{"id":10006236,"date":"2022-08-17T16:21:41","date_gmt":"2022-08-17T08:21:41","guid":{"rendered":"https:\/\/ityet.com\/?p=10006236"},"modified":"2022-08-17T16:21:41","modified_gmt":"2022-08-17T08:21:41","slug":"easynlp%e9%9b%86%e6%88%90k-bert%e7%ae%97%e6%b3%95%ef%bc%8c%e5%80%9f%e5%8a%a9%e7%9f%a5%e8%af%86%e5%9b%be%e8%b0%b1%e5%ae%9e%e7%8e%b0%e6%9b%b4%e4%bc%98finetune","status":"publish","type":"post","link":"https:\/\/itmax.cn\/?p=10006236","title":{"rendered":"EasyNLP\u96c6\u6210K-BERT\u7b97\u6cd5\uff0c\u501f\u52a9\u77e5\u8bc6\u56fe\u8c31\u5b9e\u73b0\u66f4\u4f18Finetune"},"content":{"rendered":"<p>\u5bfc\u8bfb\u77e5\u8bc6\u56fe\u8c31\uff08Knowledge Graph\uff09\u7684\u6982\u5ff5\u2fb8\u6b21\u51fa\u73b02012\u5e74\uff0c\u7531Google\u63d0\u51fa\uff0c\u5b83\u4f5c\u4e3a\u2f00\u79cd\u2f24\u89c4\u6a21\u8bed\u4e49\u2f79\u7edc\uff0c \u51c6\u786e\u5730\u63cf\u8ff0\u4e86\u5b9e\u4f53\u4ee5\u53ca\u5b9e\u4f53\u4e4b\u95f4\u7684\u5173\u7cfb\u3002\u77e5\u8bc6\u56fe\u8c31\u6700\u65e9\u5e94\u2f64\u4e8e\u641c\u7d22\u5f15\u64ce\uff0c\u2f64\u4e8e\u51c6\u5907\u8fd4\u56de\u2f64\u6237\u6240\u9700\u7684\u77e5\u8bc6\u3002\u968f\u7740\u9884\u8bad\u7ec3\u8bed\u2f94\u6a21\u578b\u7684\u4e0d\u65ad\u53d1\u5c55\uff0c\u5c06\u77e5\u8bc6\u56fe\u8c31\u4e2d\u7684\u77e5\u8bc6\u4e09\u5143\u7ec4\u878d\u2f0a\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u5bf9\u63d0\u5347\u6a21\u578b\u7684\u6548\u679c\u6709\u91cd\u8981\u7684\u4f5c\u2f64\u3002\u7ecf\u5178\u7684\u77e5\u8bc6\u6ce8\u2f0a\u7b97\u6cd5\u76f4\u63a5\u5c06\u77e5\u8bc6\u56fe\u8c31\u4e2d\u5b9e\u4f53\u7684\u8bcd\u5d4c\u2f0a\u5411\u91cf\u6ce8\u2f0a\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u8bcd\u5d4c\u2f0a\u5411\u91cf\u4e2d\uff1b\u7136\u2f7d\uff0c\u77e5\u8bc6\u56fe\u8c31\u4e2d\u5b9e\u4f53\u7684\u8bcd\u5d4c\u2f0a\u5411\u91cf\u548c\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u8bcd\u5d4c\u2f0a\u5411\u91cf\u7684\u5411\u91cf\u7a7a\u95f4\u4e0d\u2f00\u81f4\uff0c\u2f7d\u4e14\u5bb9\u6613\u5f15\u8d77\u77e5\u8bc6\u566a\u2fb3\uff08Knowledge Noise\uff09\u7684\u95ee\u9898\u3002K-BERT\u7b97\u6cd5\u63d0\u51fa\u5229\u2f64Visible Matrix\u5728BERT\u6a21\u578b\u4e2d\u5f15\u2f0a\u5916\u90e8\u77e5\u8bc6\uff0c\u5177\u6709\u8f83\u597d\u7684\u6548\u679c\u3002 \u56e0\u6b64\uff0c\u6211\u4eec\u5728EasyNLP\u8fd9\u2f00\u7b97\u6cd5\u6846\u67b6\u4e2d\u96c6\u6210\u4e86K-BERT\u7b97\u6cd5\uff0c\u4f7f\u2f64\u6237\u5728\u5177\u6709\u77e5\u8bc6\u56fe\u8c31\u7684\u60c5\u51b5\u4e0b\uff0c\u53d6\u5f97\u66f4\u597d\u7684\u6a21\u578bFinetune\u6548\u679c\u3002EasyNLP\uff08<a href=\"https:\/\/link.segmentfault.com\/?enc=UTGrRlOZ6C50vHlYcGFzGw%3D%3D.Q5XUMGvaIT3W3zXhbvmd%2B%2FjkypPQexf2a4EgjtR2PRD1gb3Bf4yHdP9Iql8JRoZz\" rel=\"nofollow\">https:\/\/github.com\/alibaba\/Ea&#8230;<\/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\u672c\u2f42\u7b80\u8981\u4ecb\u7ecdK-BERT\u7684\u6280\u672f\u89e3\u8bfb\uff0c\u4ee5\u53ca\u5982\u4f55\u5728EasyNLP\u6846\u67b6\u4e2d\u4f7f\u2f64K-BERT\u6a21\u578b\u3002K-BERT\u6a21\u578b\u8be6\u89e3BERT\u7b49\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u4ece\u5927\u89c4\u6a21\u8bed\u6599\u5e93\u4e2d\u6355\u83b7\u6587\u672c\u8bed\u8a00\u8868\u793a\uff0c\u4f46\u7f3a\u4e4f\u9886\u57df\u7279\u5b9a\u7684\u77e5\u8bc6\u3002\u800c\u9886\u57df\u4e13\u5bb6\u5728\u9605\u8bfb\u9886\u57df\u6587\u672c\u65f6\uff0c\u4f1a\u5229\u7528\u76f8\u5173\u77e5\u8bc6\u8fdb\u884c\u63a8\u7406\u3002\u4e3a\u4e86\u5b9e\u73b0\u8fd9\u4e00\u529f\u80fd\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\u3002K-BERT\u7684\u6a21\u578b\u67b6\u6784\u548c\u77e5\u8bc6\u6ce8\u5165\u7684\u5e94\u7528\u793a\u4f8b\u5982\u4e0b\u6240\u793a\uff1a<br \/><img decoding=\"async\" referrerpolicy=\"no-referrer\" src=\"https:\/\/segmentfault.com\/img\/remote\/1460000042319351\" alt=\"\u56fe\u7247\" title=\"\u56fe\u7247\" loading=\"lazy\"><br \/>\u00a0\u7279\u522b\u5730\uff0c\u5728\u6a21\u578b\u7684\u8f93\u5165\u5c42\uff0cK-BERT\u8868\u5f81\u5c42\u901a\u8fc7\u4f4d\u7f6e\u7d22\u5f15\u5c06\u8bed\u53e5\u6811\u8f6c\u6362\u4e3a\u53e5\u5b50\uff0c\u6784\u9020\u4e09\u79cd\u8868\u5f81\u65b9\u5f0f\uff1aToken\u8868\u5f81\u3001Soft-position\u8868\u5f81\u548cSegment\u8868\u5f81\u3002\u5176\u4e2dSoft-position\u8868\u5f81\u4f5c\u4e3a\u4f4d\u7f6e\u5d4c\u5165\uff0c\u4e3a\u91cd\u6392\u7684\u5e8f\u5217\u63d0\u4f9b\u53e5\u5b50\u7ed3\u6784\u4fe1\u606f\u3002\u6b64\u5916\uff0c\u901a\u8fc7Visible Matrix\u6765\u8986\u76d6\u4e0d\u5e94\u8be5\u770b\u5230\u7684Token\uff0c\u907f\u514d\u4e86\u77e5\u8bc6\u566a\u58f0\u95ee\u9898\u3002\u5982\u4e0b\u56fe\u6240\u793a\uff0c\u4ee5Beijing\u4e3a\u4f8b\uff0c\u7ed9\u5b9a\u77e5\u8bc6\u4e09\u5143\u7ec4\uff08Beijing\uff0ccapital\uff0cChina\uff09\uff0cK-BERT\u901a\u8fc7Visible Matrix\u9650\u5236\u53ea\u6709Beijing\u80fd\u201c\u770b\u5230\u201d\u5176\u76f8\u5173\u7684\u5173\u7cfb\u8bcd\u548c\u5bbe\u8bed\uff0c\u5206\u522b\u4e3a\u201ccapital\u201d\u548c\u201cChina\u201d\u3002\u4e0e\u4e4b\u76f8\u53cd\uff0c\u4e00\u4e2a\u77e5\u8bc6\u65e0\u5173\u7684\u8bcdnow\u5219\u65e0\u6cd5\u201c\u770b\u5230\u201d\u201ccapital\u201d\u548c\u201cChina\u201d\u3002<br \/><img decoding=\"async\" referrerpolicy=\"no-referrer\" src=\"https:\/\/segmentfault.com\/img\/remote\/1460000042319352\" alt=\"\u56fe\u7247\" title=\"\u56fe\u7247\" loading=\"lazy\"><br \/>\u4ece\u4f5c\u8005\u8bba\u6587\u4e2d\u6211\u4eec\u4e5f\u53ef\u4ee5\u770b\u5230K-BERT\u4e2d\u7684Attention Matrix\u7684\u8ba1\u7b97\u8fc7\u7a0b\uff1a<br \/><img decoding=\"async\" referrerpolicy=\"no-referrer\" src=\"https:\/\/segmentfault.com\/img\/remote\/1460000042319353\" alt=\"\u56fe\u7247\" title=\"\u56fe\u7247\" loading=\"lazy\"><br \/>\u5176\u4e2d\uff0cM\u5373\u4e3aVisible Matrix\uff0c\u4e3a\u4e86\u8868\u793aK-BERT\u8f93\u5165Token\u5e8f\u5217\u80fd\u5426\u4e92\u76f8\u201c\u770b\u89c1\u201d\uff0c\u5b9a\u4e49M\u5982\u4e0b\uff1a<br \/><img decoding=\"async\" referrerpolicy=\"no-referrer\" src=\"https:\/\/segmentfault.com\/img\/remote\/1460000042319354\" alt=\"\u56fe\u7247\" title=\"\u56fe\u7247\" loading=\"lazy\"><br \/>\u56e0\u6b64\uff0c\u5f53\u4e24\u4e2aToken\u4e92\u76f8\u53ef\u4ee5\u201c\u770b\u89c1\u201d\uff0cM\u77e9\u9635\u5bf9\u5e94\u7684\u5143\u7d20\u503c\u4e3a0\uff0c\u56e0\u6b64Attention Matrix\u7684\u8ba1\u7b97\u8fc7\u7a0b\u548c\u666e\u901aBERT\u76f8\u540c\u3002\u53cd\u4e4b\uff0c\u5f53\u4e24\u4e2aToken\u4e0d\u80fd\u4e92\u76f8\u201c\u770b\u89c1\u201d\uff0cM\u77e9\u9635\u5bf9\u5e94\u7684\u5143\u7d20\u503c\u4e3a\u8d1f\u65e0\u7a77\uff0c\u5bf9\u5e94SoftMax\u51fd\u6570\u540e\u7684\u6743\u91cd\u5219\u4f1a\u8d8b\u4e8e0\uff0c\u8fd9\u4f7f\u5f97\u8fd9\u4e24\u4e2aToken\u5728Self-Attention\u8ba1\u7b97\u8fc7\u7a0b\u4e2d\u4e92\u65e0\u5f71\u54cd\u3002\u8fd9\u5c31\u5728\u8ba1\u7b97\u8fc7\u7a0b\u4e2d\uff0c\u5927\u5927\u7f13\u89e3\u4e86\u77e5\u8bc6\u56fe\u8c31\u589e\u5f3a\u8fc7\u7a0b\u7684\u77e5\u8bc6\u566a\u58f0\u95ee\u9898\u3002K-BERT\u6a21\u578b\u7684\u5b9e\u73b0\u4e0e\u6548\u679c\u5728EasyNLP\u6846\u67b6\u4e2d\uff0c\u6211\u4eec\u5728\u6a21\u578b\u5c42\u6784\u5efa\u4e86K-BERT\u6a21\u578b\u7684Backbone\uff0c\u5176\u6838\u2f3c\u4ee3\u7801\u5982\u4e0b\u6240\u793a\uff1aself.kbert = KBertModel(config, add_pooling_layer=False)<br \/>self.cls = KBertOnlyMLMHead(config)<\/p>\n<p>outputs = self.kbert(<\/p>\n<pre><code>input_ids,\nattention_mask=attention_mask,\ntoken_type_ids=token_type_ids,\nposition_ids=position_ids,\nhead_mask=head_mask,\ninputs_embeds=inputs_embeds,\nencoder_hidden_states=encoder_hidden_states,\nencoder_attention_mask=encoder_attention_mask,\noutput_attentions=output_attentions,\noutput_hidden_states=output_hidden_states,\nreturn_dict=return_dict,<\/code><\/pre>\n<p>)<\/p>\n<p>sequence_output = outputs[0]<br \/>prediction_scores = self.cls(sequence_output)<br \/>loss_fct = CrossEntropyLoss()<br \/>masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))\u5176\u4e2d\uff0c\u5728K-BERT\u4e2d\uff0c\u6a21\u578bBackbone\u7684Attention Mask\u7531\u4e24\u4e2aMatrix\u52a0\u548c\u800c\u6210\uff0c\u5206\u522b\u4e3a\u666e\u901a\u7684Attention Mask\u548cVisible Matrix\uff0c\u5b9e\u73b0\u6838\u5fc3\u4ee3\u7801\u5982\u4e0b\uff1aextended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape, device) + self.get_extended_attention_mask(visible_matrix, input_shape, device)<br \/>\u5728\u6570\u636e\u9884\u5904\u7406\u8fc7\u7a0b\u4e2d\uff0c\u6211\u4eec\u9700\u8981\u83b7\u5f97\u5f53\u524d\u6837\u672c\u7684\u8f93\u5165\u6587\u672c\u548c\u77e5\u8bc6\u4e09\u5143\u7ec4\uff0c\u4ece\u800c\u8ba1\u7b97\u51faVisible Matrix\uff1a\u00a0 if self.kbert_model_prefix:<\/p>\n<pre><code>    encoding['input_ids'], encoding['token_type_ids'], encoding['attention_mask'], encoding['position_ids'], encoding['visible_matrix'] = self.kbert_row_data_process(encoding['input_ids'], encoding['token_type_ids'], encoding['attention_mask'])\u4e3a\u4e86\u9a8c\u8bc1EasyNLP\u6846\u67b6\u4e2dK-BERT\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\u3002\u6211\u4eec\u4f7f\u7528EasyNLP\u52a0\u8f7d\u4e86BERT\u6a21\u578b\uff0c\u5bf9\u6bd4\u590d\u73b0\u7ed3\u679c\u548cK-BERT\u5b98\u65b9\u8bba\u6587\u7684\u7ed3\u679c\uff0c\u5982\u4e0b\u6240\u793a\uff1a\u6570\u636e\u96c6Dev\u590d\u73b0\u7ed3\u679cDev\u8bba\u6587\u7ed3\u679cTest\u590d\u73b0\u7ed3\u679cTest\u8bba\u6587\u7ed3\u679cBook_review88.588.687.0687.2Chnsenticorp94.394.695.0895.6MSRA-NER94.5694.594.4694.5\u53ef\u4ee5\u901a\u8fc7\u4e0a\u8ff0\u7ed3\u679c\uff0c\u9a8c\u8bc1EasyNLP\u6846\u67b6\u4e2dK-BERT\u7b97\u6cd5\u5b9e\u73b0\u7684\u6b63\u786e\u6027\u3002K-BERT\u6a21\u578b\u4f7f\u2f64\u6559\u7a0b\u4ee5\u4e0b\u6211\u4eec\u7b80\u8981\u4ecb\u7ecd\u5982\u4f55\u5728EasyNLP\u6846\u67b6\u4f7f\u2f64K-BERT\u6a21\u578b\u3002\u5b89\u88c5EasyNLP\u2f64\u6237\u53ef\u4ee5\u76f4\u63a5\u53c2\u8003GitHub\uff08https:\/\/github.com\/alibaba\/EasyNLP\uff09\u4e0a\u7684\u8bf4\u660e\u5b89\u88c5EasyNLP\u7b97\u6cd5\u6846\u67b6\u3002\u6570\u636e\u51c6\u5907K-BERT\u662f\u4e00\u4e2afinetune\u6a21\u578b\uff0c\u9700\u8981\u7528\u6237\u51c6\u5907\u4e0b\u6e38\u4efb\u52a1\u7684\u8bad\u7ec3\u4e0e\u9a8c\u8bc1\u6570\u636e\uff0c\u4e3atsv\u6587\u4ef6\u3002\u5bf9\u4e8e\u6587\u672c\u5206\u7c7b\u4efb\u52a1\uff0c\u8fd9\u2f00\u2f42\u4ef6\u5305\u542b\u4ee5\u5236\u8868\u7b26\/t\u5206\u9694\u7684\u4e09\u5217\uff0c\u7b2c\u4e00\u5217\u662f\u6807\u7b7e\uff0c\u7b2c\u4e8c\u5217\u662f\u53e5\u5b50\u5e8f\u53f7\uff0c\u7b2c\u4e09\u5217\u662f\u6587\u672c\u53e5\u5b50\u3002\u6837\u4f8b\u5982\u4e0b\uff1a0   25  \u4f5c\u8005\u80af\u5b9a\u662f\u624b\u5934\u6709\u4e00\u672c\u5e74\u8868\u548c\u540d\u518c\u7684\uff0c\u4eba\u7269\u4e8b\u4ef6\u884c\u52a8\u5b8c\u5168\u6263\u5408\u6b63\u53f2\uff0c\u4eba\u7269\u6027\u683c\u4e5f\u591a\u6709\u300a\u767e\u5bb6\u8bb2\u575b\u300b\u4e3a\u8bc1\uff0c\u4eba\u7269\u5e74\u9f84\u7684bug\u6bd4\u7ea2\u697c\u68a6\u8fd8\u5c11\uff0c\u53d9\u8ff0\u4e8b\u4ef6\u67d0\u67d0\u5e74\u67d0\u67d0\u6708\u67d0\u67d0\u4f10\u67d0\u67d0\u4e0d\u538c\u5176\u70e6\uff0c\u53ef\u662f\u5207\u5408\u5386\u53f2\u7edd\u4e0d\u662f\u8bf4\u5b83\u662f\u597d\u5c0f\u8bf4\u7684\u7406\u7531\u554a\uff01\u6211\u89c9\u5f97\u739b\u4e3d\u82cf\u90fd\u4e0d\u662f\u81f4\u547d\u4f24\uff0c\u5173\u952e\u662f\u60c5\u8282\u5e7c\u7a1a\u770b\u4e0d\u4e0b\u53bb\u554a\uff01\u5bf9\u4e8eNER\u4efb\u52a1\uff0c\u540c\u6837\u4e3a\u4e00\u4e2atsv\u6587\u4ef6\uff0c\u7531\/t\u5206\u9694\u4e3a\u6587\u672c\u548clabel\u4e24\u5217\uff0c\u6587\u672c\u5b57\u4e0e\u5b57\u4e4b\u95f4\u7528\u7a7a\u683c\u9694\u5f00\u3002\u6837\u4f8b\u5982\u4e0b\uff1a\u730e \u8c79 \u79fb \u52a8 \u65b9 \u9762 \u89e3 \u91ca \u79f0 \uff0c \u79fb \u52a8 \u6536 \u5165 \u548c \u6d77 \u5916 \u6536 \u5165 \u7684 \u589e \u957f \u4e3b \u8981 \u5f97 \u76ca \u4e8e L i v e . m e \u4ea7 \u54c1 \u5728 \u6d77 \u5916 \u5e02 \u573a \u7684 \u5feb \u901f \u589e \u957f \u3002   B-ORG I-ORG I-ORG I-ORG O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O\u4e0b\u5217\u2f42\u4ef6\u5df2\u7ecf\u5b8c\u6210\u9884\u5904\u7406\uff0c\u53ef\u2f64\u4e8e\u6d4b\u8bd5\uff1ahttps:\/\/atp-modelzoo-sh.oss-cn-shanghai.aliyuncs.com\/release\/tutorials\/K-BERT\/kbert_data.zipK-BERT\u2f42\u672c\u5206\u7c7b\u793a\u4f8b\u5728\u6587\u672c\u5206\u7c7b\u4efb\u52a1\u4e2d\uff0c\u6211\u4eec\u91c7\u2f64\u4ee5\u4e0b\u547d\u4ee4\u5bf9K-BERT\u6a21\u578b\u8fdb\u2f8ffinetune\uff0c\u6a21\u578b\u5728finetune\u8fc7\u7a0b\u4e2d\u4e5f\u4f1a\u81ea\u52a8\u8f93\u51fa\u8bc4\u4f30\u7ed3\u679c\u3002\u5176\u4e2d\uff0c\u7528\u6237\u53ef\u4ee5\u5728\u524d\u8ff0\u9884\u5904\u7406\u793a\u4f8b\u6587\u4ef6\u4e2d\u627e\u5230\u8bad\u7ec3\u96c6chnsenticorp\/train.tsv\u3001\u8bc4\u6d4b\u96c6chnsenticorp\/dev.tsv\u4ee5\u53ca\u77e5\u8bc6\u56fe\u8c31\u6587\u4ef6kbert_kgs\/HowNet.spo\u3002kbert_cls.py\u4f4d\u4e8eEasyNLP\u9879\u76ee\u4e0b\u7684examples\/kbert\/\u4e0b\u3002\u76f8\u5173\u793a\u4f8b\u4ee3\u7801\u5982\u4e0b\uff1apython kbert_cls.py \/<\/code><\/pre>\n<p>&#8211;mode train \/<br \/>&#8211;tables tmp\/kbert_data\/chnsenticorp\/train.tsv,tmp\/kbert_data\/chnsenticorp\/dev.tsv \/<br \/>&#8211;input_schema label:str:1,sid1:str:1,sent1:str:1 \/<br \/>&#8211;first_sequence sent1 \/<br \/>&#8211;label_name label\/<br \/>&#8211;label_enumerate_values 0,1 \/<br \/>&#8211;checkpoint_dir .\/tmp\/kbert_classification_model\/ \/<br \/>&#8211;learning_rate 2e-5 \/<br \/>&#8211;epoch_num 2 \/<br \/>&#8211;random_seed 42 \/<br \/>&#8211;save_checkpoint_steps 50 \/<br \/>&#8211;sequence_length 128 \/<br \/>&#8211;micro_batch_size 16 \/<br \/>&#8211;app_name text_classify \/<br \/>&#8211;user_defined_parameters &#8220;pretrain_model_name_or_path=kbert-base-chinese kg_file=tmp\/kbert_data\/kbert_kgs\/HowNet.spo&#8221;K-BERT\u547d\u540d\u5b9e\u4f53\u8bc6\u522b\u793a\u4f8b\u5728NER\u4efb\u52a1\u4e2d\uff0c\u6211\u4eec\u91c7\u2f64\u4ee5\u4e0b\u547d\u4ee4\u5bf9K-BERT\u6a21\u578b\u8fdb\u2f8ffinetune\uff0c\u5176\u4f7f\u7528\u65b9\u5f0f\u4e0e\u6587\u672c\u5206\u7c7b\u76f8\u540c\uff1apython kbert_ner.py \/<br \/>&#8211;mode train \/<br \/>&#8211;tables tmp\/kbert_data\/financial_ner\/train.tsv,tmp\/kbert_data\/financial_ner\/dev.tsv \/<br \/>&#8211;input_schema content:str:1,label:str:1 \/<br \/>&#8211;first_sequence content \/<br \/>&#8211;label_name label\/<br \/>&#8211;label_enumerate_values B-ORG,B-PER,B-POS,I-ORG,I-PER,I-POS,O \/<br \/>&#8211;checkpoint_dir .\/tmp\/kbert_ner_model\/ \/<br \/>&#8211;learning_rate 2e-5 \/<br \/>&#8211;epoch_num 2 \/<br \/>&#8211;random_seed 42 \/<br \/>&#8211;save_checkpoint_steps 50 \/<br \/>&#8211;sequence_length 128 \/<br \/>&#8211;micro_batch_size 16 \/<br \/>&#8211;app_name sequence_labeling \/<br \/>&#8211;user_defined_parameters &#8220;pretrain_model_name_or_path=kbert-base-chinese kg_file=tmp\/kbert_data\/kbert_kgs\/HowNet.spo&#8221;\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\uff1a<a href=\"https:\/\/link.segmentfault.com\/?enc=qbKa4nblif4%2B99mGnz%2F8sg%3D%3D.85EHlZWrAR2BrK9BQzZFZLBJ8vc%2Bn6FSP7ojxHAdigDlrduwR13VoPfCwM6ykPxLdsGN94u6nYaYHTuHngoXow%3D%3D\" rel=\"nofollow\">https:\/\/github.com\/alibaba\/Ea&#8230;<\/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. arXivWeijie Liu, Peng Zhou, Zhe Zhao, Zhiruo Wang, Qi Ju, Haotang Deng, and Ping Wang. 2020. K-BERT: Enabling Language Representation with Knowledge Graph. In AAAI. 2901\u20132908K-BERT\u539f\u4f5c\u8005\u5f00\u6e90\u4ee3\u7801\uff1a<a href=\"https:\/\/link.segmentfault.com\/?enc=r4is7pigne%2B2UScGTyWs1g%3D%3D.51%2BCcpMsITlTflJMBj74gh9JCxZBZ2BalfiE%2FmCT%2BnxUqFUuvGDIvD1Yc%2Bpbj4Fz\" rel=\"nofollow\">https:\/\/github.com\/autoliuwei&#8230;<\/a>\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\u539f\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.segmentfault.com\/?enc=X8qy0dHUDBlv5tTSckEHgw%3D%3D.5HQ1Kkze4VoZhOlNJUvrZmnRutLqmugOSEZoJ5F1%2FZKT4uq%2BwVwuqGitTy5%2BCd56\" rel=\"nofollow\">https:\/\/click.aliyun.com\/m\/10&#8230;<\/a>\u672c\u6587\u4e3a\u963f\u91cc\u4e91\u539f\u521b\u5185\u5bb9\uff0c\u672a\u7ecf\u5141\u8bb8\u4e0d\u5f97\u8f6c\u8f7d\u3002<\/p>\n<p><a href=\"https:\/\/segmentfault.com\/a\/1190000042333254\">https:\/\/segmentfault.com\/a\/1190000042333254<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5bfc\u8bfb\u77e5\u8bc6\u56fe\u8c31\uff08Knowledge Graph\uff09\u7684\u6982\u5ff5\u2fb8\u6b21\u51fa\u73b02012\u5e74\uff0c\u7531Google\u63d0\u51fa\uff0c\u5b83\u4f5c\u4e3a\u2f00\u79cd\u2f24\u89c4\u6a21\u8bed [&hellip;]<\/p>\n","protected":false},"author":0,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[],"tags":[],"class_list":["post-10006236","post","type-post","status-publish","format-standard","hentry"],"jetpack_sharing_enabled":true,"jetpack_featured_media_url":"","_links":{"self":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts\/10006236","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"}],"replies":[{"embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=10006236"}],"version-history":[{"count":0,"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts\/10006236\/revisions"}],"wp:attachment":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=10006236"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=10006236"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=10006236"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}