{"id":5762,"date":"2022-08-12T16:08:51","date_gmt":"2022-08-12T08:08:51","guid":{"rendered":"http:\/\/139.9.1.231\/?p=5762"},"modified":"2022-08-12T16:09:04","modified_gmt":"2022-08-12T08:09:04","slug":"natural-language-processing-nlp","status":"publish","type":"post","link":"http:\/\/139.9.1.231\/index.php\/2022\/08\/12\/natural-language-processing-nlp\/","title":{"rendered":"Transformer\u7efc\u8ff0&#8212;(Natural Language Processing, NLP)"},"content":{"rendered":"\n<ul class=\"has-light-pink-background-color has-background\"><li><strong>github\u5b66\u4e60\u5730\u5740\uff1ahttps:\/\/github.com\/datawhalechina\/learn-nlp-with-transformers<\/strong><\/li><\/ul>\n\n\n\n<p>\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff08Natural Language Processing, NLP\uff09\u662f\u4e00\u79cd\u91cd\u8981\u7684\u4eba\u5de5\u667a\u80fd\uff08Artificial Intelligence, AI\uff09\u6280\u672f\u3002\u6211\u4eec\u968f\u5904\u53ef\u4ee5\u89c1\u5230NLP\u6280\u672f\u7684\u5e94\u7528\uff0c\u6bd4\u5982\u7f51\u7edc\u641c\u7d22\uff0c\u5e7f\u544a\uff0c\u7535\u5b50\u90ae\u4ef6\uff0c\u667a\u80fd\u5ba2\u670d\uff0c\u673a\u5668\u7ffb\u8bd1\uff0c\u667a\u80fd\u65b0\u95fb\u64ad\u62a5\u7b49\u7b49\u3002\u6700\u8fd1\u51e0\u5e74\uff0c\u57fa\u4e8e\u6df1\u5ea6\u5b66\u4e60\uff08Deep Learning, DL\uff09\u7684NLP\u6280\u672f\u5728\u5404\u9879\u4efb\u52a1\u4e2d\u53d6\u5f97\u4e86\u5f88\u597d\u7684\u6548\u679c\uff0c\u8fd9\u4e9b\u57fa\u4e8e\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u7684NLP\u4efb\u52a1\u89e3\u51b3\u65b9\u6848\u901a\u5e38\u4e0d\u4f7f\u7528\u4f20\u7edf\u7684\u3001\u7279\u5b9a\u4efb\u52a1\u7684\u7279\u5f81\u5de5\u7a0b\u800c\u662f\u4ec5\u4ec5\u4f7f\u7528\u4e00\u4e2a\u7aef\u5230\u7aef\uff08end-to-end\uff09\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u5c31\u53ef\u4ee5\u83b7\u5f97\u5f88\u597d\u7684\u6548\u679c\u3002\u672c\u6559\u7a0b\u5c06\u4f1a\u57fa\u4e8e\u6700\u524d\u6cbf\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u7ed3\u6784\uff08transformers\uff09\u6765\u89e3\u51b3NLP\u91cc\u7684\u51e0\u4e2a\u7ecf\u5178\u4efb\u52a1\u3002\u901a\u8fc7\u672c\u6559\u7a0b\u7684\u5b66\u4e60\uff0c\u6211\u4eec\u5c06\u80fd\u591f\u4e86\u89e3transformer\u76f8\u5173\u539f\u7406\u3001\u719f\u7ec3\u4f7f\u7528transformer\u76f8\u5173\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u6765\u89e3\u51b3NLP\u91cc\u7684\u5b9e\u9645\u95ee\u9898\u4ee5\u53ca\u5728\u5404\u7c7b\u4efb\u52a1\u4e0a\u53d6\u5f97\u5f88\u597d\u7684\u6548\u679c\u3002<\/p>\n\n\n\n<p>\u81ea\u7136\u8bed\u8a00\u4e0e\u6df1\u5ea6\u5b66\u4e60\u7684\u8bfe\u7a0b\u63a8\u8350\uff1a<a href=\"http:\/\/web.stanford.edu\/class\/cs224n\/\">CS224n: Natural Language Processing with Deep Learning<\/a>\u00a0\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7684\u4e66\u7c4d\u63a8\u8350\uff1a<a href=\"https:\/\/web.stanford.edu\/~jurafsky\/slp3\/\">Speech and Language Processing<\/a><\/p>\n\n\n\n<h2 id=\"%E5%B8%B8%E8%A7%81%E7%9A%84nlp%E4%BB%BB%E5%8A%A1\">\u5e38\u89c1\u7684NLP\u4efb\u52a1<\/h2>\n\n\n\n<p>\u672c\u6559\u7a0b\u5c06NLP\u4efb\u52a1\u5212\u5206\u4e3a4\u4e2a\u5927\u7c7b\uff1a1\u3001\u6587\u672c\u5206\u7c7b\uff0c 2\u3001\u5e8f\u5217\u6807\u6ce8\uff0c3\u3001\u95ee\u7b54\u4efb\u52a1\u2014\u2014\u62bd\u53d6\u5f0f\u95ee\u7b54\u548c\u591a\u9009\u95ee\u7b54\uff0c4\u3001\u751f\u6210\u4efb\u52a1\u2014\u2014\u8bed\u8a00\u6a21\u578b\u3001\u673a\u5668\u7ffb\u8bd1\u548c\u6458\u8981\u751f\u6210\u3002<\/p>\n\n\n\n<ul><li>\u6587\u672c\u5206\u7c7b\uff1a\u5bf9\u5355\u4e2a\u3001\u4e24\u4e2a\u6216\u8005\u591a\u6bb5\u6587\u672c\u8fdb\u884c\u5206\u7c7b\u3002\u4e3e\u4f8b\uff1a\u201c\u8fd9\u4e2a\u6559\u7a0b\u771f\u68d2\uff01\u201d\u8fd9\u6bb5\u6587\u672c\u7684\u60c5\u611f\u503e\u5411\u662f\u6b63\u5411\u7684\uff0c\u201c\u6211\u5728\u5b66\u4e60transformer\u201d\u548c\u201c\u5982\u4f55\u5b66\u4e60transformer\u201d\u8fd9\u4e24\u6bb5\u6587\u672c\u662f\u76f8\u4f3c\u7684\u3002<\/li><li>\u5e8f\u5217\u6807\u6ce8\uff1a\u5bf9\u6587\u672c\u5e8f\u5217\u4e2d\u7684token\u3001\u5b57\u6216\u8005\u8bcd\u8fdb\u884c\u5206\u7c7b\u3002\u4e3e\u4f8b\uff1a\u201c\u6211\u5728<strong>\u56fd\u5bb6\u56fe\u4e66\u9986<\/strong>\u5b66transformer\u3002\u201d\u8fd9\u6bb5\u6587\u672c\u4e2d\u7684<strong>\u56fd\u5bb6\u56fe\u4e66\u9986<\/strong>\u662f\u4e00\u4e2a<strong>\u5730\u70b9<\/strong>\uff0c\u53ef\u4ee5\u88ab\u6807\u6ce8\u51fa\u6765\u65b9\u4fbf\u673a\u5668\u5bf9\u6587\u672c\u7684\u7406\u89e3\u3002<\/li><li>\u95ee\u7b54\u4efb\u52a1\u2014\u2014\u62bd\u53d6\u5f0f\u95ee\u7b54\u548c\u591a\u9009\u95ee\u7b54\uff1a1\u3001\u62bd\u53d6\u5f0f\u95ee\u7b54\u6839\u636e<strong>\u95ee\u9898<\/strong>\u4ece\u4e00\u6bb5\u7ed9\u5b9a\u7684\u6587\u672c\u4e2d\u627e\u5230<strong>\u7b54\u6848<\/strong>\uff0c\u7b54\u6848\u5fc5\u987b\u662f\u7ed9\u5b9a\u6587\u672c\u7684\u4e00\u5c0f\u6bb5\u6587\u5b57\u3002\u4e3e\u4f8b\uff1a\u95ee\u9898\u201c\u5c0f\u5b66\u8981\u8bfb\u591a\u4e45?\u201d\u548c\u4e00\u6bb5\u6587\u672c\u201c\u5c0f\u5b66\u6559\u80b2\u4e00\u822c\u662f\u516d\u5e74\u5236\u3002\u201d\uff0c\u5219\u7b54\u6848\u662f\u201c\u516d\u5e74\u201d\u30022\u3001\u591a\u9009\u5f0f\u95ee\u7b54\uff0c\u4ece\u591a\u4e2a\u9009\u9879\u4e2d\u9009\u51fa\u4e00\u4e2a\u6b63\u786e\u7b54\u6848\u3002\u4e3e\u4f8b\uff1a\u201c\u4ee5\u4e0b\u54ea\u4e2a\u6a21\u578b\u7ed3\u6784\u5728\u95ee\u7b54\u4e2d\u6548\u679c\u6700\u597d\uff1f\u201c\u548c4\u4e2a\u9009\u9879\u201dA\u3001MLP\uff0cB\u3001cnn\uff0cC\u3001lstm\uff0cD\u3001transformer\u201c\uff0c\u5219\u7b54\u6848\u9009\u9879\u662fD\u3002<\/li><li>\u751f\u6210\u4efb\u52a1\u2014\u2014\u8bed\u8a00\u6a21\u578b\u3001\u673a\u5668\u7ffb\u8bd1\u548c\u6458\u8981\u751f\u6210\uff1a\u6839\u636e\u5df2\u6709\u7684\u4e00\u6bb5\u6587\u5b57\u751f\u6210\uff08generate\uff09\u4e00\u4e2a\u5b57\u901a\u5e38\u53eb\u505a\u8bed\u8a00\u6a21\u578b\uff0c\u6839\u636e\u4e00\u5927\u6bb5\u6587\u5b57\u751f\u6210\u4e00\u5c0f\u6bb5\u603b\u7ed3\u6027\u6587\u5b57\u901a\u5e38\u53eb\u505a\u6458\u8981\u751f\u6210\uff0c\u5c06\u6e90\u8bed\u8a00\u6bd4\u5982\u4e2d\u6587\u53e5\u5b50\u7ffb\u8bd1\u6210\u76ee\u6807\u8bed\u8a00\u6bd4\u5982\u82f1\u8bed\u901a\u5e38\u53eb\u505a\u673a\u5668\u7ffb\u8bd1\u3002<\/li><\/ul>\n\n\n\n<p>\u867d\u7136\u5404\u79cd\u57fa\u4e8etransformer\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u5df2\u7ecf\u5728\u591a\u4e2a\u4eba\u5de5\u6784\u5efa\u7684NLP\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u8272\uff0c\u4f46\u7531\u4e8e\u4eba\u7c7b\u8bed\u8a00\u535a\u5927\u7cbe\u6df1\uff0c\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u4f9d\u7136\u6709\u5f88\u957f\u7684\u8def\u8981\u8d70\u3002<\/p>\n\n\n\n<h2 class=\"has-light-pink-background-color has-background\" id=\"transformer%E7%9A%84%E5%85%B4%E8%B5%B7\">Transformer\u7684\u5174\u8d77<\/h2>\n\n\n\n<p class=\"has-light-pink-background-color has-background\">2017\u5e74\uff0c<a href=\"https:\/\/arxiv.org\/pdf\/1706.03762.pdf\">Attention Is All You Need<\/a>\u8bba\u6587\u9996\u6b21\u63d0\u51fa\u4e86<strong>Transformer<\/strong>\u6a21\u578b\u7ed3\u6784\u5e76\u5728\u673a\u5668\u7ffb\u8bd1\u4efb\u52a1\u4e0a\u53d6\u5f97\u4e86The State of the Art(SOTA, \u6700\u597d)\u7684\u6548\u679c\u30022018\u5e74\uff0c<a href=\"https:\/\/arxiv.org\/pdf\/1810.04805.pdf\">BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding<\/a>\u4f7f\u7528Transformer\u6a21\u578b\u7ed3\u6784\u8fdb\u884c\u5927\u89c4\u6a21\u8bed\u8a00\u6a21\u578b\uff08language model\uff09\u9884\u8bad\u7ec3\uff08Pre-train\uff09\uff0c\u518d\u5728\u591a\u4e2aNLP\u4e0b\u6e38\uff08downstream\uff09\u4efb\u52a1\u4e2d\u8fdb\u884c\u5fae\u8c03\uff08Finetune\uff09\uff0c\u4e00\u4e3e\u5237\u65b0\u4e86\u5404\u5927NLP\u4efb\u52a1\u7684\u699c\u5355\u6700\u9ad8\u5206\uff0c\u8f70\u52a8\u4e00\u65f6\u30022019\u5e74-2021\u5e74\uff0c\u7814\u7a76\u4eba\u5458\u5c06Transformer\u8fd9\u79cd\u6a21\u578b\u7ed3\u6784\u548c\u9884\u8bad\u7ec3+\u5fae\u8c03\u8fd9\u79cd\u8bad\u7ec3\u65b9\u5f0f\u76f8\u7ed3\u5408\uff0c\u63d0\u51fa\u4e86\u4e00\u7cfb\u5217Transformer\u6a21\u578b\u7ed3\u6784\u3001\u8bad\u7ec3\u65b9\u5f0f\u7684\u6539\u8fdb\uff08\u6bd4\u5982transformer-xl\uff0cXLnet\uff0cRoberta\u7b49\u7b49\uff09\u3002\u5982\u4e0b\u56fe\u6240\u793a\uff0c\u5404\u7c7bTransformer\u7684\u6539\u8fdb\u4e0d\u65ad\u6d8c\u73b0\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" src=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2022\/08\/1-x-formers-890x1024.png\" alt=\"\" class=\"wp-image-5764\" width=\"752\" height=\"865\" srcset=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2022\/08\/1-x-formers-890x1024.png 890w, http:\/\/139.9.1.231\/wp-content\/uploads\/2022\/08\/1-x-formers-261x300.png 261w, http:\/\/139.9.1.231\/wp-content\/uploads\/2022\/08\/1-x-formers-768x884.png 768w, http:\/\/139.9.1.231\/wp-content\/uploads\/2022\/08\/1-x-formers.png 1244w\" sizes=\"(max-width: 752px) 100vw, 752px\" \/><figcaption>\u56fe\uff1a\u5404\u7c7bTransformer\u6539\u8fdb\uff0c\u6765\u6e90\uff1a<a href=\"https:\/\/arxiv.org\/pdf\/2106.04554.pdf\">A Survey of Transformers<\/a><\/figcaption><\/figure>\n\n\n\n<p>\u53e6\u5916\uff0c\u7531\u4e8eTransformer\u4f18\u5f02\u7684\u6a21\u578b\u7ed3\u6784\uff0c\u4f7f\u5f97\u5176\u53c2\u6570\u91cf\u53ef\u4ee5\u975e\u5e38\u5e9e\u5927\u4ece\u800c\u5bb9\u7eb3\u66f4\u591a\u7684\u4fe1\u606f\uff0c\u56e0\u6b64Transformer\u6a21\u578b\u7684\u80fd\u529b\u968f\u7740\u9884\u8bad\u7ec3\u4e0d\u65ad\u63d0\u5347\uff0c\u968f\u7740\u8fd1\u51e0\u5e74\u8ba1\u7b97\u80fd\u529b\u7684\u63d0\u5347\uff0c\u8d8a\u6765\u8d8a\u5927\u7684\u9884\u8bad\u7ec3\u6a21\u578b\u4ee5\u53ca\u6548\u679c\u8d8a\u6765\u8d8a\u597d\u7684Transformers\u4e0d\u65ad\u6d8c\u73b0\uff0c\u7b80\u5355\u7684\u7edf\u8ba1\u53ef\u4ee5\u4ece\u4e0b\u56fe\u770b\u51fa\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" width=\"1024\" height=\"606\" src=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2022\/08\/image-193-1024x606.png\" alt=\"\" class=\"wp-image-5765\" srcset=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2022\/08\/image-193-1024x606.png 1024w, http:\/\/139.9.1.231\/wp-content\/uploads\/2022\/08\/image-193-300x177.png 300w, http:\/\/139.9.1.231\/wp-content\/uploads\/2022\/08\/image-193-768x454.png 768w, http:\/\/139.9.1.231\/wp-content\/uploads\/2022\/08\/image-193.png 1505w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><figcaption>\u56fe\uff1a\u9884\u8bad\u7ec3\u6a21\u578b\u53c2\u6570\u4e0d\u65ad\u53d8\u5927,\u6765\u6e90<a href=\"https:\/\/huggingface.co\/course\/chapter1\/4?fw=pt\">Huggingface<\/a><\/figcaption><\/figure>\n\n\n\n<p>\u5c3d\u7ba1\u5404\u7c7bTransformer\u7684\u7814\u7a76\u975e\u5e38\u591a\uff0c\u603b\u4f53\u4e0a\u7ecf\u5178\u548c\u6d41\u884c\u7684Transformer\u6a21\u578b\u90fd\u53ef\u4ee5\u901a\u8fc7<a href=\"https:\/\/github.com\/huggingface\/transformers\">HuggingFace\/Transformers, 48.9k Star<\/a>\u83b7\u5f97\u548c\u514d\u8d39\u4f7f\u7528\uff0c\u4e3a\u521d\u5b66\u8005\u3001\u7814\u7a76\u4eba\u5458\u63d0\u4f9b\u4e86\u5de8\u5927\u7684\u5e2e\u52a9\u3002<\/p>\n\n\n\n<p>\u672c\u6559\u7a0b\u4e5f\u5c06\u57fa\u4e8e<a href=\"https:\/\/github.com\/huggingface\/transformers\">HuggingFace\/Transformers, 48.9k Star<\/a>\u8fdb\u884c\u5177\u4f53\u7684\u7f16\u7a0b\u548c\u4efb\u52a1\u89e3\u51b3\u65b9\u6848\u5b9e\u73b0\u3002<\/p>\n\n\n\n<p>NLP\u4e2d\u7684\u9884\u8bad\u7ec3+\u5fae\u8c03\u7684\u8bad\u7ec3\u65b9\u5f0f\u63a8\u8350\u9605\u8bfb\uff1a&nbsp;<a href=\"https:\/\/zhuanlan.zhihu.com\/p\/363802308\">2021\u5e74\u5982\u4f55\u79d1\u5b66\u7684\u201c\u5fae\u8c03\u201d\u9884\u8bad\u7ec3\u6a21\u578b\uff1f&nbsp;<\/a>\u548c<a href=\"https:\/\/zhuanlan.zhihu.com\/p\/49271699\">\u4eceWord Embedding\u5230Bert\u6a21\u578b\u2014\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4e2d\u7684\u9884\u8bad\u7ec3\u6280\u672f\u53d1\u5c55\u53f2<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>github\u5b66\u4e60\u5730\u5740\uff1ahttps:\/\/github.com\/datawhalechina\/learn-nlp- &hellip; <a href=\"http:\/\/139.9.1.231\/index.php\/2022\/08\/12\/natural-language-processing-nlp\/\" class=\"more-link\">\u7ee7\u7eed\u9605\u8bfb<span class=\"screen-reader-text\">Transformer\u7efc\u8ff0&#8212;(Natural Language Processing, NLP)<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[21],"tags":[],"_links":{"self":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/5762"}],"collection":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/comments?post=5762"}],"version-history":[{"count":4,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/5762\/revisions"}],"predecessor-version":[{"id":5768,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/5762\/revisions\/5768"}],"wp:attachment":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/media?parent=5762"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/categories?post=5762"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/tags?post=5762"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}