u884c\u52a0\u901f\u3002\u7531\u4e8eCKBERT\u91c7\u7528PyTorch\u6846\u67b6\u5b9e\u73b0\uff0c\u4e0eTensorflow 1.x Graph Execution\u65b9\u5f0f\u76f8\u6bd4\uff0cPyTorch\u91c7\u7528Eager Execution\u7684\u65b9\u5f0f\u8fd0\u884c\uff0c\u5177\u6709\u5f88\u597d\u7684\u6613\u7528\u6027\u3001\u5bb9\u6613\u5f00\u53d1\u8c03\u8bd5\u7b49\u7279\u70b9\u3002\u4f46\u662f\uff0cPytorch\u7f3a\u5c11\u6a21\u578b\u7684Graph IR\uff08Intermediate Representation\uff09\u8868\u8fbe\uff0c\u56e0\u6b64\u65e0\u6cd5\u8fdb\u884c\u66f4\u6df1\u5ea6\u7684\u4f18\u5316\u3002\u53d7\u5230LazyTensor \u548cPytorch\/XLA\uff08https:\/\/github.com\/pytorch\/xla\uff09\u7684\u542f\u53d1\uff0cPAI\u56e2\u961f\u5728PyTorch\u6846\u67b6\u4e2d\u5f00\u53d1\u4e86TorchAccelerator\uff0c\u65e8\u5728\u89e3\u51b3PyTorch\u4e0a\u7684\u8bad\u7ec3\u4f18\u5316\u95ee\u9898\uff0c\u5728\u4fdd\u8bc1\u7528\u6237\u6613\u7528\u6027\u548c\u53ef\u8c03\u8bd5\u884c\u7684\u57fa\u7840\u4e0a\uff0c\u63d0\u5347\u7528\u6237\u8bad\u7ec3\u901f\u5ea6\u3002\u00a0\u7531\u4e8eLazyTensor\u5728Eager Execution\u5230Graph Execution\u8f6c\u5316\u8fc7\u7a0b\u4e2d\u8fd8\u5b58\u5728\u5f88\u591a\u7f3a\u9677\u3002\u901a\u8fc7\u5c06Custom Operation\u5c01\u88c5\u8fdbXLA CustomCall\u3001\u5bf9Python\u4ee3\u7801\u8fdb\u884cAST\u89e3\u6790\u7b49\u624b\u6bb5\uff0cTorchAccelerator\u63d0\u5347\u4e86Eager Execution\u5230Graph Execution\u7684\u5b8c\u5907\u6027\u548c\u8f6c\u5316\u6027\u80fd\uff0c\u901a\u8fc7\u591aStream\u4f18\u5316\u3001Tensor\u5f02\u6b65\u4f20\u8f93\u7b49\u624b\u6bb5\u63d0\u5347\u7f16\u8bd1\u4f18\u5316\u6548\u679c\u3002<\/code><\/pre>\n
\"\u56fe\u7247\"
\u4ece\u5b9e\u9a8c\u7ed3\u679c\u6765\u770b\uff0c\u5c06TorchAccelerator\u548cAMP\uff08Automatic Mixed Precision\uff0c\u6df7\u5408\u7cbe\u5ea6\u8bad\u7ec3\uff09\u7ed3\u5408\u8d77\u6765\u4f7f\u7528\uff0c\u8bad\u7ec3\u901f\u5ea6\u5c06\u4f1a\u670940%\u4ee5\u4e0a\u7684\u63d0\u5347\uff0c\u8bf4\u660e\u5728AMP\u548cTorchAccelerator\u8fdb\u884c\u76f8\u4e92\u4f5c\u7528\u4e0b\u6709\u6bd4\u8f83\u597d\u7684\u52a0\u901f\u6548\u679c\u3002CKBERT\u5b9e\u9a8c\u6548\u679c\u8bc4\u6d4b\u4e3a\u4e86\u9a8c\u8bc1CKBERT\u6a21\u578b\u5728\u5404\u79cd\u4efb\u52a1\u4e0a\u7684\u7cbe\u5ea6\uff0c\u6211\u4eec\u5728\u591a\u4e2a\u516c\u5f00\u6570\u636e\u96c6\u4e0a\u9a8c\u8bc1\u4e86\u53e5\u5b50\u5206\u7c7b\u548cNER\u4efb\u52a1\u7684\u6548\u679c\uff0c\u5982\u4e0b\u6240\u793a\uff1aCLUE\u6570\u636e\u96c6\u5b9e\u9a8c\u6548\u679c\u6a21\u578bText ClassificationQuestion AnsweringTotal ScoreAFQMCTNEWSIFLYTEKOCNLIWSCCSLCMRCCHIDC3BERT72.7355.2259.5466.5372.4981.7773.4079.1957.9169.72MacBERT69.9057.9360.3567.4374.7182.1373.5579.5158.8970.28PERT73.6154.5057.4266.7076.0782.7773.8080.1958.0370.18ERNIE-Baidu73.0856.2260.1167.4875.7982.1472.8680.0357.6369.83Lattice-BERT72.9656.1458.9767.5476.1081.9973.4780.2457.8070.29K-BERT73.1555.9160.1967.8376.2182.2472.7480.2957.4870.35ERNIE-THU72.8856.5959.3367.9575.8282.3572.9680.2256.3069.98CKBERT-base73.1756.4460.6568.5376.3882.6373.5581.6957.9171.36CKBERT-large74.7555.8660.6270.5778.8982.3073.4582.3458.1272.23CKBERT-huge75.0359.7260.9678.2685.1689.4777.2597.7386.5978.91CKBERT-huge(ensemble)77.0561.1661.1982.8087.1494.2380.4097.9187.2681.02\u00a0\u00a0NER\u6570\u636e\u96c6\u5b9e\u9a8c\u6548\u679c\u6a21\u578bMSRAWeiboOnto.Resu.BERT95.2054.6581.6194.86MacBERT95.0754.9381.9695.22PERT94.9953.7481.4495.10ERNIE-Baidu95.3955.1481.1795.13Lattice-BERT95.2854.9982.0195.31K-BERT94.9755.2181.9894.92ERNIE-THU95.2553.8582.0394.89CKBERT-base95.3555.9782.1995.68CKBERT-large95.5857.0982.4396.08CKBERT-huge96.7958.6683.8797.19\u00a0\u4e0a\u8ff0\u7ed3\u679c\u8bf4\u660e\uff0c\u9996\u5148\u5728CLUE\u6570\u636e\u96c6\u4e0a\uff1a\uff081\uff09\u77e5\u8bc6\u589e\u5f3a\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u6027\u80fd\u76f8\u8f83\u4e8eBERT\u5747\u6709\u8f83\u5927\u63d0\u5347\uff0c\u5728\u4e00\u5b9a\u7a0b\u5ea6\u8bf4\u660e\u4e86\u77e5\u8bc6\u7684\u6ce8\u5165\u80fd\u5e2e\u52a9\u6a21\u578b\u8fdb\u884c\u66f4\u597d\u7684\u8bed\u4e49\u63a8\u7406\uff1b\uff082\uff09\u8ddf\u5148\u524d\u7684\u8f83\u597d\u7684baseline\u6a21\u578b\u76f8\u6bd4\uff0cCKBERT\u7684\u6027\u80fd\u8fdb\u4e00\u6b65\u5f97\u5230\u4e86\u63d0\u5347\uff0c\u8fd9\u4e5f\u8bf4\u660e\u4e86\u5f02\u6784\u77e5\u8bc6\u7684\u6ce8\u5165\u6709\u5229\u4e8e\u6a21\u578b\u6027\u80fd\u7684\u63d0\u5347\uff1b\uff083\uff09\u6a21\u578b\u53c2\u6570\u91cf\u8d8a\u5927\uff0c\u5f02\u6784\u77e5\u8bc6\u7684\u7684\u6ce8\u5165\u6240\u5e26\u6765\u7684\u63d0\u5347\u8d8a\u660e\u663e\uff0c\u8fd9\u5728\u6211\u4eec\u7684huge\u6a21\u578b\u548cbase\u6a21\u578b\u4e4b\u95f4\u7684\u5bf9\u6bd4\u4e0a\u53ef\u4ee5\u770b\u51fa\u3002\u5176\u6b21\uff0c\u5728NER\u6570\u636e\u96c6\u4e0a\uff1a\uff081\uff09\u77e5\u8bc6\u589e\u5f3a\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u6027\u80fd\u76f8\u8f83\u4e8eBERT\u4e5f\u6709\u4e00\u5b9a\u7684\u63d0\u5347\uff1b\uff082\uff09CKBERT\u6a21\u578b\u76f8\u8f83\u4e8e\u5176\u4ed6baseline\u6a21\u578b\u7684\u63d0\u5347\u8f83\u5927\uff0c\u8fd9\u8fdb\u4e00\u6b65\u8bf4\u660e\u4e86\u5f02\u6784\u77e5\u8bc6\u7684\u6ce8\u5165\u5bf9\u4e8e\u6a21\u578b\u6027\u80fd\u7684\u63d0\u5347\u662f\u6709\u5e2e\u52a9\u7684\u3002CKBERT\u6a21\u578b\u4f7f\u2f64\u6559\u7a0b\u4ee5\u4e0b\u6211\u4eec\u7b80\u8981\u4ecb\u7ecd\u5982\u4f55\u5728EasyNLP\u6846\u67b6\u4f7f\u2f64CKBERT\u6a21\u578b\u3002\u5b89\u88c5EasyNLP\u2f64\u6237\u53ef\u4ee5\u76f4\u63a5\u53c2\u8003GitHub\uff08https:\/\/github.com\/alibaba\/Ea…<\/a>\uff09\u4e0a\u7684\u8bf4\u660e\u5b89\u88c5EasyNLP\u7b97\u6cd5\u6846\u67b6\u3002\u6a21\u578b\u9884\u8bad\u7ec3\u4ee5\u4e0b\u4ecb\u7ecdCKBERT\u6a21\u578b\u7684\u9884\u8bad\u7ec3\u8c03\u7528\u8fc7\u7a0b\uff0c\u5982\u679c\u7528\u6237\u6ca1\u6709\u81ea\u884c\u9884\u8bad\u7ec3\u7684\u9700\u6c42\u53ef\u4ee5\u8df3\u8fc7\u6b64\u90e8\u5206\u3002\u6570\u636e\u51c6\u5907CKBERT\u662f\u4e00\u4e2a\u77e5\u8bc6\u5d4c\u5165\u7684\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u9700\u8981\u7528\u6237\u81ea\u5df1\u51c6\u5907\u76f8\u5e94\u7684\u539f\u59cb\u8bad\u7ec3\u6570\u636e\uff08xxx.json\uff09\u548c\u77e5\u8bc6\u56fe\u8c31\uff08xxx.spo\uff09,\u5176\u4e2d\u6570\u636e\u5206\u9694\u5747\u4f7f\u7528\/t\u5206\u9694\u7b26\u3002\u8bad\u7ec3\u6570\u636e\u7684\u683c\u5f0f\u4e3a{‘text’:[‘xxx’], ‘title’:’xxx’}\uff0c\u6837\u4f8b\u5982\u4e0b\uff1a{‘text’: [‘\u6211\u60f3,\u5982\u679c\u6211\u6ca1\u6709\u53bb\u505a\u5927\u5b66\u751f\u6751\u5b98,\u6050\u6015\u6211\u8fd9\u4e2a\u5728\u6606\u660e\u957f\u5927\u7684\u5b69\u5b50\u6c38\u8fdc\u90fd\u4e0d\u80fd\u5207\u8eab\u611f\u53d7\u5230\u4e91\u5357\u8fd9\u6b2160\u5e74\u4e00\u9047\u7684\u7279\u5927\u65f1\u60c5\u7684\u4e25\u91cd\u6027,\u6050\u6015\u6211\u53ea\u662f\u6bcf\u5929\u770b\u7740\u65b0\u95fb\u4e0a\u90a3\u4e9b\u7f3a\u6c34\u7684\u955c\u5934,\u5634\u4e0a\u8bf4\u8981\u8282\u6c34,\u4f46\u4e8b\u5b9e\u884c\u52a8\u4fdd\u6301\u4e0d\u4e86\u4e09\u5929\u3002 \u6211\u4efb\u804c\u7684\u5730\u65b9\u5728\u6606\u660e\u5e02\u7984\u529d\u53bf\u7684\u4e00\u4e2a\u6751\u59d4\u4f1a,\u8bf4\u5b9e\u8bdd\u8fd9\u91cc\u8ddd\u79bb\u7984\u529d\u53bf\u57ce\u4e0d\u8fdc,\u81ea\u7136\u73af\u5883\u4e0d\u7b97\u5f88\u5dee\u3002\u76ee\u524d,\u53ea\u6709\u4e00\u4e2a\u81ea\u7136\u6751\u4fdd\u8bc1\u4e0d\u4e86\u996e\u7528\u6c34\u3002\u4e00\u4e2a\u81ea\u7136\u6751\u57fa\u672c\u80fd\u4fdd\u8bc1\u6709\u996e\u7528\u6c34\u52305\u6708\u3002\u8fd9\u91cc\u6240\u8bf4\u7684\u996e\u7528\u6c34,\u662f\u6307\u4ece\u5c71\u809a\u5b50\u91cc\u51fa\u6765\u7684\u6c34,\u79ef\u5728\u5c0f\u6c34\u575d\u6216\u662f\u6c34\u5858\u91cc\u53c8\u901a\u8fc7\u7ba1\u9053\u8f93\u9001\u5230\u6751\u5b50\u91cc\u7684\u6c34,\u548c\u6211\u4eec\u57ce\u5e02\u91cc\u771f\u6b63\u610f\u4e49\u4e0a\u6d88\u8fc7\u6bd2\u7684\u3001\u80fd\u5b89\u5168\u996e\u7528\u7684\u996e\u7528\u6c34\u4e0d\u540c\u3002\u5728\u6574\u4e2a\u8f93\u9001\u7684\u8fc7\u7a0b\u4e2d,\u53ef\u80fd\u5df2\u7ecf\u4ea7\u751f\u4e86\u6709\u5bb3\u7269\u8d28\u3002\u6211\u89c9\u5f97\u662f\u3002 \u6ca1\u6709\u996e\u7528\u6c34\u7684\u90a3\u4e2a\u81ea\u7136\u6751\u53eb\u5927\u6d77\u5b50\u6751,50\u6237,\u8fd1200\u591a\u4eba\u3002\u5730\u5904\u5c71\u5934,\u4ea4\u901a\u5f88\u4e0d\u4fbf\u5229,\u8d70\u8def\u5927\u6982\u89811\u4e2a\u534a\u5c0f\u65f6\u5230\u4e24\u4e2a\u5c0f\u65f6,\u800c\u4e14\u5761\u5ea6\u6bd4\u8f83\u5927,\u662f\u4e00\u4e2a\u82d7\u65cf\u6751\u5be8\u3002\u5730\u7406\u6761\u4ef6\u9650\u5236,\u57fa\u672c\u6ca1\u6709\u4ec0\u4e48\u7ecf\u6d4e\u4f5c\u7269,\u7b97\u662f\u9760\u5929\u5403\u996d\u7684\u90a3\u79cd\u3002\u4eca\u5e74\u9047\u523060\u5e74\u4e00\u9047\u7684\u5e72\u65f1,\u6751\u91cc\u7684\u4e24\u4e2a\u6c34\u7a96\u90fd\u57fa\u672c\u5e72\u4e86,\u4e4b\u524d\u51e0\u5929,\u4ed6\u4eec\u6751\u957f\u6765\u53cd\u6620,\u51e0\u4e2a\u8001\u4eba\u5df2\u7ecf\u662f\u62ac\u4e2a\u5c0f\u677f\u51f3\u5750\u5230\u7a96\u5e95\u7528\u7897\u8200\u6c34\u4e86\u3002 \u9762\u5bf9\u8fd9\u4e48\u4e25\u5cfb\u7684\u65f1\u60c5,\u6751\u59d4\u4f1a\u7684\u9886\u5bfc\u548c\u5404\u5c0f\u7ec4\u957f\u90fd\u5728\u60f3\u529e\u6cd5\u3002\u4f46\u662f\u4e0a\u5c71\u7684\u8def\u8def\u9762\u60c5\u51b5\u5dee,\u5927\u8f66\u91cd\u8f66\u4e0a\u4e0d\u53bb\uff1b\u5468\u8fb9\u6c34\u6e90\u5730\u5c11\u3002\u6700\u53ef\u884c\u7684\u529e\u6cd5\u5c31\u662f\u4ece\u6b66\u5b9a\u90a3\u8fb9\u7ed5\u8def\u4e0a\u53bb\u3002\u4f46\u6bcf\u5929\u8fd0\u6c34\u4e0a\u53bb\u4e5f\u4e0d\u662f\u529e\u6cd5,\u957f\u8fdc\u6765\u770b\u8fd8\u662f\u8981\u4fee\u5efa\u4e00\u4e2a\u5c0f\u6c34\u575d\u3002\u6751\u59d4\u4f1a\u7684\u9886\u5bfc\u4e3b\u52a8\u6350\u6b3e,\u6751\u6c11\u4e5f\u81ea\u884c\u7b79\u8d44,\u5f00\u59cb\u81ea\u6551\u3002 \u6700\u8fd1\u6bcf\u4e2a\u5468\u672b\u90fd\u56de\u5bb6,\u6dfb\u7f6e\u9632\u6652\u54c1,\u56e0\u4e3a\u57fa\u672c\u6bcf\u5929\u90fd\u4e0a\u5c71\u4e0b\u6751\u53bb\u4e86\u89e3\u60c5\u51b5,\u5fc5\u987b\u638c\u63e1\u8f96\u533a\u518513\u4e2a\u6751\u5c0f\u7ec4\u6c34\u8d44\u6e90\u7684\u60c5\u51b5\u3002\u6211\u6bcf\u6b21\u56de\u5bb6\u89c1\u5230\u670b\u53cb\u4eec,\u7b2c\u4e00\u53e5\u5c31\u662f,\u8981\u8282\u7ea6\u7528\u6c34\u554a \u670b\u53cb\u4eec,\u4f60\u4eec\u73b0\u5728\u770b\u5230\u7684\u53ea\u662f\u7b80\u5355\u7406\u89e3\u7684”\u7f3a\u6c34”\u3002\u4f60\u4eec\u6240\u4e0d\u77e5\u9053\u7684\u662f,\u6ca1\u6c34\u5c0f\u6625\u4f5c\u7269\u9762\u4e34\u7edd\u6536\u30014\u6708\u4efd\u63d2\u79e7\u6ca1\u6709\u6c34\u6ce1\u79e7\u7530\u30015\u6708\u4efd\u79cd\u70e4\u70df\u4e5f\u6ca1\u6709\u6c34\u3002\u3002\u3002\u90a3\u4e48\u5bf9\u519c\u6c11\u5c31\u610f\u5473\u7740\u4eca\u5e74\u4e00\u5e74\u5c31\u6ca1\u6709\u4e86\u6536\u5165\u3002\u6211\u4eec\u73b0\u5728\u80fd\u52aa\u529b\u505a\u597d\u7684,\u53ea\u662f\u4fdd\u8bc1\u4eba\u7684\u996e\u7528\u6c34\u3002 \u4e0a\u5468\u5c31\u5728\u60f3\u80fd\u4e0d\u80fd\u901a\u8fc7\u4ec0\u4e48\u6e20\u9053\u5e2e\u6751\u6c11\u4eec\u505a\u70b9\u4e8b\u3002\u53d4\u53d4\u53eb\u6211\u5f04\u4e2a\u6297\u65f1\u7684\u57fa\u91d1,\u4ed6\u53d1\u52a8\u5468\u56f4\u7684\u670b\u53cb\u6765\u6350\u8d60,\u5e0c\u671b\u80fd\u53e3\u53e3\u76f8\u4f20\u5e26\u52a8\u591a\u4e00\u70b9\u670b\u53cb\u3002\u6211\u6b63\u5728\u7b79\u5907\u9636\u6bb5,\u770b\u600e\u6837\u624d\u80fd\u505a\u5230\u6700\u5927\u7684\u516c\u5f00\u900f\u660e,\u8ba9\u6350\u8d60\u7684\u4eba\u5b8c\u5168\u653e\u5fc3\u3002 \u5468\u4e00\u63a5\u5230\u4e00\u4e2a\u670b\u53cb\u7684\u7535\u8bdd,\u8bf4\u4ed6\u4eec\u516c\u53f8\u60f3\u4e3a\u65f1\u707e\u732e\u70b9\u7231\u5fc3,\u60f3\u4e70\u70b9\u6c34\u9001\u53bb\u6211\u4eec\u90a3\u513f\u3002\u6628\u5929\u89c1\u4e86\u8d1f\u8d23\u4eba,\u5f88\u8c26\u548c\u7684\u4e00\u4e2a\u59d0\u59d0,\u5979\u8bf4\u5927\u5bb6\u90fd\u60f3\u505a\u70b9\u4ec0\u4e48,\u89c9\u5f97\u6350\u94b1\u6ca1\u4ec0\u4e48\u610f\u4e49,\u60f3\u4eb2\u81ea\u9001\u6c34\u5230\u6700\u9700\u8981\u7684\u5730\u65b9\u3002 \u5176\u5b9e\u4eba\u5bb6\u53ea\u662f\u5bb6\u79c1\u8425\u7684\u5c0f\u516c\u53f8,\u4f46\u6211\u771f\u7684\u5f88\u611f\u8c22\u4ed6\u4eec\u3002\u59d0\u59d0\u8fd8\u7279\u522b\u4ea4\u4ee3\u6211,\u6211\u4eec\u53ea\u9700\u8981\u627e\u62d6\u62c9\u673a\u4e0b\u6765\u5e2e\u5fd9\u628a\u6c34\u8fd0\u4e0a\u5c71\u53bb,\u5176\u4ed6\u7684\u4ec0\u4e48\u90fd\u4e0d\u7528\u7ba1,\u4ed6\u4eec\u4f1a\u5b89\u6392\u597d\u7684\u3002\u8fd9\u4e2a\u5468\u672b,\u5c06\u6709 400\u4ef6\u77ff\u6cc9\u6c34\u4f1a\u9001\u5230\u6751\u6c11\u5bb6\u91cc\u3002\u6211\u60f3,\u5e94\u8be5\u53ef\u4ee5\u6682\u65f6\u7f13\u89e3\u65f1\u60c5\u3002\u518d\u6b21\u4ee3\u6751\u6c11\u611f\u8c22\u4ed6\u4eec\uff01 \u4e0b\u534a\u5e74,\u65f1\u60c5\u7ed9\u519c\u6c11\u7684\u751f\u4ea7\u3001\u751f\u6d3b\u5e26\u6765\u7684\u95ee\u9898\u8fd8\u5f88\u591a\u3002\u4f46\u662f\u6211\u4e2a\u4eba\u7684\u529b\u91cf\u5f88\u6709\u9650,\u5e0c\u671b\u80fd\u591f\u770b\u5230\u8fd9\u4e9b\u5e16\u5b50\u7684\u670b\u53cb\u4eec,\u5982\u679c\u6709\u80fd\u529b,\u6709\u8fd9\u4efd\u5fc3\u610f\u7684\u8bdd,\u8bf7\u7ed9\u4e88\u65f1\u707e\u5730\u533a\u7684\u519c\u6c11\u66f4\u591a\u7684\u5e2e\u52a9\u3002 \u6211\u60f3\u5927\u5bb6\u90fd\u77e5\u9053,\u6606\u660e80%\u4ee5\u4e0a\u7684\u7528\u6c34\u90fd\u6765\u81ea\u7984\u529d\u7684\u4e91\u9f99\u6c34\u5e93,\u4e91\u9f99\u7684\u540c\u4e8b\u4eec\u4e5f\u5728”\u6297\u65f1”,\u4ed6\u4eec\u7684\u5de5\u4f5c\u4efb\u52a1\u662f\u8981\u4fdd\u8bc1\u4e25\u683c\u7684\u8282\u7ea6\u7528\u6c34,\u8981\u5bfb\u6c42\u5176\u4ed6\u6c34\u6e90\u7528\u6c34,\u4ece\u800c\u4fdd\u8bc1\u6606\u660e\u7684\u7528\u6c34\u3002\u6240\u4ee5,\u8bf7\u6bcf\u4e00\u4e2a\u6606\u660e\u4eba\u90fd\u8282\u6c34\u5427,\u7984\u529d\u7684\u5f88\u591a\u5730\u65b9\u90fd\u5728\u7f3a\u6c34,\u6211\u4eec\u90a3\u91cc\u4e0d\u7b97\u4e25\u91cd\u7684,\u8bf7\u73cd\u60dc\u4f60\u4eec\u73b0\u5728\u5728\u7528\u7684\u6bcf\u4e00\u6ef4\u6c34~ \u4e5f\u8bb8,\u8981\u7ecf\u5386\u8fc7\u8fd9\u6837\u4e00\u6b21\u89e6\u76ee\u60ca\u5fc3\u7684\u5927\u65f1\u624d\u771f\u6b63\u77e5\u9053\u6c34\u7684\u73cd\u8d35\u3002\u5e0c\u671b\u6211\u4eec\u90fd\u884c\u52a8\u8d77\u6765,\u4e0d\u8981\u518d\u8ba9\u8fd9\u6837\u7684\u65f1\u707e\u4fb5\u88ad\u6211\u4eec\u7684\u5bb6\u4e61\u3002’], ‘title’: ‘\u65f1\u60c5\u8bb0\u8981—–\u6606\u660e\u4eba,\u8bf7\u73cd\u60dc\u4f60\u4eec\u73b0\u5728\u5728\u7528\u7684\u6bcf\u4e00\u6ef4\u6c34~’}\u77e5\u8bc6\u56fe\u8c31\u6570\u636e\u683c\u5f0f\u4e3a\u4e09\u5217\u6570\u636e\uff0c\u4ece\u5de6\u5230\u53f3\u5206\u522b\u662f\u5b9e\u4f53\u5173\u7cfb\u7684\u63cf\u8ff0\uff0c\u6837\u4f8b\u5982\u4e0b\uff1a\u7ea2\u8272\u98df\u54c1  \u6807\u7b7e  \u751f\u6d3b\u6570\u636e\u9884\u5904\u7406\u53ef\u4ee5\u4f7f\u7528\u63d0\u4f9b\u7684\u6570\u636e\u9884\u5904\u7406\u811a\u672c(preprocess\/run_local_preprocess.sh)\u6765\u5bf9\u539f\u59cb\u6570\u636e\u8fdb\u884c\u4e00\u952e\u5904\u7406\uff0c\u5728\u7ecf\u8fc7LTP\u5904\u7406\u4e4b\u540e\uff0c\u6570\u636e\u6837\u4f8b\u5982\u4e0b\uff1a{“text”: [“\u6211\u60f3,\u5982\u679c\u6211\u6ca1\u6709\u53bb\u505a\u5927\u5b66\u751f\u6751\u5b98,\u6050\u6015\u6211\u8fd9\u4e2a\u5728\u6606\u660e\u957f\u5927\u7684\u5b69\u5b50\u6c38\u8fdc\u90fd\u4e0d\u80fd\u5207\u8eab\u611f\u53d7\u5230\u4e91\u5357\u8fd9\u6b2160\u5e74\u4e00\u9047\u7684\u7279\u5927\u65f1\u60c5\u7684\u4e25\u91cd\u6027,\u6050\u6015\u6211\u53ea\u662f\u6bcf\u5929\u770b\u7740\u65b0\u95fb\u4e0a\u90a3\u4e9b\u7f3a\u6c34\u7684\u955c\u5934,\u5634\u4e0a\u8bf4\u8981\u8282\u6c34,\u4f46\u4e8b\u5b9e\u884c\u52a8\u4fdd\u6301\u4e0d\u4e86\u4e09\u5929\u3002\u6211\u4efb\u804c\u7684\u5730\u65b9\u5728\u6606\u660e\u5e02\u7984\u529d\u53bf\u7684\u4e00\u4e2a\u6751\u59d4\u4f1a,\u8bf4\u5b9e\u8bdd\u8fd9\u91cc\u8ddd\u79bb\u7984\u529d\u53bf\u57ce\u4e0d\u8fdc,\u81ea\u7136\u73af\u5883\u4e0d\u7b97\u5f88\u5dee\u3002\u76ee\u524d,\u53ea\u6709\u4e00\u4e2a\u81ea\u7136\u6751\u4fdd\u8bc1\u4e0d\u4e86\u996e\u7528\u6c34\u3002\u4e00\u4e2a\u81ea\u7136\u6751\u57fa\u672c\u80fd\u4fdd\u8bc1\u6709\u996e\u7528\u6c34\u52305\u6708\u3002\u8fd9\u91cc\u6240\u8bf4\u7684\u996e\u7528\u6c34,\u662f\u6307\u4ece\u5c71\u809a\u5b50\u91cc\u51fa\u6765\u7684\u6c34,\u79ef\u5728\u5c0f\u6c34\u575d\u6216\u662f\u6c34\u5858\u91cc\u53c8\u901a\u8fc7\u7ba1\u9053\u8f93\u9001\u5230\u6751\u5b50\u91cc\u7684\u6c34,\u548c\u6211\u4eec\u57ce\u5e02\u91cc\u771f\u6b63\u610f\u4e49\u4e0a\u6d88\u8fc7\u6bd2\u7684\u3001\u80fd\u5b89\u5168\u996e\u7528\u7684\u996e\u7528\u6c34\u4e0d\u540c\u3002\u5728\u6574\u4e2a\u8f93\u9001\u7684\u8fc7\u7a0b\u4e2d,\u53ef\u80fd\u5df2\u7ecf\u4ea7\u751f\u4e86\u6709\u5bb3\u7269\u8d28\u3002\u6211\u89c9\u5f97\u662f\u3002\u6ca1\u6709\u996e\u7528\u6c34\u7684\u90a3\u4e2a\u81ea\u7136\u6751\u53eb\u5927\u6d77\u5b50\u6751,50\u6237,\u8fd1200\u591a\u4eba\u3002\u5730\u5904\u5c71\u5934,\u4ea4\u901a\u5f88\u4e0d\u4fbf\u5229,\u8d70\u8def\u5927\u6982\u89811\u4e2a\u534a\u5c0f\u65f6\u5230\u4e24\u4e2a\u5c0f\u65f6,\u800c\u4e14\u5761\u5ea6\u6bd4\u8f83\u5927,\u662f\u4e00\u4e2a\u82d7\u65cf\u6751\u5be8\u3002\u5730\u7406\u6761\u4ef6\u9650\u5236,\u57fa\u672c\u6ca1\u6709\u4ec0\u4e48\u7ecf\u6d4e\u4f5c\u7269,\u7b97\u662f\u9760\u5929\u5403\u996d\u7684\u90a3\u79cd\u3002\u4eca\u5e74\u9047\u523060\u5e74\u4e00\u9047\u7684\u5e72\u65f1,\u6751\u91cc\u7684\u4e24\u4e2a\u6c34\u7a96\u90fd\u57fa\u672c\u5e72\u4e86,\u4e4b\u524d\u51e0\u5929,\u4ed6\u4eec\u6751\u957f\u6765\u53cd\u6620,\u51e0\u4e2a\u8001\u4eba\u5df2\u7ecf\u662f\u62ac\u4e2a\u5c0f\u677f\u51f3\u5750\u5230\u7a96\u5e95\u7528\u7897\u8200\u6c34\u4e86\u3002\u9762\u5bf9\u8fd9\u4e48\u4e25\u5cfb\u7684\u65f1\u60c5,\u6751\u59d4\u4f1a\u7684\u9886\u5bfc\u548c\u5404\u5c0f\u7ec4\u957f\u90fd\u5728\u60f3\u529e\u6cd5\u3002\u4f46\u662f\u4e0a\u5c71\u7684\u8def\u8def\u9762\u60c5\u51b5\u5dee,\u5927\u8f66\u91cd\u8f66\u4e0a\u4e0d\u53bb\uff1b\u5468\u8fb9\u6c34\u6e90\u5730\u5c11\u3002\u6700\u53ef\u884c\u7684\u529e\u6cd5\u5c31\u662f\u4ece\u6b66\u5b9a\u90a3\u8fb9\u7ed5\u8def\u4e0a\u53bb\u3002\u4f46\u6bcf\u5929\u8fd0\u6c34\u4e0a\u53bb\u4e5f\u4e0d\u662f\u529e\u6cd5,\u957f\u8fdc\u6765\u770b\u8fd8\u662f\u8981\u4fee\u5efa\u4e00\u4e2a\u5c0f\u6c34\u575d\u3002\u6751\u59d4\u4f1a\u7684\u9886\u5bfc\u4e3b\u52a8\u6350\u6b3e,\u6751\u6c11\u4e5f\u81ea\u884c\u7b79\u8d44,\u5f00\u59cb\u81ea\u6551\u3002\u6700\u8fd1\u6bcf\u4e2a\u5468\u672b\u90fd\u56de\u5bb6,\u6dfb\u7f6e\u9632\u6652\u54c1,\u56e0\u4e3a\u57fa\u672c\u6bcf\u5929\u90fd\u4e0a\u5c71\u4e0b\u6751\u53bb\u4e86\u89e3\u60c5\u51b5,\u5fc5\u987b\u638c\u63e1\u8f96\u533a\u518513\u4e2a\u6751\u5c0f\u7ec4\u6c34\u8d44\u6e90\u7684\u60c5\u51b5\u3002\u6211\u6bcf\u6b21\u56de\u5bb6\u89c1\u5230\u670b\u53cb\u4eec,\u7b2c\u4e00\u53e5\u5c31\u662f,\u8981\u8282\u7ea6\u7528\u6c34\u554a<\/del>\u670b\u53cb\u4eec,\u4f60\u4eec\u73b0\u5728\u770b\u5230\u7684\u53ea\u662f\u7b80\u5355\u7406\u89e3\u7684\/"\u7f3a\u6c34\/"\u3002\u4f60\u4eec\u6240\u4e0d\u77e5\u9053\u7684\u662f,\u6ca1\u6c34\u5c0f\u6625\u4f5c\u7269\u9762\u4e34\u7edd\u6536\u30014\u6708\u4efd\u63d2\u79e7\u6ca1\u6709\u6c34\u6ce1\u79e7\u7530\u30015\u6708\u4efd\u79cd\u70e4\u70df\u4e5f\u6ca1\u6709\u6c34\u3002\u3002\u3002\u90a3\u4e48\u5bf9\u519c\u6c11\u5c31\u610f\u5473\u7740\u4eca\u5e74\u4e00\u5e74\u5c31\u6ca1\u6709\u4e86\u6536\u5165\u3002\u6211\u4eec\u73b0\u5728\u80fd\u52aa\u529b\u505a\u597d\u7684,\u53ea\u662f\u4fdd\u8bc1\u4eba\u7684\u996e\u7528\u6c34\u3002\u4e0a\u5468\u5c31\u5728\u60f3\u80fd\u4e0d\u80fd\u901a\u8fc7\u4ec0\u4e48\u6e20\u9053\u5e2e\u6751\u6c11\u4eec\u505a\u70b9\u4e8b\u3002\u53d4\u53d4\u53eb\u6211\u5f04\u4e2a\u6297\u65f1\u7684\u57fa\u91d1,\u4ed6\u53d1\u52a8\u5468\u56f4\u7684\u670b\u53cb\u6765\u6350\u8d60,\u5e0c\u671b\u80fd\u53e3\u53e3\u76f8\u4f20\u5e26\u52a8\u591a\u4e00\u70b9\u670b\u53cb\u3002\u6211\u6b63\u5728\u7b79\u5907\u9636\u6bb5,\u770b\u600e\u6837\u624d\u80fd\u505a\u5230\u6700\u5927\u7684\u516c\u5f00\u900f\u660e,\u8ba9\u6350\u8d60\u7684\u4eba\u5b8c\u5168\u653e\u5fc3\u3002\u5468\u4e00\u63a5\u5230\u4e00\u4e2a\u670b\u53cb\u7684\u7535\u8bdd,\u8bf4\u4ed6\u4eec\u516c\u53f8\u60f3\u4e3a\u65f1\u707e\u732e\u70b9\u7231\u5fc3,\u60f3\u4e70\u70b9\u6c34\u9001\u53bb\u6211\u4eec\u90a3\u513f\u3002\u6628\u5929\u89c1\u4e86\u8d1f\u8d23\u4eba,\u5f88\u8c26\u548c\u7684\u4e00\u4e2a\u59d0\u59d0,\u5979\u8bf4\u5927\u5bb6\u90fd\u60f3\u505a\u70b9\u4ec0\u4e48,\u89c9\u5f97\u6350\u94b1\u6ca1\u4ec0\u4e48\u610f\u4e49,\u60f3\u4eb2\u81ea\u9001\u6c34\u5230\u6700\u9700\u8981\u7684\u5730\u65b9\u3002\u5176\u5b9e\u4eba\u5bb6\u53ea\u662f\u5bb6\u79c1\u8425\u7684\u5c0f\u516c\u53f8,\u4f46\u6211\u771f\u7684\u5f88\u611f\u8c22\u4ed6\u4eec\u3002\u59d0\u59d0\u8fd8\u7279\u522b\u4ea4\u4ee3\u6211,\u6211\u4eec\u53ea\u9700\u8981\u627e\u62d6\u62c9\u673a\u4e0b\u6765\u5e2e\u5fd9\u628a\u6c34\u8fd0\u4e0a\u5c71\u53bb,\u5176\u4ed6\u7684\u4ec0\u4e48\u90fd\u4e0d\u7528\u7ba1,\u4ed6\u4eec\u4f1a\u5b89\u6392\u597d\u7684\u3002\u8fd9\u4e2a\u5468\u672b,\u5c06\u6709400\u4ef6\u77ff\u6cc9\u6c34\u4f1a\u9001\u5230\u6751\u6c11\u5bb6\u91cc\u3002\u6211\u60f3,\u5e94\u8be5\u53ef\u4ee5\u6682\u65f6\u7f13\u89e3\u65f1\u60c5\u3002\u518d\u6b21\u4ee3\u6751\u6c11\u611f\u8c22\u4ed6\u4eec\uff01\u4e0b\u534a\u5e74,\u65f1\u60c5\u7ed9\u519c\u6c11\u7684\u751f\u4ea7\u3001\u751f\u6d3b\u5e26\u6765\u7684\u95ee\u9898\u8fd8\u5f88\u591a\u3002\u4f46\u662f\u6211\u4e2a\u4eba\u7684\u529b\u91cf\u5f88\u6709\u9650,\u5e0c\u671b\u80fd\u591f\u770b\u5230\u8fd9\u4e9b\u5e16\u5b50\u7684\u670b\u53cb\u4eec,\u5982\u679c\u6709\u80fd\u529b,\u6709\u8fd9\u4efd\u5fc3\u610f\u7684\u8bdd,\u8bf7\u7ed9\u4e88\u65f1\u707e\u5730\u533a\u7684\u519c\u6c11\u66f4\u591a\u7684\u5e2e\u52a9\u3002\u6211\u60f3\u5927\u5bb6\u90fd\u77e5\u9053,\u6606\u660e80%\u4ee5\u4e0a\u7684\u7528\u6c34\u90fd\u6765\u81ea\u7984\u529d\u7684\u4e91\u9f99\u6c34\u5e93,\u4e91\u9f99\u7684\u540c\u4e8b\u4eec\u4e5f\u5728\/"\u6297\u65f1\/",\u4ed6\u4eec\u7684\u5de5\u4f5c\u4efb\u52a1\u662f\u8981\u4fdd\u8bc1\u4e25\u683c\u7684\u8282\u7ea6\u7528\u6c34,\u8981\u5bfb\u6c42\u5176\u4ed6\u6c34\u6e90\u7528\u6c34,\u4ece\u800c\u4fdd\u8bc1\u6606\u660e\u7684\u7528\u6c34\u3002\u6240\u4ee5,\u8bf7\u6bcf\u4e00\u4e2a\u6606\u660e\u4eba\u90fd\u8282\u6c34\u5427,\u7984\u529d\u7684\u5f88\u591a\u5730\u65b9\u90fd\u5728\u7f3a\u6c34,\u6211\u4eec\u90a3\u91cc\u4e0d\u7b97\u4e25\u91cd\u7684,\u8bf7\u73cd\u60dc\u4f60\u4eec\u73b0\u5728\u5728\u7528\u7684\u6bcf\u4e00\u6ef4\u6c34~\u4e5f\u8bb8,\u8981\u7ecf\u5386\u8fc7\u8fd9\u6837\u4e00\u6b21\u89e6\u76ee\u60ca\u5fc3\u7684\u5927\u65f1\u624d\u771f\u6b63\u77e5\u9053\u6c34\u7684\u73cd\u8d35\u3002\u5e0c\u671b\u6211\u4eec\u90fd\u884c\u52a8\u8d77\u6765,\u4e0d\u8981\u518d\u8ba9\u8fd9\u6837\u7684\u65f1\u707e\u4fb5\u88ad\u6211\u4eec\u7684\u5bb6\u4e61\u3002”], “title”: “\u65f1\u60c5\u8bb0\u8981—–\u6606\u660e\u4eba,\u8bf7\u73cd\u60dc\u4f60\u4eec\u73b0\u5728\u5728\u7528\u7684\u6bcf\u4e00\u6ef4\u6c34~”, “seg”: [“\u6211”, “\u60f3”, “,”, “\u5982\u679c”, “\u6211”, “\u6ca1\u6709”, “\u53bb”, “\u505a”, “\u5927\u5b66\u751f\u6751\u5b98”, “,”, “\u6050\u6015”, “\u6211”, “\u8fd9\u4e2a”, “\u5728”, “\u6606\u660e\u957f\u5927”, “\u7684”, “\u5b69\u5b50”, “\u6c38\u8fdc”, “\u90fd”, “\u4e0d\u80fd\u5207\u8eab\u611f\u53d7\u5230”, “\u4e91\u5357”, “\u8fd9\u6b21”, “60\u5e74”, “\u4e00\u9047\u7684”, “\u7279\u5927\u65f1\u60c5”, “\u7684”, “\u4e25\u91cd\u6027”, “,”, “\u6050\u6015”, “\u6211”, “\u53ea\u662f”, “\u6bcf\u5929”, “\u770b”, “\u7740”, “\u65b0\u95fb\u4e0a”, “\u90a3\u4e9b”, “\u7f3a\u6c34”, “\u7684”, “\u955c\u5934”, “,”, “\u5634\u4e0a\u8bf4\u8981\u8282\u6c34”, “,”, “\u4f46\u4e8b\u5b9e\u884c\u52a8”, “\u4fdd\u6301\u4e0d\u4e86”, “\u4e09\u5929”, “\u3002”, “\u6211”, “\u4efb\u804c”, “\u7684”, “\u5730\u65b9”, “\u5728”, “\u6606\u660e\u5e02\u7984\u529d\u53bf”, “\u7684”, “\u4e00\u4e2a”, “\u6751\u59d4\u4f1a”, “,”, “\u8bf4\u5b9e\u8bdd”, “\u8fd9\u91cc”, “\u8ddd\u79bb\u7984”, “\u529d\u53bf\u57ce\u4e0d\u8fdc”, “,”, “\u81ea\u7136\u73af\u5883”, “\u4e0d\u7b97”, “\u5f88\u5dee”, “\u3002”, “\u76ee\u524d”, “,”, “\u53ea\u6709”, “\u4e00\u4e2a”, “\u81ea\u7136\u6751”, “\u4fdd\u8bc1\u4e0d\u4e86”, “\u996e\u7528\u6c34”, “\u3002”, “\u4e00\u4e2a”, “\u81ea\u7136\u6751”, “\u57fa\u672c”, “\u80fd”, “\u4fdd\u8bc1”, “\u6709”, “\u996e\u7528\u6c34”, “\u5230”, “5\u6708”, “\u3002”, “\u8fd9\u91cc\u6240\u8bf4”, “\u7684”, “\u996e\u7528\u6c34”, “,”, “\u662f\u6307”, “\u4ece”, “\u5c71\u809a\u5b50\u91cc”, “\u51fa\u6765”, “\u7684”, “\u6c34”, “,\u79ef\u5728”, “\u5c0f\u6c34\u575d”, “\u6216\u662f”, “\u6c34\u5858\u91cc”, “\u53c8”, “\u901a\u8fc7”, “\u7ba1\u9053”, “\u8f93\u9001\u5230”, “\u6751\u5b50\u91cc”, “\u7684\u6c34”, “,”, “\u548c”, “\u6211\u4eec”, “\u57ce\u5e02\u91cc”, “\u771f\u6b63\u610f\u4e49\u4e0a”, “\u6d88\u8fc7\u6bd2\u7684”, “\u3001\u80fd\u5b89\u5168”, “\u996e\u7528”, “\u7684”, “\u996e\u7528\u6c34\u4e0d\u540c”, “\u3002”, “\u5728”, “\u6574\u4e2a”, “\u8f93\u9001”, “\u7684”, “\u8fc7\u7a0b”, “\u4e2d”, “,”, “\u53ef\u80fd”, “\u5df2\u7ecf”, “\u4ea7\u751f”, “\u4e86”, “\u6709\u5bb3\u7269\u8d28”, “\u3002”, “\u6211”, “\u89c9\u5f97”, “\u662f”, “\u3002”, “\u6ca1\u6709”, “\u996e\u7528\u6c34”, “\u7684”, “\u90a3”, “\u4e2a”, “\u81ea\u7136\u6751\u53eb\u5927\u6d77\u5b50\u6751”, “,”, “50\u6237”, “,\u8fd1”, “200\u591a\u4eba”, “\u3002\u5730\u5904\u5c71\u5934”, “,”, “\u4ea4\u901a”, “\u5f88\u4e0d\u4fbf\u5229”, “,”, “\u8d70\u8def”, “\u5927\u6982\u8981”, “1\u4e2a\u534a\u5c0f\u65f6\u5230”, “\u4e24\u4e2a”, “\u5c0f\u65f6”, “,”, “\u800c\u4e14\u5761\u5ea6”, “\u6bd4\u8f83\u5927”, “,”, “\u662f”, “\u4e00\u4e2a”, “\u82d7\u65cf\u6751\u5be8”, “\u3002\u5730\u7406\u6761\u4ef6”, “\u9650\u5236”, “,”, “\u57fa\u672c”, “\u6ca1\u6709\u4ec0\u4e48\u7ecf\u6d4e\u4f5c\u7269”, “,”, “\u7b97\u662f”, “\u9760\u5929”, “\u5403\u996d”, “\u7684”, “\u90a3\u79cd”, “\u3002”, “\u4eca\u5e74”, “\u9047\u5230”, “60\u5e74\u4e00\u9047”, “\u7684”, “\u5e72\u65f1”, “,\u6751\u91cc”, “\u7684”, “\u4e24\u4e2a\u6c34\u7a96”, “\u90fd”, “\u57fa\u672c”, “\u5e72”, “\u4e86”, “,\u4e4b\u524d\u51e0\u5929”, “,”, “\u4ed6\u4eec”, “\u6751\u957f\u6765\u53cd\u6620”, “,”, “\u51e0\u4e2a\u8001\u4eba”, “\u5df2\u7ecf”, “\u662f”, “\u62ac\u4e2a\u5c0f\u677f\u51f3\u5750\u5230\u7a96\u5e95\u7528\u7897\u8200\u6c34”, “\u4e86”, “\u3002\u9762\u5bf9”, “\u8fd9\u4e48”, “\u4e25\u5cfb\u7684”, “\u65f1\u60c5”, “,”, “\u6751\u59d4\u4f1a”, “\u7684\u9886\u5bfc”, “\u548c”, “\u5404\u5c0f\u7ec4\u957f”, “\u90fd”, “\u5728”, “\u60f3”, “\u529e\u6cd5”, “\u3002”, “\u4f46\u662f\u4e0a\u5c71\u7684”, “\u8def\u8def\u9762\u60c5\u51b5”, “\u5dee”, “,”, “\u5927\u8f66\u91cd\u8f66\u4e0a\u4e0d\u53bb”, “\uff1b”, “\u5468\u8fb9\u6c34\u6e90\u5730\u5c11”, “\u3002”, “\u6700”, “\u53ef\u884c”, “\u7684”, “\u529e\u6cd5”, “\u5c31\u662f”, “\u4ece\u6b66\u5b9a”, “\u90a3\u8fb9”, “\u7ed5\u8def\u4e0a\u53bb”, “\u3002”, “\u4f46”, “\u6bcf\u5929\u8fd0\u6c34\u4e0a”], “ner”: [[“Ns”, 14, 14], [“Ns”, 20, 20], [“Ns”, 51, 51]], “dep”: [[1, 2, “SBV”], [2, 0, “HED”], [3, 2, “WP”], [4, 6, “ADV”], [5, 6, “SBV”], [6, 2, “VOB”], [7, 6, “COO”], [8, 7, “COO”], [9, 8, “VOB”], [10, 8, “WP”], [11, 7, “COO”], [12, 172, “SBV”], [13, 172, “ADV”], [14, 172, “ADV”], [15, 14, “POB”], [16, 172, “RAD”], [17, 172, “SBV”], [18, 172, “ADV”], [19, 172, “ADV”], [20, 172, “ADV”], [21, 172, “SBV”], [22, 172, “ADV”], [23, 172, “ADV”], [24, 172, “ADV”], [25, 172, “ADV”], [26, 172, “RAD”], [27, 172, “VOB”], [28, 156, “WP”], [29, 33, “ADV”], [30, 33, “SBV”], [31, 33, “ADV”], [32, 33, “ADV”], [33, 156, “COO”], [34, 33, “RAD”], [35, 39, “ATT”], [36, 39, “ATT”], [37, 39, “ATT”], [38, 37, “RAD”], [39, 33, “VOB”], [40, 33, “WP”], [41, 44, “ADV”], [42, 44, “WP”], [43, 44, “ADV”], [44, 33, “COO”], [45, 44, “CMP”], [46, 44, “WP”], [47, 48, “SBV”], [48, 55, “ATT”], [49, 48, “RAD”], [50, 51, “POB”], [51, 48, “ADV”], [52, 51, “POB”], [53, 48, “RAD”], [54, 55, “ATT”], [55, 44, “SBV”], [56, 44, “WP”], [57, 44, “ADV”], [58, 59, “SBV”], [59, 44, “ADV”], [60, 44, “COO”], [61, 44, “WP”], [62, 63, “SBV”], [63, 71, “CMP”], [64, 71, “CMP”], [65, 71, “WP”], [66, 71, “ADV”], [67, 71, “WP”], [68, 71, “ADV”], [69, 70, “ATT”], [70, 71, “SBV”], [71, 44, “COO”], [72, 71, “COO”], [73, 71, “WP”], [74, 75, “ATT”], [75, 71, “SBV”], [76, 78, “ADV”], [77, 78, “ADV”], [78, 44, “COO”], [79, 78, “VOB”], [80, 78, “VOB”], [81, 80, “CMP”], [82, 81, “POB”], [83, 78, “WP”], [84, 86, “ATT”], [85, 84, “RAD”], [86, 78, “COO”], [87, 78, “WP”], [88, 44, “COO”], [89, 91, “ADV”], [90, 89, “POB”], [91, 93, “ATT”], [92, 91, “RAD”], [93, 101, “VOB”], [94, 101, “WP”], [95, 101, “CMP”], [96, 97, “LAD”], [97, 101, “VOB”], [98, 101, “ADV”], [99, 101, “ADV”], [100, 99, “POB”], [101, 156, “COO”], [102, 101, “SBV”], [103, 101, “RAD”], [104, 101, “WP”], [105, 109, “ADV”], [106, 107, “ATT”], [107, 105, “POB”], [108, 109, “ADV”], [109, 156, “COO”], [110, 111, “WP”], [111, 109, “COO”], [112, 109, “RAD”], [113, 109, “COO”], [114, 129, “WP”], [115, 129, “ADV”], [116, 119, “ATT”], [117, 119, “ATT”], [118, 117, “RAD”], [119, 120, “ATT”], [120, 115, “POB”], [121, 129, “WP”], [122, 124, “ADV”], [123, 124, “ADV”], [124, 129, “COO”], [125, 124, “RAD”], [126, 129, “COO”], [127, 129, “WP”], [128, 129, “SBV”], [129, 109, “COO”], [130, 129, “VOB”], [131, 129, “WP”], [132, 133, “COO”], [133, 109, “COO”], [134, 109, “RAD”], [135, 109, “ADV”], [136, 137, “ATT”], [137, 109, “SBV”], [138, 109, “WP”], [139, 109, “ADV”], [140, 109, “WP”], [141, 109, “ADV”], [142, 109, “WP”], [143, 109, “WP”], [144, 109, “COO”], [145, 109, “ADV”], [146, 156, “WP”], [147, 156, “SBV”], [148, 156, “ADV”], [149, 151, “ATT”], [150, 151, “ATT”], [151, 156, “VOB”], [152, 156, “WP”], [153, 156, “ADV”], [154, 156, “ADV”], [155, 156, “WP”], [156, 167, “COO”], [157, 158, “ATT”], [158, 167, “VOB”], [159, 160, “WP”], [160, 167, “COO”], [161, 160, “WP”], [162, 163, “ADV”], [163, 167, “COO”], [164, 165, “WP”], [165, 163, “COO”], [166, 163, “ADV”], [167, 27, “ATT”], [168, 167, “RAD”], [169, 172, “ADV”], [170, 172, “WP”], [171, 172, “ADV”], [172, 7, “COO”], [173, 175, “ATT”], [174, 175, “RAD”], [175, 172, “VOB”], [176, 175, “WP”], [177, 175, “RAD”], [178, 175, “ATT”], [179, 181, “ADV”], [180, 181, “ADV”], [181, 6, “COO”], [182, 181, “RAD”], [183, 181, “WP”], [184, 181, “WP”], [185, 190, “SBV”], [186, 190, “SBV”], [187, 190, “WP”], [188, 190, “SBV”], [189, 190, “ADV”], [190, 181, “COO”], [191, 190, “VOB”], [192, 191, “RAD”], [193, 191, “WP”], [194, 195, “ADV”], [195, 204, “CMP”], [196, 204, “VOB”], [197, 204, “WP”], [198, 204, “SBV”], [199, 198, “RAD”], [200, 204, “LAD”], [201, 204, “SBV”], [202, 204, “ADV”], [203, 204, “ADV”], [204, 191, “COO”], [205, 204, “VOB”], [206, 204, “WP”], [207, 204, “ADV”], [208, 209, “SBV”], [209, 191, “ADV”], [210, 209, “WP”], [211, 209, “SBV”], [212, 209, “WP”], [213, 209, “SBV”], [214, 209, “WP”], [215, 216, “ADV”], [216, 218, “ATT”], [217, 216, “RAD”], [218, 209, “SBV”], [219, 209, “ADV”], [220, 191, “ADV”], [221, 191, “ADV”], [222, 181, “COO”], [223, 181, “WP”], [224, 181, “ADV”], [225, 181, “ADV”]], “sdp”: [[1, 2, “AGT”], [1, 129, “AGT”], [2, 0, “Root”], [3, 2, “mPUNC”], [4, 7, “mRELA”], [5, 7, “AGT”], [5, 8, “AGT”], [6, 7, “mNEG”], [6, 8, “mNEG”], [7, 2, “dCONT”], [8, 7, “eSUCC”], [9, 8, “LINK”], [10, 8, “mPUNC”], [11, 8, “eSUCC”], [12, 172, “EXP”], [13, 172, “SCO”], [14, 15, “mRELA”], [15, 167, “LOC”], [15, 172, “LOC”], [16, 172, “mDEPD”], [17, 172, “EXP”], [18, 172, “mDEPD”], [19, 172, “mDEPD”], [20, 172, “mNEG”], [21, 172, “AGT”], [22, 172, “SCO”], [23, 172, “EXP”], [24, 172, “EXP”], [25, 172, “MANN”], [26, 172, “mDEPD”], [27, 172, “CONT”], [28, 172, “mPUNC”], [29, 33, “mDEPD”], [30, 33, “AGT”], [31, 33, “mDEPD”], [32, 33, “mDEPD”], [33, 172, “eSUCC”], [34, 33, “mDEPD”], [35, 39, “FEAT”], [36, 39, “SCO”], [37, 39, “rEXP”], [38, 37, “mDEPD”], [39, 33, “CONT”], [40, 33, “mPUNC”], [41, 44, “LOC”], [42, 44, “mPUNC”], [43, 44, “mRELA”], [44, 33, “eSUCC”], [45, 44, “TIME”], [46, 44, “mPUNC”], [47, 48, “AGT”], [47, 60, “PAT”], [47, 204, “AGT”], [48, 55, “rDATV”], [49, 48, “mDEPD”], [50, 48, “LOC”], [51, 50, “mRELA”], [52, 50, “FEAT”], [53, 48, “mDEPD”], [54, 55, “MEAS”], [55, 44, “EXP”], [56, 44, “mPUNC”], [57, 44, “eCOO”], [58, 59, “EXP”], [59, 44, “eSUCC”], [60, 44, “eSUCC”], [61, 60, “mPUNC”], [62, 60, “PAT”], [63, 60, “eSUCC”], [64, 63, “mDEPD”], [65, 71, “mPUNC”], [66, 71, “TIME”], [67, 66, “mPUNC”], [68, 71, “mDEPD”], [69, 70, “MEAS”], [70, 71, “AGT”], [71, 60, “eCOO”], [72, 71, “dCONT”], [73, 72, “mPUNC”], [74, 75, “MEAS”], [75, 72, “AGT”], [76, 78, “mDEPD”], [77, 78, “mDEPD”], [78, 44, “eSUCC”], [79, 78, “dCONT”], [80, 79, “LINK”], [81, 79, “eCOO”], [82, 81, “TIME”], [83, 91, “mPUNC”], [84, 91, “LOC”], [85, 91, “mDEPD”], [86, 88, “EXP”], [87, 88, “mPUNC”], [88, 91, “mDEPD”], [89, 90, “mRELA”], [90, 91, “LOC”], [91, 79, “eSUCC”], [92, 91, “mDEPD”], [93, 79, “EXP”], [94, 93, “mPUNC”], [95, 93, “FEAT”], [96, 97, “mRELA”], [97, 93, “eCOO”], [98, 78, “mDEPD”], [99, 100, “mRELA”], [100, 101, “MANN”], [101, 78, “dCONT”], [102, 107, “FEAT”], [102, 109, “SCO”], [103, 109, “mDEPD”], [104, 109, “mPUNC”], [105, 107, “mRELA”], [106, 107, “FEAT”], [107, 109, “SCO”], [108, 109, “mDEPD”], [109, 101, “ePREC”], [110, 109, “mPUNC”], [111, 109, “eSUCC”], [112, 109, “mDEPD”], [113, 109, “eSUCC”], [114, 129, “mPUNC”], [115, 119, “mRELA”], [116, 119, “SCO”], [117, 119, “FEAT”], [118, 117, “mDEPD”], [119, 124, “STAT”], [120, 119, “mDEPD”], [121, 119, “mPUNC”], [122, 124, “mDEPD”], [123, 124, “mDEPD”], [124, 129, “dCONT”], [125, 124, “mDEPD”], [126, 129, “dCONT”], [127, 129, “mPUNC”], [128, 129, “AGT”], [129, 109, “eSUCC”], [130, 129, “dCONT”], [131, 129, “mPUNC”], [132, 109, “ePREC”], [133, 109, “eSUCC”], [134, 109, “mDEPD”], [135, 109, “SCO”], [136, 137, “MEAS”], [137, 109, “FEAT”], [138, 109, “mPUNC”], [139, 109, “FEAT”], [140, 109, “mPUNC”], [141, 109, “FEAT”], [142, 109, “mPUNC”], [143, 8, “mPUNC”], [143, 109, “mPUNC”], [144, 109, “eSUCC”], [145, 160, “mDEPD”], [146, 160, “mPUNC”], [147, 160, “eSUCC”], [148, 147, “mDEPD”], [149, 147, “MEAS”], [150, 151, “MEAS”], [151, 147, “TIME”], [152, 147, “mPUNC”], [153, 156, “mRELA”], [154, 156, “mRELA”], [155, 156, “mPUNC”], [156, 160, “mDEPD”], [157, 158, “MEAS”], [158, 156, “LINK”], [159, 156, “mPUNC”], [160, 109, “eSUCC”], [161, 160, “mPUNC”], [162, 163, “mDEPD”], [163, 160, “dEXP”], [164, 165, “mPUNC”], [165, 160, “eCOO”], [166, 167, “mRELA”], [167, 165, “dEXP”], [168, 167, “mDEPD”], [169, 167, “SCO”], [170, 160, “mPUNC”], [171, 172, “TIME”], [172, 11, “dCONT”], [173, 8, “MEAS”], [174, 175, “mDEPD”], [175, 8, “eSUCC”], [176, 175, “mPUNC”], [177, 181, “mDEPD”], [178, 181, “EXP”], [179, 181, “mDEPD”], [180, 181, “mDEPD”], [181, 2, “dCONT”], [182, 44, “mDEPD”], [182, 209, “mDEPD”], [183, 209, “mPUNC”], [184, 209, “mPUNC”], [185, 44, “AGT”], [185, 209, “EXP”], [186, 185, “eCOO”], [187, 209, “mPUNC”], [188, 191, “MEAS”], [189, 191, “mDEPD”], [190, 191, “eSUCC”], [191, 209, “dEXP”], [192, 204, “mDEPD”], [193, 204, “mPUNC”], [194, 195, “SCO”], [195, 204, “FEAT”], [196, 204, “CONT”], [197, 204, “mPUNC”], [198, 204, “AGT”], [199, 48, “mDEPD”], [199, 198, “mDEPD”], [200, 204, “mRELA”], [201, 204, “AGT”], [202, 204, “mDEPD”], [203, 204, “eCOO”], [204, 209, “ePREC”], [205, 204, “CONT”], [206, 204, “mPUNC”], [207, 204, “mDEPD”], [208, 204, “LOC”], [209, 181, “eSUCC”], [210, 209, “mPUNC”], [211, 209, “EXP”], [212, 209, “mPUNC”], [213, 209, “EXP”], [214, 209, “mPUNC”], [215, 216, “mDEPD”], [216, 218, “FEAT”], [217, 37, “mDEPD”], [217, 216, “mDEPD”], [218, 209, “EXP”], [219, 209, “mDEPD”], [220, 221, “mRELA”], [221, 209, “LOC”], [222, 181, “eSUCC”], [223, 181, “mPUNC”], [224, 181, “mRELA”], [225, 181, “SCO”]]}\u7d27\u63a5\u7740\uff0c\u8c03\u7528\u76f8\u5e94\u7684mask\u7b56\u7565\u5bf9\u6570\u636e\u8fdb\u884c\u5904\u7406\uff0c\u5904\u7406\u540e\u7684\u6570\u636e\u6837\u4f8b\u5982\u4e0b\uff1a[[‘[CLS]’, ‘\u6211’, ‘\u60f3’, ‘,’, ‘\u5982’, ‘\u679c’, ‘\u6211’, ‘\u6ca1’, ‘\u6709’, ‘\u53bb’, ‘\u505a’, ‘\u5927’, ‘\u5b66’, ‘\u751f’, ‘\u6751’, ‘\u5b98’, ‘,’, ‘\u6050’, ‘\u6015’, ‘\u6211’, ‘\u8fd9’, ‘\u4e2a’, ‘\u5728’, ‘\u6606’, ‘\u660e’, ‘\u957f’, ‘\u5927’, ‘\u7684’, ‘\u5b69’, ‘\u5b50’, ‘\u6c38’, ‘\u8fdc’, ‘\u90fd’, ‘\u4e0d’, ‘\u80fd’, ‘\u5207’, ‘\u8eab’, ‘\u611f’, ‘\u53d7’, ‘\u5230’, ‘\u4e91’, ‘\u5357’, ‘\u8fd9’, ‘\u6b21’, ‘6’, ‘0’, ‘\u5e74’, ‘\u4e00’, ‘\u9047’, ‘\u7684’, ‘\u7279’, ‘\u5927’, ‘\u65f1’, ‘\u60c5’, ‘\u7684’, ‘\u4e25’, ‘\u91cd’, ‘\u6027’, ‘,’, ‘\u6050’, ‘\u6015’, ‘[sdp]’, ‘\u6211’, ‘[sdp]’, ‘\u53ea’, ‘\u662f’, ‘\u6bcf’, ‘\u5929’, ‘[sdp]’, ‘\u770b’, ‘[sdp]’, ‘\u7740’, ‘\u65b0’, ‘\u95fb’, ‘\u4e0a’, ‘\u90a3’, ‘\u4e9b’, ‘\u7f3a’, ‘\u6c34’, ‘\u7684’, ‘\u955c’, ‘\u5934’, ‘,’, ‘\u5634’, ‘\u4e0a’, ‘\u8bf4’, ‘\u8981’, ‘\u8282’, ‘\u6c34’, ‘,’, ‘\u4f46’, ‘\u4e8b’, ‘\u5b9e’, ‘\u884c’, ‘\u52a8’, ‘\u4fdd’, ‘\u6301’, ‘\u4e0d’, ‘\u4e86’, ‘\u4e09’, ‘\u5929’, ‘\u3002’, ‘\u6211’, ‘\u4efb’, ‘\u804c’, ‘\u7684’, ‘\u5730’, ‘\u65b9’, ‘\u5728’, ‘\u6606’, ‘\u660e’, ‘\u5e02’, ‘\u7984’, ‘\u529d’, ‘\u53bf’, ‘\u7684’, ‘\u4e00’, ‘\u4e2a’, ‘\u6751’, ‘\u59d4’, ‘\u4f1a’, ‘,’, ‘\u8bf4’, ‘\u5b9e’, ‘\u8bdd’, ‘\u8fd9’, ‘\u91cc’, ‘[SEP]’], [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [‘\u6606\u660e\u957f\u5927’, ‘\u4e91\u5357’, ‘\u6606\u660e\u5e02\u7984\u529d\u53bf’]]\u9884\u8bad\u7ec3\u811a\u672c\u6570\u636e\u5904\u7406\u5b8c\u6bd5\u4e4b\u540e\uff0c\u5c31\u53ef\u4ee5\u8c03\u7528\u9884\u8bad\u7ec3\u811a\u672c\u8fdb\u884c\u6a21\u578b\u7684\u9884\u8bad\u7ec3\uff0c\u811a\u672c\u5982\u4e0b\uff1agpu_number=1
negative_e_number=4
negative_e_length=16<\/p>\n
base_dir=$PWD
checkpoint_dir=$base_dir\/checkpoints
resources=$base_dir\/resources
local_kg=$resources\/ownthink_triples_small.txt
local_train_file=$resources\/train_small.txt
remote_kg=https:\/\/atp-modelzoo-sh.oss-c…<\/a>
remote_train_file=https:\/\/atp-modelzoo-sh.oss-c…<\/a><\/p>\n
if [ ! -d $checkpoint_dir ];then
mkdir $checkpoint_dir
fi<\/p>\n
if [ ! -d $resources ];then
mkdir $resources
fi<\/p>\n
if [ ! -f $local_kg ];then
wget -P $resources $remote_kg
fi<\/p>\n
if [ ! -f $local_train_file ];then
wget -P $resources $remote_train_file
fi<\/p>\n
python -m torch.distributed.launch –nproc_per_node=$gpu_number \/
–master_port=52349 \/
$base_dir\/main.py \/
–mode=train \/
–worker_gpu=$gpu_number \/
–tables=$local_train_file, \/
–learning_rate=5e-5  \/
–epoch_num=5  \/
–logging_steps=10 \/
–save_checkpoint_steps=2150 \/
–sequence_length=256 \/
–train_batch_size=20 \/
–checkpoint_dir=$checkpoint_dir \/
–app_name=language_modeling \/
–use_amp \/
–save_all_checkpoints \/
–user_defined_parameters=”pretrain_model_name_or_path=hfl\/macbert-base-zh external_mask_flag=True contrast_learning_flag=True negative_e_number=${negative_e_number} negative_e_length=${negative_e_length} kg_path=${local_kg}"\u6a21\u578bFinetuneCKBERT\u6a21\u578b\u4e0eBERT\u662f\u540c\u6837\u7684\u67b6\u6784\uff0c\u53ea\u9700\u8981\u4f7f\u7528\u901a\u7528\u7684EasyNLP\u6846\u67b6\u547d\u4ee4\u5c31\u53ef\u4ee5\u8fdb\u884c\u8c03\u7528\u3002\u4ee5\u4e0b\u547d\u4ee4\u5206\u522b\u4e3aTrain\u548cPredict\u72b6\u6001\u7684\u4f8b\u5b50\uff0c\u4f7f\u7528\u7684\u6a21\u578b\u4e3ackbert-base\u3002\u5f53\u524d\u5728EasyNLP\u6846\u67b6\u4e2d\u4e5f\u53ef\u4ee5\u8c03\u7528large\u548chuge\u6a21\u578b\u8fdb\u884c\u6d4b\u8bd5\uff0c\u53ea\u9700\u8981\u66ff\u6362\u547d\u4ee4\u4e2d\u7684\u53c2\u6570\u5373\u53efpretrain_model_name_or_path=alibaba-pai\/pai-ckbert-large-zhpretrain_model_name_or_path=alibaba-pai\/pai-ckbert-huge-zh$ easynlp \/
   –mode=train \/
   –worker_gpu=1 \/
   –tables=train.tsv,dev.tsv \/
   –input_schema=label:str:1,sid1:str:1,sid2:str:1,sent1:str:1,sent2:str:1 \/
   –first_sequence=sent1 \/
   –label_name=label \/
   –label_enumerate_values=0,1 \/
   –checkpoint_dir=.\/classification_model \/
   –epoch_num=1  \/
   –sequence_length=128 \/
   –app_name=text_classify \/
   –user_defined_parameters=’pretrain_model_name_or_path=alibaba-pai\/pai-ckbert-base-zh’$ easynlp \/
  –mode=predict \/
  –tables=dev.tsv \/
  –outputs=dev.pred.tsv \/
  –input_schema=label:str:1,sid1:str:1,sid2:str:1,sent1:str:1,sent2:str:1 \/
  –output_schema=predictions,probabilities,logits,output \/
  –append_cols=label \/
  –first_sequence=sent1 \/
  –checkpoint_path=.\/classification_model \/
  –app_name=text_classify\u5728HuggingFace\u4e0a\u4f7f\u7528CKBERT\u6a21\u578b\u4e3a\u4e86\u65b9\u4fbf\u5f00\u6e90\u7528\u6237\u4f7f\u7528CKBERT\uff0c\u6211\u4eec\u4e5f\u5c06\u4e09\u4e2aCKBERT\u6a21\u578b\u5728HuggingFace Models\u4e0a\u67b6\uff0c\u5176Model Card\u5982\u4e0b\u6240\u793a\uff1a
\"\u56fe\u7247\"
\u7528\u6237\u4e5f\u53ef\u4ee5\u76f4\u63a5\u4f7f\u7528HuggingFace\u63d0\u4f9b\u7684pipeline\u8fdb\u884c\u6a21\u578b\u63a8\u7406\uff0c\u6837\u4f8b\u5982\u4e0b\uff1afrom transformers import AutoTokenizer, AutoModelForMaskedLM, FillMaskPipeline<\/p>\n
tokenizer = AutoTokenizer.from_pretrained(“alibaba-pai\/pai-ckbert-base-zh”, use_auth_token=True)
model = AutoModelForMaskedLM.from_pretrained(“alibaba-pai\/pai-ckbert-base-zh”, use_auth_token=True)
unmasker = FillMaskPipeline(model, tokenizer)   
unmasker(“\u5df4\u9ece\u662f[MASK]\u56fd\u7684\u9996\u90fd\u3002”,top_k=5)<\/p>\n
[<\/p>\n
{'score': 0.8580496311187744, \n 'token': 3791, \n 'token_str': '\u6cd5', \n 'sequence': '\u5df4 \u9ece \u662f \u6cd5 \u56fd \u7684 \u9996 \u90fd \u3002'}, \n{'score': 0.08550138026475906, \n 'token': 2548, \n 'token_str': '\u5fb7', \n 'sequence': '\u5df4 \u9ece \u662f \u5fb7 \u56fd \u7684 \u9996 \u90fd \u3002'}, \n{'score': 0.023137662559747696, \n 'token': 5401, \n 'token_str': '\u7f8e', \n 'sequence': '\u5df4 \u9ece \u662f \u7f8e \u56fd \u7684 \u9996 \u90fd \u3002'}, \n{'score': 0.012281022034585476, \n 'token': 5739, 'token_str': '\u82f1', \n 'sequence': '\u5df4 \u9ece \u662f \u82f1 \u56fd \u7684 \u9996 \u90fd \u3002'}, \n{'score': 0.005729076452553272, \n 'token': 704, 'token_str': '\u4e2d', \n 'sequence': '\u5df4 \u9ece \u662f \u4e2d \u56fd \u7684 \u9996 \u90fd \u3002'}<\/code><\/pre>\n
]\u6216\u8005\u4e5f\u53ef\u4ee5\u4f7f\u7528Pytorch\u52a0\u8f7d\u6a21\u578b\uff0c\u6837\u4f8b\u5982\u4e0b\uff1afrom transformers import AutoTokenizer, AutoModelForMaskedLM<\/p>\n
tokenizer = AutoTokenizer.from_pretrained(“alibaba-pai\/pai-ckbert-base-zh”, use_auth_token=True)
model = AutoModelForMaskedLM.from_pretrained(“alibaba-pai\/pai-ckbert-base-zh”, use_auth_token=True)
text = “\u5df4\u9ece\u662f[MASK]\u56fd\u7684\u9996\u90fd\u3002”
encoded_input = tokenizer(text, return_tensors=’pt’)
output = model(**encoded_input)\u5728\u963f\u91cc\u4e91\u673a\u5668\u5b66\u4e60\u5e73\u53f0PAI\u4e0a\u4f7f\u7528CKBERT\u6a21\u578bPAI-DSW\uff08Data Science Workshop\uff09\u662f\u963f\u91cc\u4e91\u673a\u5668\u5b66\u4e60\u5e73\u53f0PAI\u5f00\u53d1\u7684\u4e91\u4e0aIDE\uff0c\u9762\u5411\u4e0d\u540c\u6c34\u5e73\u7684\u5f00\u53d1\u8005\uff0c\u63d0\u4f9b\u4e86\u4ea4\u4e92\u5f0f\u7684\u7f16\u7a0b\u73af\u5883\uff08\u6587\u6863\uff09\u3002\u5728DSW Gallery\u4e2d\uff0c\u63d0\u4f9b\u4e86\u5404\u79cdNotebook\u793a\u4f8b\uff0c\u65b9\u4fbf\u7528\u6237\u8f7b\u677e\u4e0a\u624bDSW\uff0c\u642d\u5efa\u5404\u79cd\u673a\u5668\u5b66\u4e60\u5e94\u7528\u3002\u6211\u4eec\u4e5f\u5728DSW Gallery\u4e2d\u4e0a\u67b6\u4e86\u4f7f\u7528CKBERT\u8fdb\u884c\u4e2d\u6587\u547d\u540d\u5b9e\u4f53\u8bc6\u522b\u7684Sample Notebook\uff08\u89c1\u4e0b\u56fe\uff09\uff0c\u6b22\u8fce\u5927\u5bb6\u4f53\u9a8c\uff01
\"\u56fe\u7247\"
\u672a\u6765\u5c55\u671b\u5728\u672a\u6765\uff0c\u6211\u4eec\u8ba1\u5212\u5728EasyNLP\u6846\u67b6\u4e2d\u96c6\u6210\u66f4\u591a\u4e2d\u2f42\u77e5\u8bc6\u6a21\u578b\uff0c\u8986\u76d6\u5404\u4e2a\u5e38\u2ec5\u4e2d\u2f42\u9886\u57df\uff0c\u656c\u8bf7\u671f\u5f85\u3002\u6211\u4eec\u4e5f\u5c06\u5728EasyNLP\u6846\u67b6\u4e2d\u96c6\u6210\u66f4\u591aSOTA\u6a21\u578b\uff08\u7279\u522b\u662f\u4e2d\u2f42\u6a21\u578b\uff09\uff0c\u6765\u2f40\u6301\u5404\u79cdNLP\u548c\u591a\u6a21\u6001\u4efb\u52a1\u3002\u6b64\u5916\uff0c \u963f\u2fa5\u4e91\u673a\u5668\u5b66\u4e60PAI\u56e2\u961f\u4e5f\u5728\u6301\u7eed\u63a8\u8fdb\u4e2d\u2f42\u591a\u6a21\u6001\u6a21\u578b\u7684\u2f83\u7814\u2f2f\u4f5c\uff0c\u6b22\u8fce\u2f64\u6237\u6301\u7eed\u5173\u6ce8\u6211\u4eec\uff0c\u4e5f\u6b22\u8fce\u52a0\u2f0a \u6211\u4eec\u7684\u5f00\u6e90\u793e\u533a\uff0c\u5171\u5efa\u4e2d\u2f42NLP\u548c\u591a\u6a21\u6001\u7b97\u6cd5\u5e93\uff01Github\u5730\u5740\uff1ahttps:\/\/github.com\/alibaba\/Ea…<\/a> Wang, Minghui Qiu, Taolin Zhang, Tingting Liu, Lei Li, Jianing Wang, Ming Wang, Jun Huang, Wei Lin. EasyNLP: A Comprehensive and Easy-to-use Toolkit for Natural Language Processing. EMNLP 2022Taolin Zhang, Junwei Dong, Jianing Wang, Chengyu Wang, Ang Wang, Yinghui Liu, Jun Huang, Yong Li, Xiaofeng He. Revisiting and Advancing Chinese Natural Language Understanding with Accelerated Heterogeneous Knowledge Pre-training. EMNLP 2022Yiming Cui, Wanxiang Che, Ting Liu, Bing Qin, Shijin Wang, Guoping Hu. Revisiting Pre-Trained Models for Chinese Natural Language Processing. EMNLP (Findings) 2020Yiming Cui, Ziqing Yang, Ting Liu. PERT: Pre-training BERT with Permuted Language Model. arXivYu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu. ERNIE: Enhanced Representation through Knowledge Integration. arXivYuxuan Lai, Yijia Liu, Yansong Feng, Songfang Huang, and Dongyan Zhao. Lattice-BERT: Leveraging Multi-Granularity Representations in Chinese Pre-trained Language Models. NAACL 2021Weijie Liu, Peng Zhou, Zhe Zhao, Zhiruo Wang, Qi Ju, Haotang Deng, Ping Wang. K-BERT: Enabling Language Representation with Knowledge Graph. AAAI 2020Zhengyan Zhang, Xu Han, Zhiyuan Liu, Xin Jiang, Maosong Sun, Qun Liu. ERNIE: Enhanced Language Representation with Informative Entities. ACL 2019\u963f\u91cc\u7075\u6770\u56de\u987e\u963f\u91cc\u7075\u6770\uff1a\u963f\u91cc\u4e91\u673a\u5668\u5b66\u4e60PAI\u5f00\u6e90\u4e2d\u6587NLP\u7b97\u6cd5\u6846\u67b6EasyNLP\uff0c\u52a9\u529bNLP\u5927\u6a21\u578b\u843d\u5730\u963f\u91cc\u7075\u6770\uff1a\u9884\u8bad\u7ec3\u77e5\u8bc6\u5ea6\u91cf\u6bd4\u8d5b\u593a\u51a0\uff01\u963f\u91cc\u4e91PAI\u53d1\u5e03\u77e5\u8bc6\u9884\u8bad\u7ec3\u5de5\u5177\u963f\u91cc\u7075\u6770\uff1aEasyNLP\u5e26\u4f60\u73a9\u8f6cCLIP\u56fe\u6587\u68c0\u7d22\u963f\u91cc\u7075\u6770\uff1aEasyNLP\u4e2d\u6587\u6587\u56fe\u751f\u6210\u6a21\u578b\u5e26\u4f60\u79d2\u53d8\u827a\u672f\u5bb6\u963f\u91cc\u7075\u6770\uff1aEasyNLP\u96c6\u6210K-BERT\u7b97\u6cd5\uff0c\u501f\u52a9\u77e5\u8bc6\u56fe\u8c31\u5b9e\u73b0\u66f4\u4f18Finetune\u963f\u91cc\u7075\u6770\uff1a\u4e2d\u6587\u7a00\u758fGPT\u5927\u6a21\u578b\u843d\u5730 \u2014 \u901a\u5f80\u4f4e\u6210\u672c&\u9ad8\u6027\u80fd\u591a\u4efb\u52a1\u901a\u7528\u81ea\u7136\u8bed\u8a00\u7406\u89e3\u7684\u5173\u952e\u91cc\u7a0b\u7891\u963f\u91cc\u7075\u6770\uff1aEasyNLP\u73a9\u8f6c\u6587\u672c\u6458\u8981\uff08\u65b0\u95fb\u6807\u9898\uff09\u751f\u6210\u963f\u91cc\u7075\u6770\uff1a\u8de8\u6a21\u6001\u5b66\u4e60\u80fd\u529b\u518d\u5347\u7ea7\uff0cEasyNLP\u7535\u5546\u6587\u56fe\u68c0\u7d22\u6548\u679c\u5237\u65b0SOTA\u963f\u91cc\u7075\u6770\uff1aEasyNLP\u5e26\u4f60\u5b9e\u73b0\u4e2d\u82f1\u6587\u673a\u5668\u9605\u8bfb\u7406\u89e3
\u539f\u6587\u94fe\u63a5\uff1ahttps:\/\/click.aliyun.com\/m\/10…<\/a>
\u672c\u6587\u4e3a\u963f\u91cc\u4e91\u539f\u521b\u5185\u5bb9\uff0c\u672a\u7ecf\u5141\u8bb8\u4e0d\u5f97\u8f6c\u8f7d\u3002<\/p>\n
https:\/\/segmentfault.com\/a\/1190000042727896<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"
\u7b80\u4ecb\uff1a\u00a0\u672c\u2f42\u7b80\u8981\u4ecb\u7ecdCKBERT\u7684\u6280\u672f\u89e3\u8bfb\uff0c\u4ee5\u53ca\u5982\u4f55\u5728EasyNLP\u6846\u67b6\u3001HuggingFace Models […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[4],"tags":[],"class_list":["post-10006803","post","type-post","status-publish","format-standard","hentry","category-share"],"jetpack_sharing_enabled":true,"jetpack_featured_media_url":"","_links":{"self":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts\/10006803","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=10006803"}],"version-history":[{"count":0,"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts\/10006803\/revisions"}],"wp:attachment":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=10006803"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=10006803"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=10006803"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}