| { |
| "paper_id": "2020", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T12:53:12.588613Z" |
| }, |
| "title": "\u57fa \u57fa \u57fa\u4e8e \u4e8e \u4e8e\u8de8 \u8de8 \u8de8\u8bed \u8bed \u8bed\u8a00 \u8a00 \u8a00\u53cc \u53cc \u53cc\u8bed \u8bed \u8bed\u9884 \u9884 \u9884\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u53ca \u53ca \u53caBi-LSTM\u7684 \u7684 \u7684\u6c49 \u6c49 \u6c49-\u8d8a \u8d8a \u8d8a\u5e73 \u5e73 \u5e73\u884c \u884c \u884c\u53e5 \u53e5 \u53e5\u5bf9 \u5bf9 \u5bf9\u62bd \u62bd \u62bd\u53d6 \u53d6 \u53d6\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5", |
| "authors": [ |
| { |
| "first": "\u5218", |
| "middle": [ |
| "\u5218" |
| ], |
| "last": "\u5218\u7545", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kunming University of Science and Technology", |
| "location": { |
| "postCode": "650500", |
| "settlement": "Kunming", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "\uff0c", |
| "middle": [ |
| "\uff0c" |
| ], |
| "last": "\uff0c\u9ad8", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u9ad8", |
| "middle": [], |
| "last": "\u9ad8\u76db \u76db \u76db\u7965 \u7965 \u7965", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kunming University of Science and Technology", |
| "location": { |
| "postCode": "650500", |
| "settlement": "Kunming", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "\uff0c", |
| "middle": [ |
| "\uff0c" |
| ], |
| "last": "\uff0c\u4f59", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u4f59", |
| "middle": [], |
| "last": "\u4f59\u6b63 \u6b63 \u6b63\u6d9b", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kunming University of Science and Technology", |
| "location": { |
| "postCode": "650500", |
| "settlement": "Kunming", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "\uff0c", |
| "middle": [ |
| "\uff0c" |
| ], |
| "last": "\uff0c\u9ec4", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u9ec4", |
| "middle": [], |
| "last": "\u9ec4\u4e8e \u4e8e \u4e8e\u6b23", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kunming University of Science and Technology", |
| "location": { |
| "postCode": "650500", |
| "settlement": "Kunming", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "\uff0c", |
| "middle": [ |
| "\uff0c" |
| ], |
| "last": "\uff0c\u5c24", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u5c24", |
| "middle": [], |
| "last": "\u5c24\u4e1b \u4e1b \u4e1b\u4e1b", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kunming University of Science and Technology", |
| "location": { |
| "postCode": "650500", |
| "settlement": "Kunming", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Chang", |
| "middle": [], |
| "last": "Liu", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kunming University of Science and Technology", |
| "location": { |
| "postCode": "650500", |
| "settlement": "Kunming", |
| "country": "China" |
| } |
| }, |
| "email": "liuxiaochang32@163.com" |
| }, |
| { |
| "first": "\uff0c", |
| "middle": [ |
| "\uff0c" |
| ], |
| "last": "\uff0cshengxiang Gao", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kunming University of Science and Technology", |
| "location": { |
| "postCode": "650500", |
| "settlement": "Kunming", |
| "country": "China" |
| } |
| }, |
| "email": "gaoshengxiang.yn@foxmail.com" |
| }, |
| { |
| "first": "\uff0c", |
| "middle": [ |
| "\uff0c" |
| ], |
| "last": "\uff0czhengtao Yu", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kunming University of Science and Technology", |
| "location": { |
| "postCode": "650500", |
| "settlement": "Kunming", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Yuxin", |
| "middle": [], |
| "last": "Huang", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kunming University of Science and Technology", |
| "location": { |
| "postCode": "650500", |
| "settlement": "Kunming", |
| "country": "China" |
| } |
| }, |
| "email": "huangyuxin2004@163.com" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "\uff0ccongcong You", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kunming University of Science and Technology", |
| "location": { |
| "postCode": "650500", |
| "settlement": "Kunming", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "The extraction of Chinese-Vietnamese parallel sentence pairs is an important method to alleviate the scarcity of Chinese-Vietnamese parallel corpus data. Parallel sentence pair extraction can be converted into sentence similarity classification task in the same semantic space, the core of which is to achieve bilingual semantic space alignment. The traditional semantic space alignment method relies on large-scale bilingual parallel corpus, and it is relatively difficult for Vietnamese to obtain large-scale parallel corpus as a low-resource language. To address this problem, this paper proposes a bilingual dictionary for cross-lingual bilingual pre-training and Bi-LSTM (Bi-directional Long Short-Term Memory) Chinese-Vietnamese parallel sentence pair extraction method. Only a large number of Chinese-Vietnamese monolingual and a Chinese-Vietnamese \u00a92020 \u4e2d\u56fd\u8ba1\u7b97\u8bed\u8a00\u5b66\u5927\u4f1a \u6839\u636e\u300aCreative Commons Attribution 4.0 International License\u300b\u8bb8\u53ef\u51fa\u7248 *", |
| "pdf_parse": { |
| "paper_id": "2020", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "The extraction of Chinese-Vietnamese parallel sentence pairs is an important method to alleviate the scarcity of Chinese-Vietnamese parallel corpus data. Parallel sentence pair extraction can be converted into sentence similarity classification task in the same semantic space, the core of which is to achieve bilingual semantic space alignment. The traditional semantic space alignment method relies on large-scale bilingual parallel corpus, and it is relatively difficult for Vietnamese to obtain large-scale parallel corpus as a low-resource language. To address this problem, this paper proposes a bilingual dictionary for cross-lingual bilingual pre-training and Bi-LSTM (Bi-directional Long Short-Term Memory) Chinese-Vietnamese parallel sentence pair extraction method. Only a large number of Chinese-Vietnamese monolingual and a Chinese-Vietnamese \u00a92020 \u4e2d\u56fd\u8ba1\u7b97\u8bed\u8a00\u5b66\u5927\u4f1a \u6839\u636e\u300aCreative Commons Attribution 4.0 International License\u300b\u8bb8\u53ef\u51fa\u7248 *", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "1 \u5f15 \u5f15 \u5f15\u8a00 \u8a00 \u8a00 \u5e73\u884c\u8bed\u6599\u5e93\u7684\u89c4\u6a21\u548c\u8d28\u91cf\u5bf9\u4e8e\u673a\u5668\u7ffb\u8bd1\u7684\u6027\u80fd\u81f3\u5173\u91cd\u8981\u3002\u5728\u5927\u89c4\u6a21\u8bed\u6599\u4e2d\u7684\u673a\u5668\u7ffb\u8bd1\u5982 \u6c49-\u82f1\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u4e2d\u90fd\u5df2\u7ecf\u83b7\u5f97\u5f88\u597d\u7684\u7ed3\u679c\u3002\u7531\u4e8e\u6c49-\u8d8a\u4f4e\u8d44\u6e90\u8bed\u8a00\u5f88\u96be\u83b7\u5f97\u8db3\u591f\u591a\u7684\u5e73\u884c\u53e5 \u5bf9\uff0c\u4ece\u800c\u5bfc\u81f4\u6c49-\u8d8a\u673a\u5668\u7ffb\u8bd1\u6027\u80fd\u8f83\u5dee\u3002\u4eba\u5de5\u624b\u52a8\u6784\u5efa\u5927\u89c4\u6a21\u9ad8\u8d28\u91cf\u7684\u5e73\u884c\u8bed\u6599\u5e93\u8017\u65f6\u8017\u529b\uff0c\u901a \u8fc7\u5927\u91cf\u7684\u6587\u672c\u7814\u7a76\u53d1\u73b0\u5728\u540c\u4e00\u6bb5\u65f6\u95f4\u5185\u62a5\u9053\u7684\u65b0\u95fb\u6216\u5177\u6709\u540c\u4e00\u7ed3\u6784\u7684\u7f51\u9875\u90fd\u80fd\u83b7\u5f97\u5927\u91cf\u7684\u53ef\u6bd4 \u8bed\u6599\uff0c\u4ece\u53ef\u6bd4\u8bed\u6599\u4e2d\u62bd\u53d6\u5e73\u884c\u53e5\u5bf9\u662f\u6269\u5145\u7ffb\u8bd1\u8bed\u6599\u7684\u91cd\u8981\u65b9\u6cd5\u4e4b\u4e00\u3002\u672c\u6587\u7684\u76ee\u7684\u662f\u4ece\u6c49\u8d8a\u53ef\u6bd4 \u8bed\u6599\u4e2d\u62bd\u53d6\u6c49\u8d8a\u5e73\u884c\u53e5\u5bf9\u3002 \u76ee\u524d\u53cc\u8bed\u5e73\u884c\u53e5\u5bf9\u62bd\u53d6\u7684\u65b9\u6cd5\u5927\u81f4\u53ef\u4ee5\u5206\u4e3a\u4ee5\u4e0b\u56db\u7c7b\uff1a\u9996\u5148\u662f\u5229\u7528\u7edf\u8ba1\u673a\u5668\u7ffb\u8bd1\u548c\u795e\u7ecf \u673a\u5668\u7ffb\u8bd1\u65b9\u6cd5\u4ece\u53ef\u6bd4\u8bed\u6599\u5e93\u4e2d\u62bd\u53d6\u5e73\u884c\u53e5\u5bf9\u662f\u6bd4\u8f83\u6709\u6548\u7684\u3002\u5728\u7edf\u8ba1\u673a\u5668\u7ffb\u8bd1\u65b9\u9762\uff0cRauf\u7b49 \u4eba(Rauf and Schwenk, 2011)\u7684\u65b9\u6cd5\u662f\u5c06\u76ee\u6807\u8bed\u8a00\u7ffb\u8bd1\u6210\u6e90\u8bed\u8a00\uff0c\u5229\u7528\u8de8\u8bed\u8a00\u4fe1\u606f\u68c0\u7d22\u6280\u672f\u4ece \u53ef\u6bd4\u8bed\u6599\u5e93\u4e2d\u62bd\u53d6\u5e73\u884c\u53e5\u5bf9\uff0c\u63d0\u9ad8\u4e86\u7edf\u8ba1\u673a\u5668\u7ffb\u8bd1\u7684\u6027\u80fd\uff1b\u5728\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u65b9\u9762\uff0c (Marie and Fujita, 2017; Choudhary et al., 2018) \u63d0\u51fa\u4e86\u57fa\u4e8e\u8bcd\u5d4c\u5165\u5728\u5927\u578b\u5355\u8bed\u8bed\u6599\u5e93\u4e2d\u62bd\u53d6\u5e73\u884c\u53e5\u5bf9\u4ece\u800c \u63d0\u5347\u4e86\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u7684\u6027\u80fd\u3002Utiyama\u7b49\u4eba(Masao Utiyama, 2013)\u7ecf\u8fc7\u4e24\u6b21\u673a\u5668\u7ffb\u8bd1\uff0c\u9996\u5148\u5c06 \u65e5\u8bed\u53e5\u5b50\u7ffb\u8bd1\u5f97\u5230n-best\u82f1\u8bed\u8bd1\u6587\uff0c\u518d\u628a\u82f1\u8bed\u8bd1\u6587\u7ffb\u8bd1\u6210\u6c49\u8bed\uff0c\u6784\u5efa\u4e2d\u65e5\u5e73\u884c\u8bed\u6599\u5e93\u3002\u8fd9\u4e9b\u65b9 \u6cd5\u90fd\u662f\u901a\u8fc7\u6709\u6548\u62bd\u53d6\u5e73\u884c\u53e5\u5bf9\u6765\u63d0\u5347\u673a\u5668\u7ffb\u8bd1\u7684\u6027\u80fd\uff0c\u4f46\u9700\u8981\u5728\u7ffb\u8bd1\u6a21\u578b\u6027\u80fd\u6bd4\u8f83\u597d\u7684\u57fa\u7840\u4e0a \u624d\u80fd\u8fdb\u884c\u3002 \u5176\u6b21\u5728\u57fa\u4e8e\u7279\u5f81\u5de5\u7a0b\u65b9\u9762\uff0c (Chuang et al., 2004; Espa\u00f1a-Bonet et al., 2017; Luong et al., 2015) \u63d0\u51fa\u4e86\u5728\u53cc\u8bed\u8bcd\u5178\u4fe1\u606f\u7684\u57fa\u7840\u4e0a\u7ed3\u5408\u4e86\u6807\u70b9\u7b26\u53f7\u7edf\u8ba1\u4fe1\u606f\u548c\u8bcd\u6c47\u4fe1\u606f\u7684\u53cc\u8bed\u5e73\u884c\u6587\u672c\u5bf9 \u9f50\u7684\u65b9\u6cd5\uff1bGale\u7b49\u4eba(Gale and Church, 1991)\u4ecb\u7ecd\u4e86\u4e00\u79cd\u57fa\u4e8e\u5b57\u7b26\u957f\u5ea6\u7684\u7edf\u8ba1\u6a21\u578b\u5bf9\u9f50\u5e73\u884c\u6587 \u672c\u4e2d\u7684\u53e5\u5b50\u7684\u65b9\u6cd5\uff0c\u8bc6\u522b\u4e00\u79cd\u8bed\u8a00\u7684\u53e5\u5b50\u548c\u53e6\u4e00\u79cd\u8bed\u8a00\u7684\u53e5\u5b50\u4e4b\u95f4\u7684\u957f\u5ea6\u5bf9\u5e94\u5173\u7cfb\u3002Peng\u7b49 \u4eba (Peng et al., 2010) \u63d0\u51fa\u4e86\u4e00\u79cdFast-Champollion\u53e5\u5b50\u5bf9\u9f50\u7b97\u6cd5\uff0c\u5b83\u7ed3\u5408\u4e86\u57fa\u4e8e\u957f\u5ea6\u548c\u57fa\u4e8e\u8bcd \u5178\u4fe1\u606f\uff0c\u901a\u8fc7\u5c06\u8f93\u5165\u7684\u53cc\u8bed\u6587\u672c\u5206\u5272\u6210\u5c0f\u5757\u8fdb\u884c\u5bf9\u9f50\u7684\u8fc7\u7a0b\uff0c\u63d0\u5347\u53e5\u5b50\u5bf9\u9f50\u7684\u6548\u679c\u3002Ann \u7b49 \u4eba(Masao Utiyama, 2013)\u57fa\u4e8e\u73b0\u6709\u7684\u7ffb\u8bd1\u7cfb\u7edf\uff0c\u5c06\u6e90\u8bed\u8a00\u7ffb\u8bd1\u6210\u76ee\u6807\u8bed\u8a00\u5f97\u5230\u5019\u9009\u53e5\u5b50\uff0c\u7136\u540e \u5bf9\u5019\u9009\u53e5\u5b50\u5bf9\u8fdb\u884c\u6253\u5206\u6392\u5e8f\uff0c\u4ece\u800c\u83b7\u5f97\u5e73\u884c\u53e5\u5b50\u3002Chu\u7b49\u4eba (Chu et al., 2016) \u4ece\u5bf9\u9f50\u7684\u6587\u7ae0\u4e2d\u901a \u8fc7\u7b1b\u5361\u5c14\u4e58\u79ef\u751f\u6210\u6240\u6709\u53ef\u80fd\u7684\u53e5\u5b50\u5bf9\uff0c\u5e76\u8fc7\u6ee4\u6389\u4e0d\u6ee1\u8db3\u6761\u4ef6\u7684\u53e5\u5b50\u5bf9\uff0c\u4fdd\u7559\u5c3d\u53ef\u80fd\u5339\u914d\u7684\u53e5\u5b50 \u5bf9\uff0c\u7136\u540e\u4f7f\u7528\u5c11\u91cf\u5e73\u884c\u53e5\u5bf9\u8bad\u7ec3\u5206\u7c7b\u5668\uff0c\u4ee5\u4ece\u5019\u9009\u8005\u4e2d\u8bc6\u522b\u5e73\u884c\u53e5\u5bf9\u3002Tillmann\u7b49\u4eba (Tillmann and Xu, 2009 (Artetxe et al., 2016; Artetxe et al., 2017; Artetxe et al., 2018 ", |
| "cite_spans": [ |
| { |
| "start": 407, |
| "end": 431, |
| "text": "(Marie and Fujita, 2017;", |
| "ref_id": "BIBREF16" |
| }, |
| { |
| "start": 432, |
| "end": 455, |
| "text": "Choudhary et al., 2018)", |
| "ref_id": "BIBREF7" |
| }, |
| { |
| "start": 644, |
| "end": 665, |
| "text": "(Chuang et al., 2004;", |
| "ref_id": "BIBREF9" |
| }, |
| { |
| "start": 666, |
| "end": 692, |
| "text": "Espa\u00f1a-Bonet et al., 2017;", |
| "ref_id": "BIBREF10" |
| }, |
| { |
| "start": 693, |
| "end": 712, |
| "text": "Luong et al., 2015)", |
| "ref_id": "BIBREF15" |
| }, |
| { |
| "start": 853, |
| "end": 872, |
| "text": "(Peng et al., 2010)", |
| "ref_id": "BIBREF20" |
| }, |
| { |
| "start": 1038, |
| "end": 1056, |
| "text": "(Chu et al., 2016)", |
| "ref_id": "BIBREF8" |
| }, |
| { |
| "start": 1150, |
| "end": 1172, |
| "text": "(Tillmann and Xu, 2009", |
| "ref_id": "BIBREF22" |
| }, |
| { |
| "start": 1173, |
| "end": 1195, |
| "text": "(Artetxe et al., 2016;", |
| "ref_id": "BIBREF0" |
| }, |
| { |
| "start": 1196, |
| "end": 1217, |
| "text": "Artetxe et al., 2017;", |
| "ref_id": "BIBREF1" |
| }, |
| { |
| "start": 1218, |
| "end": 1238, |
| "text": "Artetxe et al., 2018", |
| "ref_id": "BIBREF2" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": ")\u601d\u60f3\u542f\u53d1\uff0c\u63d0\u51fa\u4e86\u4e00\u4e2a\u57fa\u4e8e\u8de8\u8bed\u8a00\u53cc\u8bed\u9884\u8bad\u7ec3 \u53caBi-LSTM\u7684\u6c49-\u8d8a\u5e73\u884c\u53e5\u5bf9\u62bd\u53d6\u65b9\u6cd5\uff0c\u4ece\u6c49\u8d8a\u53ef\u6bd4\u8bed\u6599\u4e2d\u62bd\u53d6\u6c49\u8d8a\u5e73\u884c\u53e5\u5bf9\uff0c\u6765\u63d0\u5347\u4f4e\u8d44\u6e90 \u8bed\u8a00\u673a\u5668\u7ffb\u8bd1\u7684\u6027\u80fd\u3002\u5176\u4e3b\u8981\u601d\u60f3\u662f\u5728\u6c49\u8d8a\u53cc\u8bed\u9884\u8bad\u7ec3\u4e2d\u5c06\u6c49\u8d8a\u53cc\u8bed\u53e5\u5b50\u6620\u5c04\u5230\u516c\u5171\u8bed\u4e49\u7a7a\u95f4 \u4e0b\uff0c\u901a\u8fc7\u6c49-\u8d8a\u79cd\u5b50\u8bcd\u5178\u6765\u7f29\u5c0f\u6c49\u8d8a\u53cc\u8bed\u5728\u8bed\u4e49\u7a7a\u95f4\u4e2d\u7684\u8ddd\u79bb\uff0c\u4ece\u800c\u52a0\u5f3a\u6c49\u8d8a\u53cc\u8bed\u7684\u8bed\u4e49\u76f8\u5173 \u6027\u3002\u5728\u672c\u6587\u65b9\u6cd5\u4e2d\u9488\u5bf9\u7684\u662f\u6c49\u8bed\u5230\u8d8a\u5357\u8bed\u4e24\u79cd\u8bed\u8a00\uff0c\u7531\u4e8e\u6c49\u8bed\u5230\u8d8a\u5357\u8bed\u6ca1\u6709\u516c\u5f00\u7684\u6570\u636e\u96c6\uff0c \u56e0\u6b64\u8003\u8651\u4ece\u7ef4\u57fa\u767e\u79d1\u6587\u7ae0\u4e2d\u62bd\u53d6\u7684\u6c49-\u8d8a\u6bb5\u843d\u8bed\u6599\u4ee5\u53ca\u6536\u96c6\u7684\u6c49-\u8d8a\u6bb5\u843d\u8bed\u6599\u6dfb\u52a0\u5230\u4e00\u4e2a\u8bed\u6599\u5e93 \u4e2d\uff0c\u4ee5\u8bad\u7ec3\u6a21\u578b\u7684\u6027\u80fd\u3002 2 \u57fa \u57fa \u57fa\u4e8e \u4e8e \u4e8e\u6c49 \u6c49 \u6c49-\u8d8a \u8d8a \u8d8a\u53cc \u53cc \u53cc\u8bed \u8bed \u8bed\u9884 \u9884 \u9884\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u53ca \u53ca \u53caBi-LSTM\u7684 \u7684 \u7684\u5e73 \u5e73 \u5e73\u884c \u884c \u884c\u53e5 \u53e5 \u53e5\u5bf9 \u5bf9 \u5bf9\u62bd \u62bd \u62bd\u53d6 \u53d6 \u53d6\u6a21 \u6a21 \u6a21\u578b \u578b \u578b \u9488\u5bf9\u4e0a\u6587\u95ee\u9898\uff0c\u63d0\u51fa\u4e00\u4e2a\u57fa\u4e8e\u6c49-\u8d8a\u53cc\u8bed\u9884\u8bad\u7ec3\u53caBi-LSTM\u7684\u5e73\u884c\u53e5\u5bf9\u62bd\u53d6\u65b9\u6cd5\uff0c\u5177\u4f53\u6a21\u578b \u7ed3\u6784\u4f53\u7cfb\u5982\u56fe1\u6240\u793a\u3002\u8be5\u6a21\u578b\u4e3b\u8981\u5206\u4e3a\u4e09\u4e2a\u90e8\u5206\u3002\u7b2c\u4e00\u90e8\u5206\u662f\u57fa\u4e8e\u6c49-\u8d8a\u53cc\u8bed\u9884\u8bad\u7ec3\uff0c\u7b2c\u4e8c\u90e8\u5206 \u662f\u7531Bi-LSTM\u548cCNN\u7ec4\u6210\u7684\u6c49-\u8d8a\u53e5\u5b50\u7279\u5f81\u63d0\u53d6\u90e8\u5206\u7684\u7f16\u7801\u5668\uff0c\u7b2c\u4e09\u90e8\u5206\u662f\u5168\u8fde\u63a5\u5c42\u8fdb\u884c\u6c49-\u8d8a \u5e73\u884c\u548c\u53e5\u975e\u5e73\u884c\u53e5\u5206\u7c7b\u3002 \u9996\u5148\uff0c\u5c06\u6c49\u8bed-\u8d8a\u5357\u8bed\u8de8\u8bed\u8a00\u53cc\u8bed\u8bcd\u5d4c\u5165\u6620\u5c04\u5230\u516c\u5171\u7684\u8bed\u4e49\u7a7a\u95f4\u8fdb\u884c\u9884\u8bad\u7ec3\uff0c\u4f7f\u5f97\u6c49 \u8bed-\u8d8a\u5357\u8bed\u7684\u8bed\u4e49\u76f8\u4f3c\u8bcd\u5728\u8be5\u7a7a\u95f4\u4e2d\u63a5\u8fd1\uff0c\u589e\u5f3a\u6c49\u8bed\u548c\u8d8a\u5357\u8bed\u8bed\u4e49\u7a7a\u95f4\u4e2d\u7684\u76f8\u5173\u6027\u3002\u8bbex = (x 1 , x 2 , ..., x m ) \u8868\u793a\u8868\u793a\u8f93\u5165\u7684\u6c49\u8bed\u5355\u8bcd, y = (y 1 , y 2 , ..., y n ) \u8868\u793a\u8f93\u5165\u7684\u8d8a\u5357\u8bed\u5355\u8bcd\u3002\u5728\u53cc \u8bed\u9884\u8bad\u7ec3\u4e2d\uff0c\u6c49-\u8d8a\u79cd\u5b50\u8bcd\u5178\u5728\u6ca1\u6709\u5927\u89c4\u6a21\u5e73\u884c\u8bed\u6599\u60c5\u51b5\u4e0b\u53ef\u4ee5\u5b9e\u73b0\u5728\u6c49\u8d8a\u7edf\u4e00\u7a7a\u95f4\u8bed\u4e49\u5bf9 \u9f50\uff0c\u5e76\u4ee5\u81ea\u5b66\u4e60\u7684\u65b9\u5f0f\u8fed\u4ee3\u5730\u751f\u6210\u65b0\u8bcd\u5178\u3002\u518d\u5229\u7528\u6c49-\u8d8a\u79cd\u5b50\u8bcd\u5178\u6765\u5b66\u4e60\u8bcd\u5d4c\u5165\u5e76\u6307\u5bfc\u540e\u9762Bi- LSTM\u548cCNN\u5728\u516c\u5171\u8bed\u4e49\u7a7a\u95f4\u8fdb\u884c\u7edf\u4e00\u7f16\u7801\u3002\u5c06\u8bad\u7ec3\u597d\u7684\u8bcd\u5411\u91cf\u8f93\u5165Bi-LSTM\u6765\u83b7\u53d6\u5355\u8bcd\u524d\u540e \u4fe1\u606f\u7279\u5f81\uff0c\u5e76\u7528CNN\u6765\u63d0\u53d6\u53cc\u8bed\u53e5\u5b50\u66f4\u6df1\u5c42\u8bed\u4e49\u7279\u5f81\u3002\u6700\u540e\u5bf9\u6c49\u8bed\u53e5\u5b50\u548c\u8d8a\u5357\u8bed\u53e5\u5b50\u8fdb\u884c\u7f16 \u7801\uff0c\u901a\u8fc7\u4f7f\u7528\u5143\u7d20\u4e58\u79ef\u548c\u5143\u7d20\u7edd\u5bf9\u5dee\u5c06\u5b83\u4eec\u63d0\u4f9b\u7ed9\u5168\u8fde\u63a5\u5c42\uff0c\u4f7f\u7528\u8f93\u51fa\u6982\u7387\u4f5c\u4e3a\u6c49\u8d8a\u53e5\u5bf9\u662f\u5426 \u4e3a\u5e73\u884c\u8bed\u53e5\u5bf9\u7684\u5ea6\u91cf\u6765\u6355\u83b7\u5176\u5339\u914d\u4fe1\u606f\u3002 3 \u6c49 \u6c49 \u6c49\u8d8a \u8d8a \u8d8a\u8de8 \u8de8 \u8de8\u8bed \u8bed \u8bed\u8a00 \u8a00 \u8a00\u8bcd \u8bcd \u8bcd\u5411 \u5411 \u5411\u91cf \u91cf \u91cf\u9884 \u9884 \u9884\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3 3.1 \u8bcd \u8bcd \u8bcd\u5411 \u5411 \u5411\u91cf \u91cf \u91cf\u9884 \u9884 \u9884\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5 \u5728 \u53cc \u8bed \u4e2d \uff0c \u5229 \u7528 \u5355 \u72ec \u8bed \u6599 \u8fdb \u884c \u72ec \u7acb \u8bad \u7ec3 \u7684 \u65b9 \u6cd5 \u5982Mikolov\u7b49 \u4eba(Mikolov et al., 2013)\u7684word2vec(CBOW/Skip-grim)\u8bad\u7ec3\u51fa\u6709\u8bed\u4e49\u76f8\u4f3c\u6027\u7684\u8bcd\u5d4c\u5165\u5411\u91cf\u3002\u5728\u5404\u81ea\u8bed\u6599\u4e0a\u8fdb\u884c \u72ec\u7acb\u8bad\u7ec3\uff0c\u5bfc\u81f4\u4e24\u79cd\u8bed\u8a00\u8bcd\u5d4c\u5165\u77e9\u9635\u5728\u5206\u5e03\u4e0a\u4e5f\u662f\u72ec\u7acb\u4e0d\u76f8\u5173\u3002\u5728\u6c49\u8bed\u548c\u8d8a\u5357\u8bed\u8bcd\u5411\u91cf\u8868\u5f81\u4e2d \u4e5f\u662f\u5982\u6b64\u3002\u53cc\u8bed\u8bcd\u5d4c\u5165\u5c06\u4e24\u79cd\u4e0d\u540c\u8bed\u8a00\u7684\u8bcd\u6620\u5c04\u5230\u516c\u5171\u7684\u8bed\u4e49\u7a7a\u95f4\uff0c\u516c\u5171\u8bed\u4e49\u7a7a\u95f4\u4e2d\u6bcf\u4e2a\u5355\u8bcd \u5d4c\u5165\u4e4b\u95f4\u7684\u8ddd\u79bb\u5219\u6697\u793a\u7740\u4e00\u5b9a\u7684\u8bed\u4e49\u5173\u7cfb\u3002\u8fd9\u53ef\u4ee5\u4fdd\u8bc1\u5728\u5355\u8bed\u8bed\u4e49\u4e0d\u53d8\u6027\u60c5\u51b5\u4e0b\u786e\u4fdd\u5177\u6709\u4e24\u4e2a \u76f8\u540c\u8bed\u4e49\u7684\u8bcd\u5728\u516c\u5171\u8bed\u4e49\u7a7a\u95f4\u4e2d\u7684\u8ddd\u79bb\u975e\u5e38\u8fd1\uff0c\u4f46\u53cc\u8bed\u8bcd\u5d4c\u5165\u7684\u5b66\u4e60\u90fd\u4f9d\u8d56\u4e8e\u5927\u89c4\u6a21\u5e73\u884c\u8bed\u6599 \u5e93\uff0c\u8fd9\u5bf9\u4e8e\u8d44\u6e90\u7a00\u7f3a\u578b\u8bed\u8a00\u5bf9(\u6c49\u8bed-\u8d8a\u5357\u8bed)\u662f\u96be\u4ee5\u83b7\u5f97\u7684\u3002 \u6211\u4eec\u5728\u6c49\u8d8a\u8de8\u8bed\u8a00\u8bcd\u5411\u91cf\u9884\u8bad\u7ec3\u4e2d\u63d0\u51fa\u4e86\u4e00\u79cd\u81ea\u5b66\u4e60\u7684\u65b9\u6cd5\u3002\u8be5\u65b9\u6cd5\u5229\u7528\u4e86\u5d4c\u5165\u7a7a\u95f4\u7684\u7ed3 \u6784\u76f8\u4f3c\u6027\uff0c\u7ed3\u5408\u57fa\u4e8e\u6c49\u8bed-\u8d8a\u5357\u8bed\u79cd\u5b50\u8bcd\u5178\u7684\u6620\u5c04\u6280\u672f\uff0c\u964d\u4f4e\u4e86\u6c49\u8bed-\u8d8a\u5357\u8bed\u53cc\u8bed\u8d44\u6e90\u7684\u9700\u6c42\u3002 \u8be5\u81ea\u5b66\u4e60\u7684\u65b9\u6cd5\u6846\u67b6\u5148\u662f\u5bf9\u6c49\u8bed\u548c\u8d8a\u5357\u8bed\u5728\u5404\u81ea\u7684\u5355\u8bed\u8bed\u6599\u5e93\u4e0a\u8fdb\u884c\u72ec\u7acb\u8bad\u7ec3\uff0c\u518d\u901a\u8fc7\u7ebf\u6027\u53d8 \u6362\u6765\u6700\u5c0f\u5316\u6c49\u8d8a\u53cc\u8bed\u8bcd\u5178\u4e2d\u7684\u8ddd\u79bb\u4ece\u800c\u5c06\u6c49\u8bed-\u8d8a\u5357\u8bed\u8de8\u8bed\u8a00\u6620\u5c04\u5728\u540c\u4e00\u8bed\u4e49\u7a7a\u95f4\u3002\u901a\u5e38\u9700\u8981\u5927 \u89c4\u6a21\u53cc\u8bed\u8bcd\u5178\u8fdb\u884c\u8bad\u7ec3\uff0c\u9488\u5bf9\u6c49\u8bed-\u8d8a\u5357\u8bed\u96be\u4ee5\u83b7\u53d6\u5927\u89c4\u6a21\u8bcd\u5178\uff0c\u8de8\u8bed\u8a00\u9884\u8bad\u7ec3\u4e2d\u5c06\u5927\u578b\u53cc\u8bed\u8bcd \u5178\u7684\u9700\u6c42\u51cf\u5c11\u5230\u8f83\u5c0f\u7684\u79cd\u5b50\u8bcd\u5178\uff0c\u901a\u8fc7\u4e0d\u65ad\u8fed\u4ee3\u4f7f\u66f4\u65b0\u7684\u79cd\u5b50\u8bcd\u5178\u6765\u5b66\u4e60\u65b0\u7684\u6620\u5c04\u77e9\u9635\uff0c\u76f4\u81f3 \u6536\u655b\u3002\u6c49-\u8d8a\u8de8\u8bed\u8a00\u53cc\u8bed\u8bcd\u5d4c\u5165\u9884\u8bad\u7ec3\u5177\u4f53\u7ec6\u8282\u5982\u4e0b\u56fe2\u6240\u793a\uff1a Figure 1: \u57fa\u4e8e\u6c49-\u8d8a\u53cc\u8bed\u9884\u8bad\u7ec3\u53caBi-LSTM\u7684\u5e73\u884c\u53e5\u5bf9\u62bd\u53d6\u6a21\u578b Figure 2: \u6c49-\u8d8a\u8de8\u8bed\u8a00\u53cc\u8bed\u8bcd\u5d4c\u5165\u9884\u8bad\u7ec3\u8fc7\u7a0b 3.2 \u8bcd \u8bcd \u8bcd\u5411 \u5411 \u5411\u91cf \u91cf \u91cf\u9884 \u9884 \u9884\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u7684 \u7684 \u7684\u57fa \u57fa \u57fa\u672c \u672c \u672c\u6b65 \u6b65 \u6b65\u9aa4 \u9aa4 \u9aa4 \u9996\u5148\uff0c\u6784\u5efa\u4e00\u4e2a\u6c49\u8bed\u548c\u8d8a\u5357\u8bed\u540c\u65f6\u6620\u5c04\u7684\u7279\u5f81\u5411\u91cf\u7a7a\u95f4\uff0c\u6c49\u8bed\u8bed\u6599\u8bad\u7ec3\u5f97\u5230\u7684\u8bcd\u5d4c\u5165\u77e9 \u9635X\uff0c\u8d8a\u5357\u8bed\u8bed\u6599\u4e2d\u8bad\u7ec3\u7684\u8bcd\u5d4c\u5165\u77e9\u9635Z\u3002\u5c06\u79cd\u5b50\u5b57\u5178\u8868\u793a\u4e3a\u4e00\u4e2a\u4e8c\u8fdb\u5236\u77e9\u9635D\u3002D ij = 1\u65f6\u4ee3 \u8868\u8d8a\u5357\u8bed\u4e2d\u7684\u7b2cj\u4e2a\u5355\u8bcd\u662f\u6c49\u8bed\u4e2d\u7b2ci\u4e2a\u5355\u8bcd\u7684\u7ffb\u8bd1\u3002\u7136\u540e\u627e\u5230\u6700\u4f73\u6620\u5c04\u77e9\u9635W ,\u8ba9\u6c49\u8bed\u8bcd\u5411\u91cf\u548c \u8d8a\u5357\u8bed\u8bcd\u5411\u91cf\u5206\u5e03\u5728\u540c\u4e00\u4e2a\u5411\u91cf\u7a7a\u95f4\uff0c\u4f7f\u5f97\u6620\u5c04\u6c49\u8bed\u8bcd\u5d4c\u5165X i * W \u4e0e\u8d8a\u5357\u8bed\u8bcd\u5d4c\u5165Z j * \u4e4b\u95f4\u7684\u6b27 \u51e0\u91cc\u5fb7\u8ddd\u79bb\u7684\u5e73\u65b9\u548c\u6700\u5c0f\uff0c\u6620\u5c04\u77e9\u9635\uff1a W * = argmin W i j D ij ||X i * W \u2212 Z j * || 2 (1) \u5176\u4e2d\uff0c\u9884\u5904\u7406\u6b65\u9aa4\u5bf9\u8bcd\u5d4c\u5165\u77e9\u9635X\u548cZ\u8fdb\u884c\u957f\u5ea6\u5f52\u4e00\u5316\u548c\u5e73\u5747\u5c45\u4e2d\uff0c\u6700\u540e\u518d\u8fdb\u884c\u4e00\u6b21\u5f52\u4e00 \u5316\u5904\u7406\uff0c\u5e76\u5c06W \u7ea6\u675f\u4e3a\u6b63\u4ea4\u77e9\u9635\u5373W W T = W T W = I\uff0c\u4ee5\u5f3a\u5236\u6267\u884c\u6c49\u8bed\u548c\u8d8a\u5357\u8bed\u7684\u5355\u8bed\u4e0d\u53d8 \u6027\uff0c\u9632\u6b62\u5355\u8bed\u6027\u80fd\u7684\u964d\u4f4e\uff0c\u540c\u65f6\u53ef\u4ee5\u4ea7\u751f\u66f4\u597d\u7684\u6c49\u8bed-\u8d8a\u5357\u8bed\u8de8\u539f\u8bed\u8a00\u53cc\u8bed\u6620\u5c04\u3002\u5728\u8fd9\u79cd\u6b63\u4ea4 \u6027\u7ea6\u675f\u4e0b\uff0c\u6700\u5c0f\u5316\u5e73\u65b9\u6b27\u51e0\u91cc\u5fb7\u8ddd\u79bb\u5c31\u7b49\u4e8e\u6700\u5927\u5316\u70b9\u79ef\uff0c\u56e0\u6b64\u56e0\u6b64\u6620\u5c04\u77e9\u9635\u88ab\u5b9a\u4e49\u4e3a\u5982\u4e0b\u516c \u5f0f(2)\u6240\u793a\uff1a W * = argmax W T r(XW Z T D T ) (2) \u5176\u4e2d\uff0cT r(\u2022)\u8868\u793a\u4e3b\u5bf9\u89d2\u7ebf\u4e0a\u7684\u6240\u6709\u5143\u7d20\u4e4b\u548c\uff0cW * = U V T \u7ed9\u51fa\u4e86\u6b64\u95ee\u9898\u7684\u6700\u4f73\u6b63\u4ea4\u89e3\uff0c\u5176 \u4e2dX T DZ = U V T \u662fX T DZ\u7684\u5947\u5f02\u503c\u5206\u89e3\u3002\u7531\u4e8e\u5b57\u5178\u77e9\u9635D\u662f\u7a00\u758f\u7684\uff0c\u8fd9\u53ef\u4ee5\u6709\u6548\u5730\u5728\u7ebf\u6027 \u65f6\u95f4\u5185\u5bf9\u5b57\u5178\u6761\u76ee\u6570\u8fdb\u884c\u8ba1\u7b97\u3002 \u83b7\u5f97\u4e86\u8fd9\u4e2a\u6620\u5c04\u77e9\u9635W \u4e4b\u540e\uff0c\u5bf9\u4e8e\u5b57\u5178\u5916\u7684\u4efb\u4f55\u4e00\u4e2a\u6ca1\u6709\u7ffb\u8bd1\u7684\u5355\u8bcd\uff0c\u53ef\u4ee5\u6839\u636e\u6620\u5c04\u540e\u7684 \u7a7a\u95f4\u4f59\u5f26\u76f8\u4f3c\u5ea6\u6765\u8fdb\u884c\u8bcd\u5bf9\u9f50\u3002\u5728\u6700\u8fd1\u90bb\u68c0\u7d22\u4e2d\uff0c\u4e3a\u6bcf\u4e2a\u6e90\u8bed\u8a00\u5355\u8bcd\u5206\u914d\u4e86\u76ee\u6807\u8bed\u8a00\u4e2d\u6700\u63a5\u8fd1 \u7684\u5355\u8bcd\uff0c\u6211\u4eec\u5c06\u6620\u5c04\u7684\u6e90\u8bed\u8a00\u5d4c\u5165\u548c\u76ee\u6807\u8bed\u8a00\u5d4c\u5165\u4e4b\u95f4\u7684\u70b9\u79ef\u7528\u4f5c\u76f8\u4f3c\u5ea6\u5ea6\u91cf\u3002\u6700\u540e\uff0c\u901a\u8fc7\u77e2 \u91cf\u5316\u76f8\u4f3c\u77e9\u9635XW Z T \u5e76\u8fdb\u884c\u4e0d\u65ad\u8fed\u4ee3\u8ba1\u7b97\uff0c\u627e\u5230\u8be5\u77e9\u9635\u7684\u6700\u5927\u503c\uff0c\u4ece\u800c\u8fbe\u5230\u4f18\u5316\u76ee\u6807\u3002 cos dic (wx i , z j ) = n i=1 wx i z j n i=1 (wx i ) 2 n j=1 (z j ) 2 (3) 4 \u57fa \u57fa \u57fa\u4e8e \u4e8e \u4e8eBi-LSTM\u548c \u548c \u548cCNN\u516c \u516c \u516c\u5171 \u5171 \u5171\u8bed \u8bed \u8bed\u4e49 \u4e49 \u4e49\u7a7a \u7a7a \u7a7a\u95f4 \u95f4 \u95f4\u7f16 \u7f16 \u7f16\u7801 \u7801 \u7801 \u57fa\u4e8eLSTM\u6a21\u578b\u5145\u5206\u8003\u8651\u4e86\u957f\u8ddd\u79bb\u5355\u8bcd\u4e4b\u95f4\u7684\u4f9d\u8d56\u6027\uff0c\u5e76\u4fdd\u7559\u4e86\u8bf8\u5982\u5355\u8bcd\u987a\u5e8f\u4e4b\u7c7b\u7684\u529f \u80fd\u3002\u540c\u65f6CNN\u6a21\u578b\u53ef\u4ee5\u63d0\u53d6\u4e30\u5bcc\u7684\u7ec4\u5408\u7279\u5f81\u53ca\u5377\u79ef\u6838\u7684\u591a\u6837\u6027\u3002\u4f46\u662f\u7531\u4e8eLSTM\u4e0d\u4f7f\u7528\u53cd\u5411\u5355 \u8bcd\u7f16\u7801\u4fe1\u606f\uff0c\u56e0\u6b64\u4e0d\u80fd\u5728\u53cc\u5411\u5355\u8bcd\u7f16\u7801\u4e2d\u5b66\u4e60\u5230\u8bed\u4e49\u4fe1\u606f\u7279\u5f81\uff0c\u800cBi-LSTM\u53ef\u4ee5\u8003\u8651\u5355\u8bcd\u7684 \u53cc\u5411\u7f16\u7801\u3002\u518d\u4f7f\u7528CNN\u5377\u79ef\u5e76\u5408\u5e76Bi-LSTM\u7684\u8f93\u51fa\u4ee5\u63d0\u53d6\u53e5\u5b50\u7684\u5173\u952e\u8bed\u4e49\u7279\u5f81\u3002\u4e3a\u4e86\u8003\u8651\u4e0a \u8ff0\u7279\u5f81\uff0c\u7f16\u7801\u5668\u7531\u4e24\u5c42Bi-LSTM\u548cCNN\u5806\u53e0\u6210\u4e00\u4e2a\u57fa\u672c\u7684\u7f16\u7801\u5355\u5143\uff0c\u4f9d\u6b21\u4ece\u6e90\u8bed\u53e5\u548c\u76ee\u6807\u53e5 \u4e2d\u63a5\u53d7\u6bcf\u4e2a\u5355\u8bcd\u7684\u5355\u8bcd\u5d4c\u5165\u77e9\u9635W x \u2208 R d\u00d7|V x| \u6765\u8f93\u5165\u5355\u8bcdx,\u5176\u4e2dd\u4e3a\u5355\u8bcd\u5d4c\u5165\u5411\u91cf\u7684\u7ef4\u6570,V x \u4e3a \u6240\u6709\u8f93\u5165\u5355\u8bcd\u7684\u96c6\u5408\u3002\u6bcf\u4e2a\u65f6\u523b\u5185\uff0c\u7531\u8bcd\u6c47\u8868V x \u4e2d\u7684\u6574\u6570\u7d22\u5f15k\u5b9a\u4e49\u7684\u7b2ci\u4e2a\u53e5\u5b50\u4e2d\u7684\u6807\u8bb0\u8868\u793a \u4e3aone-hot\u5411\u91cfw S k \u2208 0, 1 |V | ,\u8be5one-hot\u5411\u91cf\u4e0e\u8bcd\u5d4c\u5165\u77e9\u9635E S T \u2208 R |V x|\u00d7de \u76f8\u4e58\uff0c\u4ee5\u83b7\u5f97\u8be5\u6807\u8bb0\u7684 \u8fde\u7eed\u5411\u91cf\u8868\u793aw S i \uff0c\u5176\u4f5c\u4e3aBi-LSTM\u7f16\u7801\u5668\u7684\u524d\u5411\u548c\u540e\u5411\u5faa\u73af\u72b6\u6001\u7684\u8f93\u5165\u3002\u524d\u5411LSTM\u8bfb\u53d6\u53d8\u957f \u53e5\uff0c\u5e76\u4ece\u7b2c\u4e00\u4e2a\u6807\u8bb0\u5230\u6700\u540e\u4e00\u4e2a\u6807\u8bb0\u66f4\u65b0\u5176\u9012\u5f52\u72b6\u6001\uff0c\u4ece\u800c\u521b\u5efa\u4e00\u4e2a\u56fa\u5b9a\u5927\u5c0f\u7684\u53e5\u5b50\u8fde\u7eed\u5411\u91cf \u8868\u793a\uff1b\u540e\u5411LSTM\u53cd\u5411\u5904\u7406\u8be5\u53e5\u5b50\uff0c\u7136\u540e\u5c06\u7b2c\u4e8c\u5c42\u76f8\u540c\u4f4d\u7f6e\u4e0a\u6bcf\u4e2a\u65f6\u95f4\u6b65\u957f\u7684\u4e24\u4e2a\u65b9\u5411\u7684\u7f16\u7801 \u5668\u8f93\u51fa\u90fd\u62fc\u63a5\u5728\u4e00\u8d77h i = [ \u2212 \u2192 h S i , \u2190 \u2212 h S i ]\uff0c\u4f5c\u4e3a\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u7684\u8f93\u5165\u3002\u524d\u5411\u9012\u5f52\u72b6\u6001\u548c\u540e\u5411\u9012\u5f52\u72b6\u6001 \u5206\u522b\u8ba1\u7b97\u5982\u4e0b\uff1a w S i = E S T w S k (4) \u2212 \u2192 h S i = \u03c6( \u2212 \u2192 h S i\u22121 , w S i ) (5) \u2190 \u2212 h S i = \u03c6( \u2190 \u2212 h S i\u22121 , w S i )", |
| "eq_num": "(6)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "h i = [ \u2212 \u2192 h S i , \u2190 \u2212 h S i ]", |
| "eq_num": "(7)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\u5176\u4e2dE\u8868\u793a\u5355\u8bcd\u5d4c\u5165\uff0c\u03c6(\u2022)\u662fLSTM\u6a21\u5757\u3002 \u539f\u59cb\u7684CNN\u7531\u5377\u79ef\u5c42\uff0c\u6c60\u5316\u5c42\u548c\u5168\u8fde\u63a5\u5c42\u7ec4\u6210\u3002\u5bf9\u4e8e\u53e5\u5b50\u957f\u5ea6\u4e3an\u7684\u53e5\u5b50\uff0c\u53ef\u4ee5\u5c06\u5b83\u8868\u793a \u6210x 1:n = x 1 \u2295 x 2 \u2295 ... \u2295 x n \uff0c\u2295\u8868\u793a\u5168\u8fde\u63a5\uff0cx i \u2208 d \u8868\u793a\u7684\u662f\u7b2ci\u4e2a\u8bcd\u5411\u91cf\uff0cd\u8868\u793a\u7684\u662f\u8bcd\u5411\u91cf\u7684 \u7ef4\u5ea6\u3002\u5377\u79ef\u8fd0\u7b97\u7684\u6838\u5fc3\u662f\u5bf9\u6ed1\u52a8\u7a97\u53e3\u7684\u5927\u5c0f\u7684\u5e8f\u5217\u5e94\u7528\u5728\u8fc7\u6ee4\u5668\u4e0a\u4ee5\u4ea7\u751f\u65b0\u7684\u7279\u5f81\uff0c\u5982\u4e0b\u516c\u5f0f \u6240\u793a\uff0c c i = f (W \u2022 x i:i+h\u22121 + b) (8) \u5176\u4e2d\uff0cb \u2208 \u662f\u4e00\u4e2a\u504f\u79fb\u5411\u91cf\uff0cf \u662f\u975e\u7ebf\u6027\u51fd\u6570(\u6bd4\u5982Sigmoid\uff0cReLU)\u3002\u957f\u5ea6\u4e3an\u7684\u53e5\u5b50 \u53ef\u4ee5\u901a\u8fc7\u5377\u79ef\u5c42\u83b7\u5f97\u53e5\u5b50\u4e2d\u4efb\u4f55\u8fde\u7eed\u5355\u8bcd\u5e8f\u5217\u7684\u6df1\u5c42\u8bed\u4e49\u7279\u5f81\uff0c\u5982\u516c\u5f0f\u6240\u793a\uff0c c = [c 1 , c 2 , ..., c n\u2212h+1 ]", |
| "eq_num": "(9)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u672c\u6587\u5c06\u7a97\u53e3\u5927\u5c0f\u4e3aF = [F (0)...F (m \u2212 1)]\u7684\u5377\u79ef\u6838\u4e0eBi-LSTM\u7684\u8f93\u51fa\u5411\u91cf\u8fdb\u884c\u5377\u79ef\u4ee5\u83b7\u5f97 \u7279\u5f81\u5411\u91cf\uff0c\u5982\u516c\u5f0f\u6240\u793a\uff1a c = tanh[( m\u22121 i=0 h(t + i) T F (i)) + b] (10) b\u662f\u504f\u79fb\u5411\u91cf\uff0cF \u548cb\u662f\u8fc7\u6ee4\u5668\u7684\u53c2\u6570\u3002\u4ece\u5178\u578b\u7684CNN\u7ed3\u6784\u53ef\u4ee5\u770b\u51fa\uff0c\u6c60\u5316\u5c42\u6784\u5efa\u5728\u5377\u79ef\u5c42\u4e4b \u4e0a\u3002\u5728\u672c\u6587\u4e2d\uff0c\u901a\u8fc7K-Max Pooling\uff0c\u6bcf\u4e2a\u6ee4\u6ce2\u5668\u6700\u5927\u503ck\u4f1a\u88ab\u4fdd\u7559\uff0c\u0109 = c k\u2212max \u3002 5 \u6a21 \u6a21 \u6a21\u578b \u578b \u578b\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u4e0e \u4e0e \u4e0e\u5206 \u5206 \u5206\u7c7b \u7c7b \u7c7b \u57fa\u4e8e\u4ee5\u4e0a\u6b65\u9aa4\uff0c\u5177\u6709\u878d\u5408\u529f\u80fd\u7684Bi-LSTM \u548cCNN\u63d0\u53d6\u51fa\u6e90\u8bed\u53e5\u548c\u76ee\u6807\u53e5\u7684\u8bed\u4e49\u7279\u5f81\uff0c \u5373C S i \uff0cC T i \uff0c\u7136\u540e\u4f7f\u7528\u5143\u7d20\u79ef\u548c\u7edd\u5bf9\u5143\u7d20\u5dee\u6765\u6355\u83b7\u5b83\u4eec\u7684\u5339\u914d\u4fe1\u606f\uff0c\u7136\u540e\u53cd\u9988\u5230\u5168\u8fde\u63a5\u7684\u5c42\u4ee5 \u8bc4\u4f30\u6c49\u8bed-\u8d8a\u5357\u8bed\u53e5\u5bf9\u76f8\u4e92\u7ffb\u8bd1\u7684\u53ef\u80fd\u6027\u5927\u5c0f\u3002\u5177\u4f53\u516c\u5f0f\u5982\u4e0b\uff1a C a i = C S i C T i (11) C a i = |C S i \u2212 C T i | (12) C i = tanh(W a C a i + W b C b i + b) (13) p(y i |c i ) = \u03c3(W c c i + c) (14) L = \u2212 n(1+m) i=1 y i log\u03c3(W c h i + c) \u2212 (1 \u2212 y i )log(1 \u2212 \u03c3(W c h i + c)) (15) \u5176 \u4e2d\u03c3(\u2022)\u662fsigmoid\u6fc0 \u6d3b \u51fd \u6570W a \uff0cW b \uff0cW c \uff0cb\uff0cc\u662f \u6a21 \u578b \u53c2 \u6570 \uff0c \u5176 \u4e2dn\u662f \u6c49 \u8bed \u53e5 \u5b50 \u7684 \u6570 \u91cf\uff0cm\u662f\u5019\u9009\u8d8a\u5357\u8bed\u53e5\u5b50\u7684\u6570\u91cf\u3002\u901a\u8fc7\u6700\u5c0f\u5316\u6807\u8bb0\u7684\u6c49\u8d8a\u53e5\u5bf9\u7684\u4ea4\u53c9\u71b5\u4f5c\u4e3a\u635f\u5931\u51fd\u6570\u6765\u8bad \u7ec3\u6a21\u578b\uff1a\u5bf9\u4e8e\u9884\u6d4b\uff0c\u5982\u679c\u53e5\u5b50\u5bf9\u7684\u6982\u7387\u5927\u4e8e\u6216\u7b49\u4e8e\u8bbe\u7f6e\u7684\u51b3\u7b56\u9608\u503c\u03c1\uff0c\u5219\u5c06\u5176\u5206\u7c7b\u4e3a\u5e73\u884c\uff1b\u5982\u679c \u5c0f\u4e8e\u51b3\u7b56\u9608\u503c\u03c1\uff0c\u5219\u5c06\u5176\u5206\u7c7b\u4e3a\u4e0d\u5e73\u884c\u3002 p( y i ) = 0 if p(y i = 1 | h i ) \u2265 \u03c1, 1 otherwise (16) 6 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u4e0e \u4e0e \u4e0e\u5206 \u5206 \u5206\u6790 \u6790 \u6790 6.1 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u6570 \u6570 \u6570\u636e \u636e \u636e \u6c49 \u6c49 \u6c49\u8d8a \u8d8a \u8d8a\u5e73 \u5e73 \u5e73\u884c \u884c \u884c\u53e5 \u53e5 \u53e5\u5bf9 \u5bf9 \u5bf9 \u6c49 \u6c49 \u6c49\u8d8a \u8d8a \u8d8a\u975e \u975e \u975e\u5e73 \u5e73 \u5e73\u884c \u884c \u884c\u53e5 \u53e5 \u53e5\u5bf9 \u5bf9 \u5bf9 \u8bad\u7ec3\u96c6 130k 130k \u6d4b\u8bd5\u96c6 10k 10k \u9a8c\u8bc1\u96c6 10k 10k", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u672c\u6587\u5c06\u6c49\u8d8a\u5e73\u884c\u53e5\u5bf9\u62bd\u53d6\u95ee\u9898\u8f6c\u5316\u4e3a\u4e8c\u5206\u7c7b\u95ee\u9898\u3002\u7531\u4e8e\u5728\u6c49\u8bed\u5230\u8d8a\u5357\u8bed\u4f4e\u8d44\u6e90\u8bed\u8a00\u4e0a\uff0c\u76ee \u524d\u5c1a\u672a\u627e\u5230\u7528\u4e8e\u8bad\u7ec3\u7684\u516c\u5f00\u6570\u636e\u96c6\uff0c\u6240\u4ee5\u672c\u6587\u5b9e\u9a8c\u6570\u636e\u96c6\u7684\u6765\u6e90\u4e3b\u8981\u662f\u4ece\u6c49\u8d8a\u65b0\u95fb\u7f51\u7ad9\u4e0a\u68c0\u7d22 \u548c\u4ece\u7ef4\u57fa\u767e\u79d1dump\u83b7\u5f97\u7684\u6c49\u8bed-\u8d8a\u5357\u8bed\u7ffb\u8bd1\u6587\u7ae0\uff0c\u8be5\u6587\u7ae0\u7ecf\u8fc7\u5904\u7406\u83b7\u53d6\u6c49\u8bed\u53e5\u5b50315211\uff0c\u8d8a\u5357 \u8bed\u5355\u8bed\u53e5\u5b50316243\uff0c\u624b\u52a8\u5bf9\u9f50\u4ee5\u83b7\u5f9713\u4e07\u4e2a\u6c49\u8d8a\u5e73\u884c\u53e5\u5bf9\uff0c\u4f7f\u7528VecMap\u5de5\u5177\u8bad\u7ec3\u9ad8\u8d28\u91cf\u7684\u8de8\u8bed \u8a00\u53cc\u8bed\u8bcd\u5411\u91cf\u3002\u540c\u65f6\u57fa\u4e8e\u6bcf\u4e2a\u5e73\u884c\u53e5\u5bf9\u7684\u8d1f\u91c7\u6837\u6837\u672c\u6570\u8bbe\u7f6e\u4e3a1:1\uff0c\u968f\u673a\u6784\u9020\u4e8613\u4e07\u4e2a\u6c49\u8d8a\u975e\u5e73 \u884c\u53e5\u5bf9\uff0c\u8bbe\u7f6e\u79cd\u5b50\u8bcd\u5178\u89c4\u6a21\u5927\u5c0f\u8bbe\u7f6e\u4e3a3852\u4e2a\u8bcd\u6761\u3002\u540c\u65f6\u4e3a\u4e86\u8861\u91cf\u672c\u6587\u4e2d\u6c49\u8d8a\u5e73\u884c\u53e5\u5bf9\u62bd\u53d6\u6a21 \u578b\u5206\u7c7b\u5668\u7684\u6027\u80fd\uff0c\u8bbe\u7f6e1\u4e07\u53e5\u6c49\u8d8a\u5e73\u884c\u8bed\u6599\u548c1\u4e07\u53e5\u6c49\u8d8a\u975e\u5e73\u884c\u8bed\u6599\u4f5c\u4e3a\u6d4b\u8bd5\u96c6\u3002\u88681\u4e3a\u672c\u6587\u57fa\u4e8e \u8de8\u8bed\u8a00\u53cc\u8bed\u9884\u8bad\u7ec3\u53caBi-LSTM\u7684\u6c49-\u8d8a\u5e73\u884c\u53e5\u5bf9\u62bd\u53d6\u6a21\u578b\u7684\u8bed\u6599\u89c4\u6a21\u3002 6.2 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u8bbe \u8bbe \u8bbe\u7f6e \u7f6e \u7f6e\u4e0e \u4e0e \u4e0e\u8bc4 \u8bc4 \u8bc4\u4ef7 \u4ef7 \u4ef7\u6307 \u6307 \u6307\u6807 \u6807 \u6807 \u5229\u7528TensorFlow\u7f16\u5199\u5b9e\u73b0\uff0c\u5355\u8bcd\u5d4c\u5165\u7ef4\u5ea6\u548c\u9690\u85cf\u5355\u5143\u6570\u5747\u4e3a300\uff0c\u9690\u85cf\u5c42\u4e3a1\u3002\u6279\u5904\u7406\u5927\u5c0f", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u8bbe\u7f6e\u4e3a64\uff0c\u8bad\u7ec3\u7684epochs\u4e3a15\uff0c\u68af\u5ea6\u622a\u53d6\u8bbe\u7f6e\u4e3a5.0\uff0c\u5b66\u4e60\u7387\u8bbe\u7f6e\u4e3a0.0002\uff0cDropout\u8bbe\u7f6e\u4e3a0.7-0.8\uff0c\u4f7f\u7528Adam\u4f18\u5316\u5668\uff0c\u6fc0\u6d3b\u51fd\u6570\u91c7\u7528sigmoid\u51fd\u6570\uff0c\u635f\u5931\u51fd\u6570\u4e3a\u4ea4\u53c9\u71b5\u635f\u5931\u51fd\u6570\u3002\u5728\u8bc4\u4f30\u6307\u6807 \u65b9\u9762\uff0c\u4f7f\u7528\"\u7cbe\u5ea6\"\uff0c\"\u53ec\u56de\u7387\"\u548c\" F1\u503c\"\u4f5c\u4e3a\u8861\u91cf\u6a21\u578b\u662f\u5426\u53ef\u4ee5\u6b63\u786e\u5206\u7c7b\u6c49\u8bed-\u8d8a\u5357\u8bed\u662f\u5426\u4e3a\u5e73 \u884c\u53e5\u5b50\u7684\u6307\u6807\u3002\u5176\u4e2d\u7cbe\u5ea6\u662f\u6240\u6709\u63d0\u53d6\u7684\u53e5\u5b50\u5bf9\u4e2d\u771f\u6b63\u5e73\u884c\u53e5\u5b50\u5bf9\u7684\u6bd4\u4f8b\uff1b\u53ec\u56de\u7387\u662f\u6d4b\u8bd5\u96c6\u4e2d\u6240 \u6709\u5e73\u884c\u53e5\u5b50\u5bf9\u4e2d\u771f\u6b63\u5e73\u884c\u63d0\u53d6\u7684\u53e5\u5b50\u5bf9\u7684\u6bd4\u4f8b\uff1bF1\u503c\u662f\u7cbe\u5ea6\u548c\u53ec\u56de\u7387\u7684\u8c03\u548c\u5e73\u5747\u503c\u3002\u5177\u4f53\u516c\u5f0f \u5982\u4e0b\uff1a ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "P recision = |T P | |T P + F P | (17) Recall = |T P | |T P + F N | (18) F 1 = 2 \u00d7 precision \u00d7 recall precision + recall \u00d7 100 (19) \u5176 \u4e2d \uff0cT P \u662f \u63d0 \u53d6 \u53e5 \u5b50 \u4e2d \u771f \u6b63 \u5e73 \u884c \u7684 \u53e5 \u5bf9 \u7684 \u6570 \u91cf \uff0cF P \u662f \u63d0 \u53d6 \u53e5 \u5b50 \u4e2d \u975e \u5e73 \u884c \u53e5 \u5bf9 \u7684 \u6570 \u91cf\uff0cF N \u662f\u6d4b\u8bd5\u96c6\u4e2d\u672a\u88ab\u63d0\u53d6\u7684\u5e73\u884c\u53e5\u5bf9\u7684\u6570\u91cf\u3002 6.3 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u7ed3 \u7ed3 \u7ed3\u679c \u679c \u679c\u4e0e \u4e0e \u4e0e\u5206 \u5206 \u5206\u6790 \u6790 \u6790 \u8de8\u8bed\u8a00\u53cc\u8bed\u9884\u8bad\u7ec3\u4f7f\u7528Artetxe\u7b49\u4eba(Artetxe et al., 2017)\u63d0\u51fa\u7684VecMap\u5f00\u6e90\u6846\u67b6\u5bf9\u52a0\u5f3a\u6c49 \u8d8a\u8bed\u8a00\u76f8\u5173\u6027\u53caBi-LSTM\u548cCNN\u66f4\u597d\u5730\u6355\u83b7\u53d6\u53e5\u5b50\u4e0a\u4e0b\u6587\u4fe1\u606f\u548c\u5c40\u90e8\u4fe1\u606f\uff0c\u8bbe\u8ba1\u4e86\u4ee5\u4e0b\u4e09\u7ec4\u5bf9 \u6bd4\u5b9e\u9a8c\uff0c\u518d\u901a\u8fc7\u4e0a\u9762\u7684\u8bc4\u4ef7\u6307\u6807\u8fdb\u884c\u5b9e\u9a8c\u8bc4\u4ef7\u4e0e\u5206\u6790\u3002 \u5b9e \u9a8c \u4e00 \uff1a \u4e3a \u4e86 \u9a8c \u8bc1 \u9884 \u8bad \u7ec3 \u65b9 \u6cd5 \u7684 \u6709 \u6548 \u6027 \uff0c \u8bbe \u7f6e \u9608 \u503c \u4e3a0.90\uff0c \u5c06 \u7ecf \u8fc7 \u9884 \u8bad \u7ec3 \u7684Bi- LSTM\u548cCNN\u6c49\u8d8a\u5e73\u884c\u53e5\u5bf9\u62bd\u53d6\u6a21\u578b\u4e0e\u4e0d\u7ecf\u8fc7\u9884\u8bad\u7ec3\u7684\u6548\u679c\u8fdb\u884c\u5bf9\u6bd4\u3002\u6211\u4eec\u8fd8\u5c06\u4ec5\u4f7f\u7528Bi- LSTM\u62bd\u53d6\u6c49\u8d8a\u53cc\u8bed\u5e73\u884c\u53e5\u5bf9\u7684\u57fa\u7ebf\u65b9\u6cd5\u8fdb\u884c\u6bd4\u8f83\uff0c\u540c\u65f6\uff0c\u4e3a\u4e86\u7a81\u51fa\u5206\u7c7b\u5668\u6784\u9020\u6bd4\u4f20\u7edf\u673a\u5668 \u5b66\u4e60\u66f4\u6df1\u5165\u5b66\u4e60\u5177\u6709\u66f4\u597d\u7684\u51c6\u786e\u6027\uff0c\u540c\u65f6\u8fd8\u6bd4\u8f83\u4e86Munteanu D", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Learning principled bilingual mappings of word embeddings while preserving monolingual invariance", |
| "authors": [ |
| { |
| "first": "Mikel", |
| "middle": [], |
| "last": "Artetxe", |
| "suffix": "" |
| }, |
| { |
| "first": "Gorka", |
| "middle": [], |
| "last": "Labaka", |
| "suffix": "" |
| }, |
| { |
| "first": "Eneko", |
| "middle": [], |
| "last": "Agirre", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. 2016. Learning principled bilingual mappings of word embeddings while preserving monolingual invariance. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Learning bilingual word embeddings with (almost) no bilingual data", |
| "authors": [ |
| { |
| "first": "Mikel", |
| "middle": [], |
| "last": "Artetxe", |
| "suffix": "" |
| }, |
| { |
| "first": "Gorka", |
| "middle": [], |
| "last": "Labaka", |
| "suffix": "" |
| }, |
| { |
| "first": "Eneko", |
| "middle": [], |
| "last": "Agirre", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics", |
| "volume": "1", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. 2017. Learning bilingual word embeddings with (almost) no bilingual data. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Generalizing and improving bilingual word embedding mappings with a multi-step framework of linear transformations", |
| "authors": [ |
| { |
| "first": "Mikel", |
| "middle": [], |
| "last": "Artetxe", |
| "suffix": "" |
| }, |
| { |
| "first": "Gorka", |
| "middle": [], |
| "last": "Labaka", |
| "suffix": "" |
| }, |
| { |
| "first": "Eneko", |
| "middle": [], |
| "last": "Agirre", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Thirty-Second AAAI Conference on Artificial Intelligence", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. 2018. Generalizing and improving bilingual word embedding mappings with a multi-step framework of linear transformations. In Thirty-Second AAAI Conference on Artificial Intelligence.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Extracting parallel sentences from comparable corpora with stacc variants", |
| "authors": [ |
| { |
| "first": "Andoni", |
| "middle": [], |
| "last": "Azpeitia", |
| "suffix": "" |
| }, |
| { |
| "first": "Eva Mart\u0131nez", |
| "middle": [], |
| "last": "Thierry Etchegoyhen", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Garcia", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Proceedings of the 11th Workshop on Building and Using Comparable Corpora", |
| "volume": "", |
| "issue": "", |
| "pages": "48--52", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Andoni Azpeitia, Thierry Etchegoyhen, and Eva Mart\u0131nez Garcia. 2018. Extracting parallel sentences from comparable corpora with stacc variants. In Proceedings of the 11th Workshop on Building and Using Comparable Corpora, pages 48-52.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "A factory of comparable corpora from wikipedia", |
| "authors": [ |
| { |
| "first": "Alberto", |
| "middle": [], |
| "last": "Barr\u00f3n-Cede\u00f1o", |
| "suffix": "" |
| }, |
| { |
| "first": "Cristina", |
| "middle": [], |
| "last": "Espa\u00f1a-Bonet", |
| "suffix": "" |
| }, |
| { |
| "first": "Josu", |
| "middle": [], |
| "last": "Boldoba", |
| "suffix": "" |
| }, |
| { |
| "first": "Llu\u00eds", |
| "middle": [], |
| "last": "M\u00e0rquez", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Eighth Workshop on Building And Using Comparable Corpora", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Alberto Barr\u00f3n-Cede\u00f1o, Cristina Espa\u00f1a-Bonet, Josu Boldoba, and Llu\u00eds M\u00e0rquez. 2015. A factory of comparable corpora from wikipedia. In Eighth Workshop on Building And Using Comparable Corpora.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "H2@ bucc18: Parallel sentence extraction from comparable corpora using multilingual sentence embeddings", |
| "authors": [ |
| { |
| "first": "Houda", |
| "middle": [], |
| "last": "Bouamor", |
| "suffix": "" |
| }, |
| { |
| "first": "Hassan", |
| "middle": [], |
| "last": "Sajjad", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Proc. Workshop on Building and Using Comparable Corpora", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Houda Bouamor and Hassan Sajjad. 2018. H2@ bucc18: Parallel sentence extraction from compara- ble corpora using multilingual sentence embeddings. In Proc. Workshop on Building and Using Comparable Corpora.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Automatically extracting parallel sentences from wikipedia using sequential matching of language resources", |
| "authors": [ |
| { |
| "first": "Juryong", |
| "middle": [], |
| "last": "Cheon", |
| "suffix": "" |
| }, |
| { |
| "first": "K", |
| "middle": [ |
| "O" |
| ], |
| "last": "Youngjoong", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Ieice Transactions on Information And Systems", |
| "volume": "100", |
| "issue": "2", |
| "pages": "405--408", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Juryong Cheon and K. O. Youngjoong. 2017. Automatically extracting parallel sentences from wikipedia using sequential matching of language resources. Ieice Transactions on Information And Systems, E100.D(2):405-408.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Neural machine translation for english-tamil", |
| "authors": [ |
| { |
| "first": "Himanshu", |
| "middle": [], |
| "last": "Choudhary", |
| "suffix": "" |
| }, |
| { |
| "first": "Aditya", |
| "middle": [ |
| "Kumar" |
| ], |
| "last": "Pathak", |
| "suffix": "" |
| }, |
| { |
| "first": "Rajiv Ratan", |
| "middle": [], |
| "last": "Saha", |
| "suffix": "" |
| }, |
| { |
| "first": "Ponnurangam", |
| "middle": [], |
| "last": "Kumaraguru", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Proceedings of the Third Conference on Machine Translation: Shared Task Papers", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Himanshu Choudhary, Aditya Kumar Pathak, Rajiv Ratan Saha, and Ponnurangam Kumaraguru. 2018. Neural machine translation for english-tamil. In Proceedings of the Third Conference on Machine Translation: Shared Task Papers.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Parallel sentence extraction from comparable corpora with neural network features", |
| "authors": [ |
| { |
| "first": "Chenhui", |
| "middle": [], |
| "last": "Chu", |
| "suffix": "" |
| }, |
| { |
| "first": "Raj", |
| "middle": [], |
| "last": "Dabre", |
| "suffix": "" |
| }, |
| { |
| "first": "Sadao", |
| "middle": [], |
| "last": "Kurohashi", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)", |
| "volume": "", |
| "issue": "", |
| "pages": "2931--2935", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Chenhui Chu, Raj Dabre, and Sadao Kurohashi. 2016. Parallel sentence extraction from comparable corpora with neural network features. In Proceedings of the Tenth International Conference on Lan- guage Resources and Evaluation (LREC'16), pages 2931-2935, Portoro\u017e, Slovenia, May. European Language Resources Association (ELRA).", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Bilingual sentence alignment based on punctuation statistics and lexicon", |
| "authors": [ |
| { |
| "first": "C", |
| "middle": [], |
| "last": "Thomas", |
| "suffix": "" |
| }, |
| { |
| "first": "Jian", |
| "middle": [ |
| "Cheng" |
| ], |
| "last": "Chuang", |
| "suffix": "" |
| }, |
| { |
| "first": "Tracy", |
| "middle": [], |
| "last": "Wu", |
| "suffix": "" |
| }, |
| { |
| "first": "Wen", |
| "middle": [ |
| "Chie" |
| ], |
| "last": "Lin", |
| "suffix": "" |
| }, |
| { |
| "first": "Jason", |
| "middle": [ |
| "S" |
| ], |
| "last": "Shei", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Chang", |
| "suffix": "" |
| } |
| ], |
| "year": 2004, |
| "venue": "Natural Language Processingijcnlp, First International Joint Conference", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Thomas C. Chuang, Jian Cheng Wu, Tracy Lin, Wen Chie Shei, and Jason S. Chang. 2004. Bilingual sentence alignment based on punctuation statistics and lexicon. In Natural Language Processing- ijcnlp, First International Joint Conference, Hainan Island, China, March, Revised Selected Papers.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "An empirical analysis of nmt-derived interlingual embeddings and their use in parallel sentence identification", |
| "authors": [ |
| { |
| "first": "Cristina", |
| "middle": [], |
| "last": "Espa\u00f1a-Bonet", |
| "suffix": "" |
| }, |
| { |
| "first": "\u00c1d\u00e1m", |
| "middle": [ |
| "Csaba" |
| ], |
| "last": "Varga", |
| "suffix": "" |
| }, |
| { |
| "first": "Alberto", |
| "middle": [], |
| "last": "Barr\u00f3n-Cede\u00f1o", |
| "suffix": "" |
| }, |
| { |
| "first": "Josef", |
| "middle": [], |
| "last": "Van Genabith", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "IEEE Journal of Selected Topics in Signal Processing", |
| "volume": "11", |
| "issue": "8", |
| "pages": "1340--1350", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Cristina Espa\u00f1a-Bonet,\u00c1d\u00e1m Csaba Varga, Alberto Barr\u00f3n-Cede\u00f1o, and Josef Van Genabith. 2017. An empirical analysis of nmt-derived interlingual embeddings and their use in parallel sentence identification. IEEE Journal of Selected Topics in Signal Processing, 11(8):1340-1350.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "A program for aligning sentences in bilingual corpora", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "William", |
| "suffix": "" |
| }, |
| { |
| "first": "Kenneth", |
| "middle": [ |
| "Ward" |
| ], |
| "last": "Gale", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Church", |
| "suffix": "" |
| } |
| ], |
| "year": 1991, |
| "venue": "Meeting of the Association for Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "William A. Gale and Kenneth Ward Church. 1991. A program for aligning sentences in bilingual corpora. In Meeting of the Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Bilingual word embeddings with bucketed CNN for parallel sentence extraction", |
| "authors": [ |
| { |
| "first": "Jeenu", |
| "middle": [], |
| "last": "Grover", |
| "suffix": "" |
| }, |
| { |
| "first": "Pabitra", |
| "middle": [], |
| "last": "Mitra", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Proceedings of ACL 2017", |
| "volume": "", |
| "issue": "", |
| "pages": "11--16", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jeenu Grover and Pabitra Mitra. 2017. Bilingual word embeddings with bucketed CNN for parallel sen- tence extraction. In Proceedings of ACL 2017, Student Research Workshop, pages 11-16, Vancouver, Canada, July. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "A deep neural network approach to parallel sentence extraction", |
| "authors": [ |
| { |
| "first": "Francis", |
| "middle": [], |
| "last": "Gr\u00e9goire", |
| "suffix": "" |
| }, |
| { |
| "first": "Philippe", |
| "middle": [], |
| "last": "Langlais", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Francis Gr\u00e9goire and Philippe Langlais. 2017. A deep neural network approach to parallel sentence extraction.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "Extracting parallel sentences with bidirectional recurrent neural networks to improve machine translation", |
| "authors": [ |
| { |
| "first": "Francis", |
| "middle": [], |
| "last": "Gr\u00e9goire", |
| "suffix": "" |
| }, |
| { |
| "first": "Philippe", |
| "middle": [], |
| "last": "Langlais", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Francis Gr\u00e9goire and Philippe Langlais. 2018. Extracting parallel sentences with bidirectional recurrent neural networks to improve machine translation.", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Bilingual word representations with monolingual quality in mind", |
| "authors": [ |
| { |
| "first": "Thang", |
| "middle": [], |
| "last": "Luong", |
| "suffix": "" |
| }, |
| { |
| "first": "Hieu", |
| "middle": [], |
| "last": "Pham", |
| "suffix": "" |
| }, |
| { |
| "first": "Christopher", |
| "middle": [ |
| "D" |
| ], |
| "last": "Manning", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of the 1st Workshop on Vector Space Modeling for Natural Language Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Thang Luong, Hieu Pham, and Christopher D. Manning. 2015. Bilingual word representations with monolingual quality in mind. In Proceedings of the 1st Workshop on Vector Space Modeling for Natural Language Processing.", |
| "links": null |
| }, |
| "BIBREF16": { |
| "ref_id": "b16", |
| "title": "Efficient extraction of pseudo-parallel sentences from raw monolingual data using word embeddings", |
| "authors": [ |
| { |
| "first": "Benjamin", |
| "middle": [], |
| "last": "Marie", |
| "suffix": "" |
| }, |
| { |
| "first": "Atsushi", |
| "middle": [], |
| "last": "Fujita", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics", |
| "volume": "2", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Benjamin Marie and Atsushi Fujita. 2017. Efficient extraction of pseudo-parallel sentences from raw monolingual data using word embeddings. In Proceedings of the 55th Annual Meeting of the Asso- ciation for Computational Linguistics (Volume 2: Short Papers).", |
| "links": null |
| }, |
| "BIBREF17": { |
| "ref_id": "b17", |
| "title": "A comparison of pivot methods for phrase-based statistical machine translation", |
| "authors": [ |
| { |
| "first": "Hitoshi Isahara Masao", |
| "middle": [], |
| "last": "Utiyama", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "Human Language Technology Conference of the North American Chapter of the Association of Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hitoshi Isahara Masao Utiyama. 2013. A comparison of pivot methods for phrase-based statistical machine translation. In Human Language Technology Conference of the North American Chapter of the Association of Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF18": { |
| "ref_id": "b18", |
| "title": "Efficient estimation of word representations in vector space", |
| "authors": [ |
| { |
| "first": "Tomas", |
| "middle": [], |
| "last": "Mikolov", |
| "suffix": "" |
| }, |
| { |
| "first": "Kai", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "Greg", |
| "middle": [], |
| "last": "Corrado", |
| "suffix": "" |
| }, |
| { |
| "first": "Jeffrey", |
| "middle": [], |
| "last": "Dean", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "Computer Science", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient estimation of word repre- sentations in vector space. Computer Science.", |
| "links": null |
| }, |
| "BIBREF19": { |
| "ref_id": "b19", |
| "title": "Improving machine translation performance by exploiting non-parallel corpora", |
| "authors": [ |
| { |
| "first": "Stefan", |
| "middle": [], |
| "last": "Dragos", |
| "suffix": "" |
| }, |
| { |
| "first": "Daniel", |
| "middle": [], |
| "last": "Munteanu", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Marcu", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Computational Linguistics", |
| "volume": "31", |
| "issue": "4", |
| "pages": "477--504", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Dragos Stefan Munteanu and Daniel Marcu. 2005. Improving machine translation performance by exploiting non-parallel corpora. Computational Linguistics, 31(4):477-504.", |
| "links": null |
| }, |
| "BIBREF20": { |
| "ref_id": "b20", |
| "title": "Fast-champollion: A fast and robust sentence alignment algorithm", |
| "authors": [ |
| { |
| "first": "Li", |
| "middle": [], |
| "last": "Peng", |
| "suffix": "" |
| }, |
| { |
| "first": "Maosong", |
| "middle": [], |
| "last": "Sun", |
| "suffix": "" |
| }, |
| { |
| "first": "Xue", |
| "middle": [], |
| "last": "Ping", |
| "suffix": "" |
| } |
| ], |
| "year": 2010, |
| "venue": "COLING 2010, 23rd International Conference on Computational Linguistics, Posters Volume", |
| "volume": "", |
| "issue": "", |
| "pages": "23--27", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Li Peng, Maosong Sun, and Xue Ping. 2010. Fast-champollion: A fast and robust sentence alignment algorithm. In COLING 2010, 23rd International Conference on Computational Linguistics, Posters Volume, 23-27 August 2010, Beijing, China.", |
| "links": null |
| }, |
| "BIBREF21": { |
| "ref_id": "b21", |
| "title": "Parallel sentence generation from comparable corpora for improved smt", |
| "authors": [ |
| { |
| "first": "Abdul", |
| "middle": [], |
| "last": "Sadaf", |
| "suffix": "" |
| }, |
| { |
| "first": "Holger", |
| "middle": [], |
| "last": "Rauf", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Schwenk", |
| "suffix": "" |
| } |
| ], |
| "year": 2011, |
| "venue": "Machine Translation", |
| "volume": "25", |
| "issue": "4", |
| "pages": "341--375", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Sadaf Abdul Rauf and Holger Schwenk. 2011. Parallel sentence generation from comparable corpora for improved smt. Machine Translation, 25(4):p.341-375.", |
| "links": null |
| }, |
| "BIBREF22": { |
| "ref_id": "b22", |
| "title": "A simple sentence-level extraction algorithm for comparable data", |
| "authors": [ |
| { |
| "first": "Christoph", |
| "middle": [], |
| "last": "Tillmann", |
| "suffix": "" |
| }, |
| { |
| "first": "Jian-Ming", |
| "middle": [], |
| "last": "Xu", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "Proceedings of Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Companion Volume: Short Papers", |
| "volume": "", |
| "issue": "", |
| "pages": "93--96", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Christoph Tillmann and Jian-ming Xu. 2009. A simple sentence-level extraction algorithm for comparable data. In Proceedings of Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Companion Volume: Short Papers, pages 93-96, Boulder, Colorado, June. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF23": { |
| "ref_id": "b23", |
| "title": "Attention is all you need", |
| "authors": [ |
| { |
| "first": "Ashish", |
| "middle": [], |
| "last": "Vaswani", |
| "suffix": "" |
| }, |
| { |
| "first": "Noam", |
| "middle": [], |
| "last": "Shazeer", |
| "suffix": "" |
| }, |
| { |
| "first": "Niki", |
| "middle": [], |
| "last": "Parmar", |
| "suffix": "" |
| }, |
| { |
| "first": "Jakob", |
| "middle": [], |
| "last": "Uszkoreit", |
| "suffix": "" |
| }, |
| { |
| "first": "Llion", |
| "middle": [], |
| "last": "Jones", |
| "suffix": "" |
| }, |
| { |
| "first": "Aidan", |
| "middle": [ |
| "N" |
| ], |
| "last": "Gomez", |
| "suffix": "" |
| }, |
| { |
| "first": "Lukasz", |
| "middle": [], |
| "last": "Kaiser", |
| "suffix": "" |
| }, |
| { |
| "first": "Illia", |
| "middle": [], |
| "last": "Polosukhin", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Advances in neural information processing systems", |
| "volume": "", |
| "issue": "", |
| "pages": "5998--6008", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in neural information processing systems, pages 5998-6008.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "TABREF1": { |
| "text": "\u5b9e\u9a8c\u6570\u636e\u96c6\u8868", |
| "html": null, |
| "content": "<table/>", |
| "type_str": "table", |
| "num": null |
| }, |
| "TABREF3": { |
| "text": "", |
| "html": null, |
| "content": "<table><tr><td/><td colspan=\"2\">: \u4e0d\u540c\u6a21\u578b\u5bf9\u6bd4\u5b9e\u9a8c\u7ed3\u679c</td><td/></tr><tr><td>\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5</td><td>R(%)</td><td>P(%)</td><td>F1(%)</td></tr><tr><td>word2vec-</td><td>72.8%</td><td>64.2%</td><td>68.2%</td></tr><tr><td>BiLSTM-</td><td/><td/><td/></tr><tr><td>CNN</td><td/><td/><td/></tr><tr><td>\u672c\u6587\u6cd5\u65b9\u6cd5</td><td>75.6%</td><td>66.5%</td><td>70.7%</td></tr><tr><td>(VecMap)</td><td/><td/><td/></tr></table>", |
| "type_str": "table", |
| "num": null |
| }, |
| "TABREF4": { |
| "text": "\u4e0d\u540c\u8bcd\u5411\u91cf\u8868\u5f81\u65b9\u6cd5\u5bf9\u6bd4\u5b9e\u9a8c\u7ed3\u679c", |
| "html": null, |
| "content": "<table><tr><td colspan=\"4\">\u5b9e\u9a8c\u4e09\uff1a\u4e3a\u4e86\u9a8c\u8bc1\u9009\u53d6\u4e0d\u540c\u9608\u503c\u65f6\u662f\u5426\u4f1a\u5f71\u54cd\u6a21\u578b\u7684\u6548\u679c\uff0c\u4e3a\u62bd\u53d6\u5230\u66f4\u9ad8\u8d28\u91cf\u7684\u6c49\u8d8a\u53cc\u8bed</td></tr><tr><td colspan=\"4\">\u5e73\u884c\u53e5\u5bf9\u63d0\u4f9b\u9608\u503c\u53c2\u6570\u57fa\u7840\uff0c\u8bbe\u7f6e\u4e86\u5728\u672c\u6587\u63d0\u51fa\u65b9\u6cd5\u4e0a\u4e0d\u540c\u9608\u503c\u7684\u5bf9\u6bd4\u5b9e\u9a8c\uff0c\u5b9e\u9a8c\u7ed3\u679c\u5982\u88684\u6240</td></tr><tr><td>\u793a\u3002</td><td/><td/><td/></tr><tr><td>\u4e0d \u4e0d \u4e0d\u540c \u540c \u540c\u7684 \u7684 \u7684\u9608 \u9608 \u9608\u503c \u503c \u503cM</td><td>R(%)</td><td>P(%)</td><td>F1(%)</td></tr><tr><td>M=0.8</td><td>77.3%</td><td>68.6%</td><td>72.7%</td></tr><tr><td>M=0.85</td><td>75.6%</td><td>66.5%</td><td>70.7%</td></tr><tr><td>M=0.90</td><td>73.9%</td><td>64.7%</td><td>68.9%</td></tr></table>", |
| "type_str": "table", |
| "num": null |
| }, |
| "TABREF5": { |
| "text": "", |
| "html": null, |
| "content": "<table><tr><td>: \u4e0d\u540c\u9608\u503c\u5bf9\u6bd4\u5b9e\u9a8c\u7ed3\u679c</td></tr><tr><td>\u4ece\u88684\u4e2d\u53ef\u4ee5\u770b\u51fa\uff0c\u4e0d\u540c\u7684\u9608\u503cM\u5bf9\u5b9e\u9a8c\u7ed3\u679c\u7684\u5f71\u54cd\u3002\u5176\u4e2d\uff0c\u5b9e\u9a8c\u8bbe\u7f6e\u9608\u503c\u53c2\u6570\u8d8a\u5927\uff0c\u62bd\u53d6</td></tr><tr><td>\u6c49\u8d8a\u53cc\u8bed\u5e73\u884c\u53e5\u5bf9\u7684F1\u5206\u503c\u53cd\u800c\u8d8a\u4f4e\u3002\u9608\u503cM\u4f5c\u4e3a\u6c49\u8d8a\u53cc\u8bed\u5e73\u884c\u53e5\u5bf9\u62bd\u53d6\u7684\u5224\u522b\u503c\u3002</td></tr><tr><td>\u5b9e\u9a8c\u56db\uff1a\u4e3a\u4e86\u9a8c\u8bc1\u672c\u6587\u65b9\u6cd5\u62bd\u53d6\u51fa\u7684\u6c49-\u8d8a\u5e73\u884c\u53e5\u5bf9\u5bf9\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u6a21\u578b\u6027\u80fd\u7684\u5f71\u54cd\u3002\u672c</td></tr><tr><td>\u6587\u9009\u62e9\u4e86\u76ee\u524d\u6bd4\u8f83\u4e3b\u6d41\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578bSeq2seq+Attention(Vaswani et al., 2017)\u4f5c\u4e3a\u673a\u5668\u7ffb\u8bd1</td></tr><tr><td>\u6a21\u578b\uff0c\u7f16\u7801\u5668\u548c\u89e3\u7801\u5668\u7684\u5355\u8bcd\u5d4c\u5165\u548c\u5faa\u73af\u72b6\u6001\u7684\u7ef4\u5ea6\u90fd\u8bbe\u7f6e\u4e3a512\uff0c\u8bad\u7ec320\u4e2aepochs\uff0c\u5176\u4e2d\u53e5\u5bf9</td></tr><tr><td>\u7684\u6279\u91cf\u5927\u5c0f\u4e3a64\u3002\u6211\u4eec\u6311\u9009\u4e8610\u4e07\u6761\u6c49-\u8d8a\u5e73\u884c\u53e5\u5bf9\u4f5c\u4e3a\u57fa\u7840\u8bad\u7ec3\u96c6\uff0c\u5e76\u4e0e\u6dfb\u52a0\u4e86\u4ece\u53ef\u6bd4\u8bed\u6599\u4e2d</td></tr><tr><td>\u62bd\u53d6\u76845\u4e07\u5e73\u884c\u53e5\u5bf9\u4f5c\u5bf9\u6bd4\uff0c\u88685\u663e\u793a\u4e86\u4e0d\u540c\u89c4\u6a21\u6570\u636e\u7684BLEU\u5f97\u5206\u3002\u4ece\u88685\u53ef\u4ee5\u770b\u51fa\uff0c\u5728\u8bad\u7ec3\u96c6</td></tr><tr><td>\u4e2d\u6dfb\u52a0\u672c\u6587\u7cfb\u7edf\u62bd\u53d6\u5230\u76845\u4e07\u5e73\u884c\u53e5\u5bf9\u540e\uff0c\u7ffb\u8bd1\u7cfb\u7edf\u7684BLEU\u5f97\u5206\u5206\u4e3a15.89\uff0c\u63d0\u9ad8\u4e860.34\uff0c\u4f18\u4e8e</td></tr><tr><td>\u76f4\u63a5\u5229\u752810\u4e07\u5e73\u884c\u53e5\u5bf9\u8bad\u7ec3\u7684\u7ffb\u8bd1\u6a21\u578b\u3002\u5b9e\u9a8c\u7ed3\u679c\u8bc1\u5b9e\u4e86\u672c\u6587\u6a21\u578b\u62bd\u53d6\u5230\u7684\u6c49\u8d8a\u5e73\u884c\u53e5\u5bf9\u7684\u8d28</td></tr><tr><td>\u91cf\uff0c\u8868\u660e\u4e86\u53ef\u6bd4\u8bed\u6599\u5e93\u4e2d\u5b58\u5728\u5927\u91cf\u8bed\u4e49\u7a7a\u95f4\u76f8\u8fd1\u7684\u6c49\u8d8a\u5e73\u884c\u53e5\u5bf9\u3002</td></tr></table>", |
| "type_str": "table", |
| "num": null |
| }, |
| "TABREF6": { |
| "text": "\u5e73\u884c\u53e5\u5bf9\u5bf9\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u6027\u80fd\u7684\u5f71\u54cd7 \u7ed3 \u7ed3 \u7ed3\u8bba \u8bba \u8bba", |
| "html": null, |
| "content": "<table><tr><td>\u9488\u5bf9\u6c49-\u8d8a\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u6570\u636e\u7a00\u7f3a\u7684\u95ee\u9898\uff0c\u672c\u6587\u63d0\u51fa\u4e86\u4e00\u79cd\u57fa\u4e8e\u8de8\u8bed\u8a00\u9884\u8bad\u7ec3\u53caBi-LSTM\u65b9</td></tr><tr><td>\u6cd5\u62bd\u53d6\u6c49\u8d8a\u53cc\u8bed\u5e73\u884c\u53e5\u5bf9\u3002\u5728\u6ca1\u6709\u5927\u89c4\u6a21\u6c49\u8d8a\u5e73\u884c\u8bed\u6599\u60c5\u51b5\u4e0b\uff0c\u8be5\u65b9\u6cd5\u5229\u7528\u6c49\u8d8a\u79cd\u5b50\u8bcd\u5178\u8fdb</td></tr><tr><td>\u884c\u6c49\u8d8a\u8de8\u8bed\u8a00\u9884\u8bad\u7ec3\uff0c\u5c06\u6c49\u8d8a\u53cc\u8bed\u8868\u5f81\u5230\u540c\u4e00\u8bed\u4e49\u7a7a\u95f4\u4e2d\uff0c\u5b9e\u73b0\u8bed\u4e49\u5bf9\u9f50\u3002\u5229\u7528\u6df1\u5ea6\u795e\u7ecf\u7f51</td></tr><tr><td>\u7edcBi-LSTM\u548cCNN\u5206\u522b\u63d0\u8d77\u6c49\u8d8a\u53e5\u5bf9\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\u548c\u5c40\u90e8\u4fe1\u606f\u4ece\u800c\u62bd\u53d6\u51fa\u5339\u914d\u5ea6\u66f4\u9ad8\uff0c\u566a\u58f0\u66f4</td></tr><tr><td>\u5c0f\u7684\u6c49\u8d8a\u53cc\u8bed\u5e73\u884c\u53e5\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u7ecf\u8fc7\u8de8\u8bed\u8a00\u9884\u8bad\u7ec3\u7684\u5e73\u884c\u53e5\u63d0\u53d6\u65b9\u6cd5\u5728\u51c6\u786e\u7387\u548c\u53ec\u56de\u7387</td></tr><tr><td>\u4e0a\u9ad8\u4e8e\u57fa\u7ebf\u6a21\u578b\uff0c\u5e76\u4e14\u62bd\u53d6\u5230\u7684\u6c49-\u8d8a\u5e73\u884c\u53e5\u7684\u8bed\u4e49\u66f4\u8fd1\u3002\u5728\u672a\u6765\u7684\u5de5\u4f5c\u4e2d\uff0c\u6211\u4eec\u4f1a\u63a2\u7d22\u8be5\u6a21\u578b</td></tr><tr><td>\u7528\u4e8e\u591a\u8bed\u8a00\u5e73\u884c\u53e5\u5bf9\u62bd\u53d6\u4e14\u5728\u673a\u5668\u7ffb\u8bd1\u7684\u6548\u679c\u3002</td></tr></table>", |
| "type_str": "table", |
| "num": null |
| } |
| } |
| } |
| } |
