{"id":670,"date":"2021-03-25T15:27:08","date_gmt":"2021-03-25T15:27:08","guid":{"rendered":"http:\/\/falbang.com\/?p=670"},"modified":"2021-06-04T10:23:44","modified_gmt":"2021-06-04T02:23:44","slug":"2001-2021%ef%bc%8c%e4%b8%80%e6%96%87%e8%af%bb%e6%87%82nlp%e5%8f%91%e5%b1%95%e7%ae%80%e5%8f%b2","status":"publish","type":"post","link":"https:\/\/falbang.com\/?p=670","title":{"rendered":"2001-2021\uff0c\u4e00\u6587\u8bfb\u61c2NLP\u53d1\u5c55\u7b80\u53f2"},"content":{"rendered":"\n

\u81ea\u7136\u8bed\u8a00<\/strong>\u4e3b\u8981\u6307\u4eba\u4eec\u4ea4\u6d41\u7528\u7684\u6b63\u5f0f\u6216\u975e\u6b63\u5f0f\u8bed\u8a00\uff0c\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff08NLP\uff09\u5219\u662f\u5c06\u4eba\u7c7b\u8bed\u8a00\u8f6c\u6362\u6210\u8ba1\u7b97\u673a\u80fd\u7406\u89e3\u7684\u7b26\u53f7\u6216\u5c06\u673a\u5668\u8bed\u8a00\u8f6c\u6362\u6210\u4eba\u53ef\u4ee5\u7406\u89e3\u7684\u8bed\u8a00\u3002<\/p>\n\n\n\n

\u81ea\u7136\u8bed\u8a00\u5904\u7406\u662f\u4eba\u5de5\u667a\u80fd\u9886\u57df\u7684\u4e00\u4e2a\u91cd\u8981\u7814\u7a76\u65b9\u5411\uff0c\u662f\u8ba1\u7b97\u673a\u79d1\u5b66\u4e0e\u8bed\u8a00\u5b66\u7684\u4ea4\u53c9\u5b66\u79d1\u3002<\/p>\n\n\n\n

\u81ea\u7136\u8bed\u8a00\u5904\u7406\u67092\u5927\u6838\u5fc3\u4efb\u52a1\uff0c\u81ea\u7136\u8bed\u8a00\u7406\u89e3\uff08NLU\uff09\u548c\u81ea\u7136\u8bed\u8a00\u751f\u6210\uff08NLG\uff09\u3002\u5bf9\u4eba\u6765\u8bf4\uff0c\u7406\u89e3\u8bed\u8a00\u662f\u4e00\u4ef6\u5f88\u81ea\u7136\u7684\u4e8b\u60c5\uff0c\u4f46\u5bf9\u673a\u5668\u6765\u8bf4\u5374\u662f\u5f88\u56f0\u96be\u7684\u4e8b\u60c5\u3002\u6240\u6709\u8bed\u8a00\u7684\u9c81\u68d2\u6027\u90fd\u662f\u5bfc\u81f4\u81ea\u7136\u8bed\u8a00\u7406\u89e3\u7684\u4e3b\u8981\u96be\u70b9\uff0c\u5176\u4e2d\u5305\u62ec\uff1a\u8bed\u8a00\u7684\u591a\u6837\u6027\u3001\u6b67\u4e49\u6027\u3001\u77e5\u8bc6\u4f9d\u8d56\u3001\u4e0a\u4e0b\u6587\u5173\u7cfb\u7b49\u3002<\/strong><\/p>\n\n\n\n

2001-2021\uff0c\u4e00\u6587\u8bfb\u61c2NLP\u53d1\u5c55\u7b80\u53f2\u81ea\u7136\u8bed\u8a00\u751f\u6210\u662f\u5c06\u673a\u5668\u7406\u89e3\u5230\u7684\u4fe1\u606f\uff0c\u4e00\u822c\u662f\u975e\u6587\u5b57\u7684\u6570\u636e\u5185\u5bb9\uff0c\u8f6c\u5316\u5e76\u8868\u8fbe\u6210\u8bed\u8a00\u4f20\u9012\u7ed9\u4eba\u7684\u8fc7\u7a0b\u3002\u8fd9\u4e2a\u8fc7\u7a0b\u9762\u4e34\u7684\u56f0\u96be\u5219\u662f\u751f\u6210\u8bed\u53e5\u7684\u8bed\u6cd5\u7ed3\u6784\u3001\u8bed\u4e49\u8868\u8fbe\u662f\u5426\u51c6\u786e\uff0c\u4fe1\u606f\u662f\u5426\u91cd\u590d\u7b49\u3002<\/p>\n\n\n\n

\u4e3a\u4e86\u89e3\u51b3\u4e0a\u8ff0\u95ee\u9898\uff0c\u4e00\u4e9b\u57fa\u672c\u7684\u81ea\u7136\u8bed\u8a00\u5904\u7406\u65b9\u5411\u4fbf\u5e94\u8fd0\u800c\u751f\uff0c\u5305\u62ec\uff1a\u5206\u8bcd\u3001\u8bcd\u6027\u6807\u6ce8\u3001\u8bcd\u5f62\u8fd8\u539f\u3001\u4f9d\u8d56\u5173\u7cfb\u89e3\u6790\u3001\u547d\u540d\u5b9e\u4f53\u8bc6\u522b\u3001\u5e8f\u5217\u6807\u6ce8\u3001\u53e5\u5b50\u5173\u7cfb\u8bc6\u522b\u7b49\u3002<\/p>\n\n\n\n

\u9664\u6b64\u4e4b\u5916\uff0c\u81ea\u7136\u8bed\u8a00\u5904\u7406\u8fd8\u5305\u62ec\u4e86\u5f88\u591a\u5177\u4f53\u5e94\u7528\uff0c\u4f8b\u5982\uff1a\u4fe1\u606f\u68c0\u7d22\u3001\u4fe1\u606f\u62bd\u53d6\u3001\u6587\u672c\u5206\u7c7b\u4e0e\u805a\u7c7b\u3001\u673a\u5668\u7ffb\u8bd1\u3001\u6458\u8981\u751f\u6210\u3001\u804a\u5929\u673a\u5668\u4eba\u7b49\u7b49\u3002\u5b83\u6d89\u53ca\u4e0e\u8bed\u8a00\u5904\u7406\u76f8\u5173\u7684\u6570\u636e\u6316\u6398\u3001\u673a\u5668\u5b66\u4e60\u3001\u8bed\u8a00\u5b66\u7814\u7a76\uff0c\u4ee5\u53ca\u8fd1\u5e74\u6765\u975e\u5e38\u6d41\u884c\u7684\u6df1\u5ea6\u5b66\u4e60\u7b49\u3002<\/p>\n\n\n\n

\u4ee5\u524d\u4eba\u4eec\u901a\u5e38\u901a\u8fc7\u7ecf\u9a8c\u4e3b\u4e49\uff0c\u5373\u7279\u5f81\u5de5\u7a0b\u7684\u65b9\u5f0f\uff0c\u6216\u7edf\u8ba1\u4fe1\u606f\u7684\u65b9\u5f0f\u6765\u89e3\u51b3NLP\u95ee\u9898\uff0c\u6587\u672c\u7684\u8868\u5f81\u4e5f\u53ea\u662f\u7b80\u5355\u7684\u8bcd\u888b\u8868\u793a\uff08Bag-of-Words, see Figure 1\uff09\uff0c\u8fd9\u79cd\u65b9\u5f0f\u4f7f\u5f97\u6587\u672c\u4e22\u5931\u4e86\u5e8f\u5217\u4fe1\u606f\u4ee5\u53ca\u80cc\u666f\u4f9d\u8d56\u4fe1\u606f\u7b49\u3002<\/p>\n\n\n\n

\u4ece2001\u5e74\u7684\u5d4c\u5165\u5f0f\u8bcd\u5411\u91cf\u8868\u793a\u52302013\u5e74\u7684word2vec (see Figure 2)\uff0c\u7814\u7a76\u5df2\u7ecf\u9010\u6b65\u89e3\u51b3\u4f20\u7edf\u8bcd\u888b\u8868\u5f81\u65b9\u5f0f\u7684\u65e0\u5e8f\u6027\uff0c\u5e76\u8fdb\u4e00\u6b65\u4e30\u5bcc\u4e86\u8868\u5f81\u5411\u91cf\u7684\u8868\u8fbe\u80fd\u529b\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

                                                     Figure 1: Bag-of-Words Example<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Figure 2: \u5d4c\u5165\u5f0f\u8bcd\u5411\u91cf\u8868\u793a \uff08word embedding\uff09<\/p>\n\n\n\n

2013\u52302014\u5e74\uff0c\u5404\u79cd\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u5728NLP\u4e0a\u7684\u5e94\u7528\u9010\u6e10\u589e\u591a\uff0c\u5176\u4e2d\u5e94\u7528\u6700\u5e7f\u6cdb\u7684\u662f\uff1a\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff08convolutional neural networks\uff0c CNN\uff09\u3001\u5faa\u73af\u795e\u7ecf\u7f51\u7edc\uff08recurrent neural networks\uff0cRNN\uff09\u548c\u7ed3\u6784\u9012\u5f52\u795e\u7ecf\u7f51\u7edc\uff08recursive neural networks\uff09\u3002<\/p>\n\n\n\n

\u56e0\u4e3a\u6587\u672c\u662f\u6709\u5e8f\u7684\uff0c\u800cRNN (see Figure 3)\u7684\u5e8f\u5217\u6027\u7ed3\u6784\u4f7f\u5176\u6210\u4e3a\u4e86\u6700\u8d34\u5408\u6587\u672c\u8f93\u5165\u7684\u7f51\u7edc\u6a21\u578b\uff0c\u4e5f\u662f\u5927\u91cf\u7814\u7a76\u5de5\u4f5c\u7684\u9996\u9009\u57fa\u7840\u6846\u67b6\u3002\u4e3a\u4e86\u89e3\u51b3RNN\u6a21\u578b\u68af\u5ea6\u6d88\u5931\u548c\u7206\u70b8\u7684\u95ee\u9898\uff0c\u7ecf\u5178\u7684\u957f\u77ed\u671f\u8bb0\u5fc6\u7f51\u7edc\uff08long-short term memory networks\uff0cLSTM\uff09\u4ee5\u53ca\u5176\u6539\u8fdb\u7248\u672c\u95e8\u63a7\u5faa\u73af\u5355\u5143\uff08gated recurrent unit\uff0c GRU\uff09\u5f00\u59cb\u88ab\u5e7f\u6cdb\u5e94\u7528\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Figure 3: Recurrent neural network, RNN<\/p>\n\n\n\n

2014 \u5e74Sutskever \u7b49\u4eba\u63d0\u51fa\u4e86\u5e8f\u5217\u5230\u5e8f\u5217\u5b66\u4e60\uff08sequence-to-sequence see Figure 4\uff09\uff0c\u5373\u662f\u7528\u795e\u7ecf\u7f51\u7edc\u5c06\u4e00\u4e2a\u5e8f\u5217\u6620\u5c04\u5230\u53e6\u4e00\u4e2a\u5e8f\u5217\u7684\u6846\u67b6\u6a21\u5f0f\u3002\u7f16\u7801\u5668\u5c06\u8f93\u5165\u6587\u672c\u8f6c\u6362\u4e3a\u8868\u793a\u5411\u91cf\uff0c\u89e3\u7801\u5668\u5c06\u5411\u91cf\u8f6c\u5316\u4e3a\u8f93\u51fa\u6587\u672c\u5e8f\u5217\u3002<\/p>\n\n\n\n

\u8fd9\u4e2a\u7ed3\u6784\u4e00\u7ecf\u63d0\u51fa\u4fbf\u8fc5\u901f\u88ab\u5e94\u7528\u5230\u673a\u5668\u7ffb\u8bd1\uff0c\u5e76\u9010\u6b65\u53d6\u4ee3\u4e86\u57fa\u4e8e\u77ed\u8bed\u7684\u6574\u53e5\u673a\u5668\u7ffb\u8bd1\u6a21\u578b\u3002<\/p>\n\n\n\n

\u4f46\u7f16\u7801\u5668\u9700\u8981\u5c06\u8f93\u5165\u5e8f\u5217\u6574\u5408\u6210\u56fa\u5b9a\u5927\u5c0f\u7684\u5411\u91cf\uff0c\u4f7f\u5f97\u5f88\u591a\u8f93\u5165\u4fe1\u606f\u88ab\u5ffd\u7565\u751a\u81f3\u4e22\u5931\uff0c\u5bfc\u81f4\u89e3\u7801\u6548\u679c\u4e0d\u4f73\uff0c\u56e0\u6b64\u6ce8\u610f\u529b\u673a\u5236\u88ab\u63d0\u51fa\uff08attention model\uff09\u3002<\/p>\n\n\n\n

\u6ce8\u610f\u529b\u673a\u5236\u4f7f\u5f97\u89e3\u7801\u5668\u6709\u9488\u5bf9\u6027\u7684\u5bf9\u7f16\u7801\u5e8f\u5217\u8fdb\u884c\u89e3\u8bfb\uff0c\u5728\u8f93\u5165\u5e8f\u5217\u4e0a\u5f15\u5165\u6ce8\u610f\u6743\u91cd\uff0c\u4ee5\u4f18\u5148\u8003\u8651\u5b58\u5728\u76f8\u5173\u4fe1\u606f\u7684\u8f93\u5165\u4f4d\u7f6e\u4fe1\u606f\uff0c\u4ee5\u751f\u6210\u4e0b\u4e00\u4e2a\u8f93\u51fa\u3002\u6ce8\u610f\u529b\u673a\u5236\u88ab\u5e7f\u6cdb\u5e94\u7528\u5230\u673a\u5668\u7ffb\u8bd1\u3001\u9605\u8bfb\u7406\u89e3\u7b49\u4efb\u52a1\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u00a0Figure 4: Sequence-to-sequence model<\/p>\n\n\n\n

\u53d7\u56fe\u50cf\u5b66\u4e60\u7684\u5f71\u54cd\uff0c\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u57282015 \u5e74\u88ab\u9996\u6b21\u63d0\u51fa\uff0c\u4f46\u57282018\u5e74\u624d\u53d1\u751f\u4e86\u91cd\u5927\u8fdb\u5c55\uff0c\u4ee5Google\u63d0\u51fa\u7684BERT (see Figure 5)\u4e3a\u4ee3\u8868\u7684\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\uff0c\u88ab\u8bc1\u5b9e\u5728\u5927\u91cfNLP\u4efb\u52a1\u4e0a\u90fd\u5f88\u6709\u6548\uff0c\u4f8b\u5982ELMO\uff0cXLNet\uff0cGPT\uff0cRoBERTa, ALBERT\u7b49\u3002<\/p>\n\n\n\n

\u8fd9\u4e9b\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u5747\u662f\u65e0\u76d1\u7763\u5f62\u5f0f\uff0c\u4f7f\u5f97\u6a21\u578b\u80fd\u591f\u4ece\u65e0\u6807\u6ce8\u8bed\u6599\u4e2d\u83b7\u5f97\u901a\u7528\u7684\u8bed\u8a00\u5efa\u6a21\u80fd\u529b\u3002\u800c\u540e\u6a21\u578b\u53ef\u4ee5\u6839\u636e\u5c11\u91cf\u7684\u6807\u6ce8\u6570\u636e\u8fdb\u884c\u4efb\u52a1\u5bfc\u5411\u7684\u5fae\u8c03\uff0c\u4f7f\u5f97\u6a21\u578b\u4e0d\u9700\u8981\u5927\u91cf\u7684\u8bad\u7ec3\u6570\u636e\u4e5f\u80fd\u591f\u8fbe\u5230\u4e0d\u9519\u7684\u6548\u679c\uff0c\u53ef\u5f88\u5927\u7a0b\u5ea6\u7f13\u89e3\u4f4e\u8d44\u6e90\u4efb\u52a1\u5bf9\u5927\u91cf\u6807\u6ce8\u6570\u636e\u7684\u9700\u6c42\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Figure 5: BERT<\/p>\n\n\n\n

\u00a0\u867d\u7136\u8fd1\u5e74\u6765NLP\u9886\u57df\u8fc5\u901f\u7684\u53d1\u5c55\uff0c\u4f46\u5f88\u591a\u4efb\u52a1\u7684\u8868\u73b0\u8fd8\u8fdc\u8fbe\u4e0d\u5230\u4eba\u4eec\u7684\u671f\u671b\uff0c\u5176\u7814\u7a76\u4ecd\u00a0 \u00a0\u4efb\u91cd\u800c\u9053\u8fdc\u3002<\/p>\n\n\n\n

\u66f4\u591a\u6570\u636e\u79d1\u5b66\u5e72\u8d27\uff0c\u5173\u6ce8\u516c\u53f7\u3010\u6570\u636e\u79d1\u5b66\u5e94\u7528\u7814\u9662\u3011\uff0c\u56de\u590d\u3010\u5b98\u7f51\u3011\uff0c\u8fd8\u80fd\u83b7\u5f97\u6570\u636e\u79d1\u5b66\u5927\u793c\u5305\u4e00\u4efd\u54e6\uff01<\/p>\n\n\n\n

\"\"<\/figure>\n","protected":false},"excerpt":{"rendered":"

\u81ea\u7136\u8bed\u8a00\u4e3b\u8981\u6307\u4eba\u4eec\u4ea4\u6d41\u7528\u7684\u6b63\u5f0f\u6216\u975e\u6b63\u5f0f\u8bed\u8a00\uff0c\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff08NLP\uff09\u5219\u662f\u5c06\u4eba\u7c7b\u8bed\u8a00\u8f6c\u6362\u6210\u8ba1\u7b97\u673a\u80fd\u7406\u89e3\u7684\u7b26\u53f7\u6216\u5c06\u673a … <\/p>\n

Continue reading “2001-2021\uff0c\u4e00\u6587\u8bfb\u61c2NLP\u53d1\u5c55\u7b80\u53f2”<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[33],"tags":[],"_links":{"self":[{"href":"https:\/\/falbang.com\/index.php?rest_route=\/wp\/v2\/posts\/670"}],"collection":[{"href":"https:\/\/falbang.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/falbang.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/falbang.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/falbang.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=670"}],"version-history":[{"count":2,"href":"https:\/\/falbang.com\/index.php?rest_route=\/wp\/v2\/posts\/670\/revisions"}],"predecessor-version":[{"id":923,"href":"https:\/\/falbang.com\/index.php?rest_route=\/wp\/v2\/posts\/670\/revisions\/923"}],"wp:attachment":[{"href":"https:\/\/falbang.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=670"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/falbang.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=670"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/falbang.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=670"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}