| { | |
| "paper_id": "2020", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T12:52:35.659310Z" | |
| }, | |
| "title": "Chinese Semantic Composition Model with Dependency Parsing", | |
| "authors": [ | |
| { | |
| "first": "Yuanmeng", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Beijing Jiaotong University", | |
| "location": { | |
| "postCode": "10004", | |
| "settlement": "Beijing" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Yujie", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Beijing Jiaotong University", | |
| "location": { | |
| "postCode": "10004", | |
| "settlement": "Beijing" | |
| } | |
| }, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "Jinan", | |
| "middle": [], | |
| "last": "Xu", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Beijing Jiaotong University", | |
| "location": { | |
| "postCode": "10004", | |
| "settlement": "Beijing" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Yufeng", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Beijing Jiaotong University", | |
| "location": { | |
| "postCode": "10004", | |
| "settlement": "Beijing" | |
| } | |
| }, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "In the semantic composition methods, the structural methods emphasize the combination mode of words' meaning representation guided by structural information. Existing structural semantic composition methods use external parser to obtain syntactic structure information, resulting in the separation of syntactic parsing and semantic composition. The accuracy of syntactic analysis will severely restrict the performance of semantic composition models, and the inconsistent training data fields will further aggravate the performance degradation. To solve this problem, this paper proposes a semantic composition model combined with dependency parsing. On the one hand, the dependency model is fine-tuned when training the semantic composition model, so that it can be more suitable for the domain characteristics of the training data used \u56fd\u5bb6\u81ea\u7136\u79d1\u5b66\u57fa\u91d1(61876198,61976015,61976016)\u8d44\u52a9 \u8ba1\u7b97\u8bed\u8a00\u5b66 by the semantic composition model. On the other hand, we add the intermediate information representation of dependency to the semantic composition part to obtain more abundant structural information and semantic information, so as to reduce the sensitivity of semantic composition model to erroneous results of dependency parsing and improve the robustness of the model. We take Chinese as the specific research object, apply semantic combination model to retelling recognition task, and verify the model proposed in this paper on CTB5 Chinese dependency parsing data and LCQMC Chinese retelling recognition data. The experimental results show that the prediction accuracy and F1 value of the method proposed in this paper reach 76.81% and 78.03% respectively in retelling recognition tasks. We further designed experiments to verify the effectiveness of joint learning and intermediate information utilization, and made comparative analysis with relevant representative work.", | |
| "pdf_parse": { | |
| "paper_id": "2020", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "In the semantic composition methods, the structural methods emphasize the combination mode of words' meaning representation guided by structural information. Existing structural semantic composition methods use external parser to obtain syntactic structure information, resulting in the separation of syntactic parsing and semantic composition. The accuracy of syntactic analysis will severely restrict the performance of semantic composition models, and the inconsistent training data fields will further aggravate the performance degradation. To solve this problem, this paper proposes a semantic composition model combined with dependency parsing. On the one hand, the dependency model is fine-tuned when training the semantic composition model, so that it can be more suitable for the domain characteristics of the training data used \u56fd\u5bb6\u81ea\u7136\u79d1\u5b66\u57fa\u91d1(61876198,61976015,61976016)\u8d44\u52a9 \u8ba1\u7b97\u8bed\u8a00\u5b66 by the semantic composition model. On the other hand, we add the intermediate information representation of dependency to the semantic composition part to obtain more abundant structural information and semantic information, so as to reduce the sensitivity of semantic composition model to erroneous results of dependency parsing and improve the robustness of the model. We take Chinese as the specific research object, apply semantic combination model to retelling recognition task, and verify the model proposed in this paper on CTB5 Chinese dependency parsing data and LCQMC Chinese retelling recognition data. The experimental results show that the prediction accuracy and F1 value of the method proposed in this paper reach 76.81% and 78.03% respectively in retelling recognition tasks. We further designed experiments to verify the effectiveness of joint learning and intermediate information utilization, and made comparative analysis with relevant representative work.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "2 \u76f8 \u76f8 \u76f8\u5173 \u5173 \u5173\u5de5 \u5de5 \u5de5\u4f5c \u4f5c \u4f5c \u76ee\u524d\u8bed\u4e49\u7ec4\u5408\u65b9\u6cd5\u4e3b\u8981\u53ef\u4ee5\u5206\u4e3a\u4e24\u7c7b\uff1a\u4e00\u79cd\u662f\u5c06\u53e5\u5b50\u89c6\u4e3a\u5e8f\u5217\u7ed3\u6784\u8fdb\u884c\u7ec4\u5408\uff0c\u5c06\u53e5\u5b50\u4e2d\u5404 \u4e2a\u8bcd\u7684\u4fe1\u606f\u8fdb\u884c\u52a0\u6743\u6574\u5408\uff0c\u4ece\u800c\u5f97\u5230\u80fd\u591f\u6709\u6548\u8868\u8fbe\u53e5\u5b50\u8bed\u4e49\u7684\u8868\u793a\uff1b\u53e6\u4e00\u79cd\u5219\u662f\u5229\u7528\u53e5\u6cd5\u7ed3\u6784 \u4f5c\u4e3a\u8bed\u4e49\u7ec4\u5408\u7684\u6307\u5bfc\uff0c\u6839\u636e\u53e5\u5b50\u4e2d\u7684\u7ed3\u6784\u5173\u7cfb\u5bf9\u8bcd\u4e49\u8868\u793a\u7684\u7ec4\u5408\u987a\u5e8f\u548c\u65b9\u5f0f\u52a0\u4ee5\u9650\u5236\uff0c\u5f97\u5230\u80fd \u66f4\u51c6\u786e\u8868\u8fbe\u53e5\u5b50\u8bed\u4e49\u7684\u8868\u793a\u3002 \u5bf9 \u4e8e \u5982 \u4f55 \u901a \u8fc7 \u7ec4 \u5408 \u8bcd \u6c47 \u8bed \u4e49 \u5f97 \u5230 \u53e5 \u5b50 \u8bed \u4e49 \uff0c \u4e00 \u79cd \u6734 \u7d20 \u7684 \u601d \u60f3 \u662f \u5c06 \u8bcd \u4e49 \u8868 \u793a \u76f8 \u52a0 \u5f97 \u5230 \u53e5 \u5b50 \u8bed \u4e49 \u8868 \u793a \uff0c \u4e00 \u822c \u8fd9 \u79cd \u65b9 \u5f0f \u88ab \u79f0 \u4e3a \u52a0 \u6cd5 \u7ec4 \u5408 (Additive Compositionality)(Mikolov et \u00a92020 \u4e2d\u56fd\u8ba1\u7b97\u8bed\u8a00\u5b66\u5927\u4f1a \u6839\u636e\u300aCreative Commons Attribution 4.0 International License\u300b\u8bb8\u53ef\u51fa\u7248 al., 2013)\u3002Hu et al. (2015)\u501f \u9274 \u56fe \u50cf \u5904 \u7406 \u7684 \u6280 \u672f \uff0c \u63d0 \u51fa \u57fa \u4e8e \u591a \u5c42 \u5377 \u79ef \u64cd \u4f5c \u7684 \u8bed \u4e49 \u7ec4 \u5408 \u65b9 \u6cd5\u3002Sutskever et al.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u548cCho et al. (2014) \u5728\u4ed6\u4eec\u63d0\u51fa\u7684seq2seq(sequence to sequence)\u6a21 \u578b\u4e2d\uff0c\u5c06RNN\u6a21\u578b\u5728\u6700\u540e\u4e00\u6b65\u7684\u8f93\u51fa\u4f5c\u4e3a\u6574\u4e2a\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a,\u5229\u7528RNN\u7c7b\u6a21\u578b\u80fd\u591f\u5145\u5206\u5229\u7528 \u957f\u8ddd\u79bb\u4fe1\u606f\u7684\u7279\u70b9(\u4ee5LSTM\u548cGRU\u7b49\u53d8\u79cd\u4e3a\u4e3b)\uff0c\u5c06\u53e5\u5b50\u4fe1\u606f\u8fdb\u884c\u6709\u6548\u5730\u878d\u5408\u3002\u8be5\u65b9\u6cd5\u4e00\u5ea6 \u968fseq2seq\u6a21\u578b\u4e00\u9053\u6210\u4e3a\u673a\u5668\u7ffb\u8bd1\u3001\u8bed\u97f3\u8bc6\u522b\u7b49\u751f\u6210\u4efb\u52a1\u4e2d\u5e38\u7528\u7684\u8bed\u4e49\u7ec4\u5408\u65b9\u6cd5\u3002Chen et al. 2017 ", | |
| "cite_spans": [ | |
| { | |
| "start": 12, | |
| "end": 18, | |
| "text": "(2014)", | |
| "ref_id": "BIBREF0" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u4f9d \u5b58 \u53e5 \u6cd5 \u6811 x1 x2 x3 x4 \u8bed\u4e49\u8868\u793a \u5b57 \u4e49 \u8868 \u793a \u4e2d\u95f4\u4fe1\u606f \u7ed3\u6784\u4fe1\u606f \u56fe 1: \u8054\u5408\u4f9d\u5b58\u5206\u6790\u7684\u6c49\u8bed\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u6a21\u578b 3 \u8054 \u8054 \u8054\u5408 \u5408 \u5408\u4f9d \u4f9d \u4f9d\u5b58 \u5b58 \u5b58\u5206 \u5206 \u5206\u6790 \u6790 \u6790\u7684 \u7684 \u7684\u6c49 \u6c49 \u6c49\u8bed \u8bed \u8bed\u8bed \u8bed \u8bed\u4e49 \u4e49 \u4e49\u7ec4 \u7ec4 \u7ec4\u5408 \u5408 \u5408\u6a21 \u6a21 \u6a21\u578b \u578b \u578b \u9488\u5bf9\u73b0\u6709\u7ed3\u6784\u5316\u8bed\u4e49\u7ec4\u5408\u65b9\u6cd5\u5b58\u5728\u7684\u95ee\u9898\uff0c\u672c\u6587\u5728Ma et al. (2018)\u7684\u57fa\u7840\u4e0a\uff0c\u8054\u5408\u57fa\u4e8e\u6ce8 \u610f\u529b\u7684\u8bed\u4e49\u7ec4\u5408\u6a21\u578b\uff0c\u63d0\u51fa\u8054\u5408\u4f9d\u5b58\u5206\u6790\u7684\u6c49\u8bed\u8bed\u4e49\u7ec4\u5408\u6a21\u578b\u3002\u5982\u56fe1\u6240\u793a\uff0c\u6211\u4eec\u7684\u6a21\u578b\u4e3b\u8981\u5305 \u542b\u4f9d\u5b58\u5206\u6790\u548c\u8bed\u4e49\u7ec4\u5408\u4e24\u5927\u90e8\u5206\u3002\u4f9d\u5b58\u5206\u6790\u90e8\u5206\u5bf9\u8f93\u5165\u53e5\u5b50\u8fdb\u884c\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\uff0c\u5e76\u5c06\u5206\u6790\u8fc7\u7a0b \u4e2d\u4ea7\u751f\u7684\u4e2d\u95f4\u4fe1\u606f\u548c\u6700\u7ec8\u5f97\u5230\u7684\u4f9d\u5b58\u53e5\u6cd5\u6811\u4f20\u9012\u7ed9\u8bed\u4e49\u7ec4\u5408\u90e8\u5206\uff1b\u8bed\u4e49\u7ec4\u5408\u90e8\u5206\u6839\u636e\u53e5\u5b50\u4e2d\u6bcf s 4 h 1 h 2 h 3", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Biaffine attention", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "s 3 s 2 s 1 s 0 BiLSTM x 1 x 0 x 2 x 3 x 4 LSTM s t1 \u7ed3\u6784\u4fe1\u606f \u2026 \u2026 s g1 s s1 + + s t2 s g2 s s2 + + s t2 s g2 s s2 + + \u4e2d\u95f4\u4fe1\u606f x1 x2 x3 x4 \u4f9d \u5b58 \u53e5 \u6cd5 \u6811 \u56fe 2: \u8054\u5408\u6a21\u578b\u4e2d\u7684\u4f9d\u5b58\u5206\u6790\u90e8\u5206 \u4e2a\u5b57\u7684\u5b57\u4e49\u8868\u793a\uff0c\u52a0\u5165\u4f9d\u5b58\u5206\u6790\u90e8\u5206\u4ea7\u751f\u7684\u4e2d\u95f4\u4fe1\u606f\uff0c\u4ee5\u4f9d\u5b58\u53e5\u6cd5\u6811\u4f5c\u4e3a\u6307\u5bfc\u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\uff0c \u4ece\u800c\u5f97\u5230\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a\u3002 3.1 \u57fa \u57fa \u57fa\u4e8e \u4e8e \u4e8eStack-Pointer Networks\u7684 \u7684 \u7684\u4f9d \u4f9d \u4f9d\u5b58 \u5b58 \u5b58\u5206 \u5206 \u5206\u6790 \u6790 \u6790\u6a21 \u6a21 \u6a21\u578b \u578b \u578b Stack-Pointer Networks(StackPTR)\u662f\u4e00\u4e2a\u57fa\u4e8e\u8f6c\u79fb\u7684\u4f9d\u5b58\u5206\u6790\u6a21\u578b\u3002\u6211\u4eec\u5728StackPTR\u7684 \u57fa\u7840\u4e0a\uff0c\u9488\u5bf9\u6c49\u8bed\u6ca1\u6709\u660e\u663e\u5206\u8bcd\u6807\u8bb0\u7684\u7279\u70b9\uff0c\u4ee5\u6bcf\u4e2a\u8bcd\u7684\u6700\u540e\u4e00\u4e2a\u5b57\u4e3a\u6839\u8282\u70b9\u6784\u5efa\u4e24\u5c42\u8bcd\u5185\u4f9d \u5b58\u7ed3\u6784\uff0c\u4ee5\u6b64\u8fdb\u884c\u5b57\u7b26\u7ea7\u6c49\u8bed\u4f9d\u5b58\u5206\u6790\u3002StackPTR\u6a21\u578b\u5927\u81f4\u6846\u67b6\u5982\u56fe2\u6240\u793a\uff0c\u5176\u4e2d\u6bcf\u4e00\u6b65\u8ba1\u7b97 \u4e2d\u5f97\u5230\u7684\u5934\u8282\u70b9\u8868\u793ah i \u5305\u542b\u8f83\u4e3a\u4e30\u5bcc\u7684\u7ed3\u6784\u4fe1\u606f\uff0c\u56e0\u6b64\u6211\u4eec\u5c06\u5176\u4f5c\u4e3a\u4f9d\u5b58\u5206\u6790\u7684\u4e2d\u95f4\u4fe1\u606f\u8868\u793a \u4f20\u9012\u7ed9\u8bed\u4e49\u7ec4\u5408\u90e8\u5206\u3002StackPTR\u4e2d\u7684\u5177\u4f53\u7ec6\u8282\u8bf7\u89c1Ma et al. (2018)\u3002 3.2 \u57fa \u57fa \u57fa\u4e8e \u4e8e \u4e8e\u6ce8 \u6ce8 \u6ce8\u610f \u610f \u610f\u529b \u529b \u529b\u7684 \u7684 \u7684\u8bed \u8bed \u8bed\u4e49 \u4e49 \u4e49\u7ec4 \u7ec4 \u7ec4\u5408 \u5408 \u5408\u6a21 \u6a21 \u6a21\u578b \u578b \u578b \u8651\u5230\u6bcf\u4e2a\u5b57\u5bf9\u53e5\u5b50\u8bed\u4e49\u7684\u8d21\u732e\u7a0b\u5ea6\u4e0d\u540c\uff0c\u6211\u4eec\u5728\u8bed\u4e49\u8ba1\u7b97\u90e8\u5206\u63d0\u51fa\u57fa\u4e8e\u6ce8\u610f\u529b\u7684\u8bed\u4e49\u7ec4\u5408 \u6a21\u578b\uff0c\u5229\u7528\u4f9d\u5b58\u5206\u6790\u90e8\u5206\u5f97\u5230\u7684\u7ed3\u6784\u4fe1\u606f\u4f5c\u4e3a\u6307\u5bfc\uff0c\u7528\u6ce8\u610f\u529b\u5f97\u5206\u4f5c\u4e3a\u4fe1\u606f\u7684\u6743\u91cd\uff0c\u8fdb\u884c\u5b57\u4e49 \u8868\u793a\u7684\u7ec4\u5408\u3002 \u5982\u56fe3\u6240\u793a\uff0c\u6a21\u578b\u4e3b\u8981\u5206\u4e3a\u5b57\u4e49\u7f16\u7801\u5c42\u3001\u5b57\u4e49\u7ec4\u5408\u5c42\u548c\u53e5\u4e49\u8f93\u51fa\u5c42\u4e09\u4e2a\u90e8\u5206\u3002\u5b57\u4e49\u7f16\u7801\u5c42\u5bf9 \u6bcf\u4e2a\u5b57\u7684\u8bed\u4e49\u8868\u793a\u8fdb\u884c\u7f16\u7801\uff1b\u5b57\u4e49\u7ec4\u5408\u5c42\u4ee5\u4f9d\u5b58\u53e5\u6cd5\u6811\u4f5c\u4e3a\u8bed\u4e49\u8ba1\u7b97\u7684\u7ed3\u6784\uff0c\u5c06\u6bcf\u4e2a\u4f9d\u5b58\u8282\u70b9 \u7684\u4fe1\u606f\u4f20\u9012\u7ed9\u5934\u8282\u70b9\uff1b\u53e5\u4e49\u8f93\u51fa\u5c42\u5c06\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u5f97\u5230\u7684\u6bcf\u4e2a\u5b57\u7684\u7ed3\u6784\u5316\u8bed\u4e49\u4fe1\u606f\u8fdb\u884c\u6c60\u5316\u5408 \u5e76\uff0c\u5f97\u5230\u8868\u793a\u53e5\u5b50\u8bed\u4e49\u7684\u5411\u91cf\u8868\u793a\u3002 3.2.1 \u5b57 \u5b57 \u5b57\u4e49 \u4e49 \u4e49\u7f16 \u7f16 \u7f16\u7801 \u7801 \u7801\u5c42 \u5c42 \u5c42", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u53c2\u8003Hochreiter and Schmidhuber (1997)\uff0c\u6211\u4eec\u4f7f\u7528\u53cc\u5411LSTM\u8fdb\u884c\u5b57\u4e49\u8868\u793a\u7684\u7f16\u7801\u3002\u4f9d\u5b58", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u5206\u6790\u6a21\u578b\u7684\u4e2d\u95f4\u4fe1\u606f\u4ee5\u5b57\u5411\u91cf\u7684\u5f62\u5f0f\uff0c\u4f20\u9012\u4e86\u4e30\u5bcc\u7684\u7ed3\u6784\u4fe1\u606f\u548c\u8bed\u4e49\u4fe1\u606f\u3002\u6211\u4eec\u5c06\u5176\u4f5c\u4e3a\u989d\u5916 \u7684\u5b57\u4e49\u4fe1\u606f\uff0c\u5bf9\u9884\u8bad\u7ec3\u5b57\u5411\u91cf\u8fdb\u884c\u6269\u5145\u3002\u5bf9\u4e8e\u7ed9\u5b9a\u7684\u53e5\u5b50x = {x 1 , x 2 , . . . , x n }\uff0c\u7f16\u7801\u5c42\u9996\u5148\u5c06 \u6bcf\u4e2a\u5b57\u7684\u539f\u59cb\u5411\u91cf\u8868\u793ax i \u4e0e\u4f9d\u5b58\u5206\u6790\u4e2d\u95f4\u4fe1\u606f\u8868\u793ah i \u8fdb\u884c\u62fc\u63a5\uff0c\u5f97\u5230\u5b57\u7684\u5411\u91cf\u8868\u793ax i \u3002\u7136\u540e\u6211 \u4eec\u5c06\u8f93\u5165\u53cc\u5411LSTM\uff0c\u7f16\u7801\u53e5\u5b50\u4fe1\u606f\u5f97\u5230\u6bcf\u4e2a\u5b57\u5728\u53e5\u5b50\u4e2d\u7684\u8bed\u4e49\u8868\u793am i \u3002 \u6b64\u5916\uff0c\u4e3a\u4e86\u63d0\u5347\u53e5\u5b50\u5168\u5c40\u4fe1\u606f\u7684\u5229\u7528\uff0c\u6211\u4eec\u5c06\u53cc\u5411LSTM\u4e24\u4e2a\u65b9\u5411\u7684\u6700\u540e\u4e00\u6b65\u8f93\u51fa\u8fdb\u884c\u62fc \u63a5\uff0c\u5f97\u5230\u53e5\u5b50\u8868\u793am x \uff0c\u4f5c\u4e3a\u5b57\u4e49\u7ec4\u5408\u5c42\u4e2d\u7684\u989d\u5916\u4fe1\u606f\u8f93\u5165\u3002 \u8ba1\u7b97\u8bed\u8a00\u5b66 \u4e2d\u95f4\u4fe1\u606f Attention-Composition m 4 m 3 m 2 m 1 BiLSTM h 2 h 3 h 4 m x h 1 x 2 x 3 x 4 x 1 \u7ed3\u6784\u4fe1\u606f x1 x2 x3 x4 \u4f9d \u5b58 \u53e5 \u6cd5 \u6811 Mean-Pooling s x \u8bed\u4e49\u8868\u793a \u5b57\u4e49\u7f16\u7801\u5c42 \u5b57\u4e49\u7ec4\u5408\u5c42 \u53e5\u4e49\u8f93\u51fa\u5c42 \u56fe 3: \u57fa\u4e8e\u6ce8\u610f\u529b\u7684\u8bed\u4e49\u7ec4\u5408\u6a21\u578b 3.2.2 \u5b57 \u5b57 \u5b57\u4e49 \u4e49 \u4e49\u7ec4 \u7ec4 \u7ec4\u5408 \u5408 \u5408\u5c42 \u5c42 \u5c42", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u6211\u4eec\u5728\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u5c42\u4f7f\u7528\u56fe\u6ce8\u610f\u529b\u7f51\u7edc (Hochreiter and Schmidhuber, 1997 ", | |
| "cite_spans": [ | |
| { | |
| "start": 19, | |
| "end": 52, | |
| "text": "(Hochreiter and Schmidhuber, 1997", | |
| "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": ")\u8fdb\u884c\u5b57\u4e49\u8868 \u793a\u7684\u7ec4\u5408\u8ba1\u7b97\uff0c\u5176\u4e2d\u4f9d\u5b58\u5206\u6790\u90e8\u5206\u9884\u6d4b\u51fa\u7684\u4f9d\u5b58\u53e5\u6cd5\u6811\u4f5c\u4e3a\u6307\u793a\u8282\u70b9\u76f8\u5173\u6027\u7684\u6709\u5411\u56fe\uff0c\u5bf9\u5b57\u4e49 \u7f16\u7801\u5c42\u8f93\u51fa\u7684\u5b57\u4e49\u8868\u793a\u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\uff0c\u5176\u4e2d\u6bcf\u4e2a\u8282\u70b9\u5728\u8ba1\u7b97\u65f6\u4ec5\u8003\u8651\u5176\u4f9d\u5b58\u8282\u70b9\uff0c\u5982\u56fe4\u6240\u793a\u3002 m 1 m 2 m 3 m 4 m 1 ' m 2 ' m 3 ' m 4 ' m 2 m 1 m 4 \u03b1 21 \u03b1 24 m x \u03b1 2x \u56fe 4: \u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u5c42\u4e2d\u7684\u4fe1\u606f\u8ba1\u7b97\u65b9\u5f0f \u6211\u4eec\u5c06\u5b57\u4e49\u7f16\u7801\u5c42\u4e2d\u5f97\u5230\u7684\u53e5\u5b50\u8868\u793am x \u4f5c\u4e3a\u4e00\u4e2a\u989d\u5916\u7684\u8282\u70b9\uff0c\u4f5c\u4e3a\u6240\u6709\u8282\u70b9\u7684\u4f9d\u5b58\u8282\u70b9\u53c2 \u4e0e\u56fe\u6ce8\u610f\u529b\u7f51\u7edc\u7684\u8ba1\u7b97\uff0c\u4f7f\u7ec4\u5408\u8ba1\u7b97\u540e\u6bcf\u4e2a\u5b57\u7684\u8868\u793a\u90fd\u5305\u542b\u4e0d\u540c\u7a0b\u5ea6\u7684\u53e5\u5b50\u5168\u5c40\u4fe1\u606f\u3002\u6211\u4eec\u9009 \u62e9\u4f7f\u7528\u53cc\u7ebf\u6027\u53d8\u6362\u4f5c\u4e3a\u6ce8\u610f\u529b\u5f97\u5206\u7684\u8ba1\u7b97\u673a\u5236\uff0c\u56e0\u6b64\u5b57\u4e49\u7ec4\u5408\u5c42\u7684\u8ba1\u7b97\u516c\u5f0f\u5373\u4e3a\uff1a m i = m i + j\u2208V (i) m i W m j + m i W m x \u00d7 m x (1) \u5176\u4e2dm i \u548cm i \u5206\u522b\u8868\u793a\u7b2ci\u4e2a\u5b57\u5728\u8fdb\u884c\u5b57\u4e49\u7ec4\u5408\u8ba1\u7b97\u524d\u540e\u7684\u8bed\u4e49\u8868\u793a\uff1bj \u2208 V (i)\u8868\u793a\u8282\u70b9i\u7684\u6240\u6709 \u4f9d\u5b58\u8282\u70b9\uff0cW \u4e3a\u53cc\u7ebf\u6027\u53d8\u6362\u7684\u53c2\u6570\u77e9\u9635\u3002 3.2.3 \u53e5 \u53e5 \u53e5\u4e49 \u4e49 \u4e49\u8f93 \u8f93 \u8f93\u51fa \u51fa \u51fa\u5c42 \u5c42 \u5c42 \u501f\u9274Mou et al. (2016)\u7684\u5de5\u4f5c\uff0c\u6211\u4eec\u4f7f\u7528\u6c60\u5316\u64cd\u4f5c\u5bf9\u5b57\u4e49\u7ec4\u5408\u8ba1\u7b97\u540e\u7684\u5b57\u5411\u91cf\u8fdb\u884c\u6c60\u5316\u64cd \u4f5c\uff0c\u83b7\u5f97\u6700\u7ec8\u7684\u53e5\u5b50\u8bed\u4e49\u8868\u793a\u3002\u4e3a\u4e86\u4f7f\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a\u4e0d\u53d7\u53e5\u5b50\u957f\u5ea6\u7684\u5f71\u54cd\uff0c\u540c\u65f6\u5c3d\u53ef\u80fd\u4fdd\u5b58 \u66f4\u591a\u7684\u8bed\u4e49\u4fe1\u606f\uff0c\u6211\u4eec\u9009\u62e9\u4f7f\u7528\u5e73\u5747\u6c60\u5316\u5bf9\u8bed\u4e49\u7ec4\u5408\u5c42\u8f93\u51fa\u7684\u5b57\u5411\u91cf\u8868\u793a\u8fdb\u884c\u5408\u5e76\u3002\u6700\u7ec8\u53e5\u5b50 \u8bed\u4e49\u8868\u793as x \u7684\u8ba1\u7b97\u516c\u5f0f\u5982\u4e0b\uff1a s x = 1 n n i=1 m i (2) x 13 x 12 x 11 \u2026 x n1 \u53e5\u5b501 Classifier e x1 e x2 Label x 23 x 22 x 21 \u2026 x n2 \u53e5\u5b502 \u8bed\u4e49\u7ec4\u5408 \u8bed\u4e49\u7ec4\u5408 \u56fe 5: \u590d\u8ff0\u8bc6\u522b\u6a21\u578b 3.3 \u6a21 \u6a21 \u6a21\u578b \u578b \u578b\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3 3.3.1 \u590d \u590d \u590d\u8ff0 \u8ff0 \u8ff0\u8bc6 \u8bc6 \u8bc6\u522b \u522b \u522b\u4efb \u4efb \u4efb\u52a1 \u52a1 \u52a1 \u590d\u8ff0\u8bc6\u522b\u7684\u4e3b\u8981\u76ee\u6807\u662f\u5224\u65ad\u7ed9\u5b9a\u7684\u4e24\u4e2a\u53e5\u5b50\u662f\u5426\u8868\u8fbe\u76f8\u540c(\u6216\u76f8\u8fd1)\u7684\u8bed\u4e49\u3002\u672c\u6587\u8ba4\u4e3a\u5728 \u6240\u6709\u76f8\u5173\u4efb\u52a1\u4e2d\uff0c\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e0e\u53e5\u5b50\u8bed\u4e49\u5173\u8054\u6700\u4e3a\u7d27\u5bc6\uff0c\u56e0\u6b64\u63d0\u51fa\u57fa\u4e8e\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u7684\u8bed\u4e49 \u7ec4\u5408\u8ba1\u7b97\u7684\u8bad\u7ec3\u548c\u8bc4\u6d4b\u65b9\u6cd5\u3002\u4e3a\u4e86\u80fd\u591f\u66f4\u76f4\u89c2\u5730\u53cd\u5e94\u8bed\u4e49\u7ec4\u5408\u6a21\u578b\u5bf9\u590d\u8ff0\u8bc6\u522b\u6027\u80fd\u7684\u5f71\u54cd\uff0c\u6211 \u4eec\u672a\u4f7f\u7528\u76ee\u524d\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e2d\u6700\u6d41\u884c\u7684\u4ea4\u53c9\u6ce8\u610f\u529b\u673a\u5236(Veli\u010dkovi\u0107 et al., 2017)\u548c\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21 \u578b(Wang et al., 2017)\uff0c\u800c\u662f\u501f\u9274\u65e9\u671f\u5e38\u7528\u7684\u83b7\u5f97\u53e5\u5b50\u8bed\u4e49\u8868\u793a\u540e\u8fdb\u884c\u5173\u7cfb\u9884\u6d4b\u7684\u65b9\u5f0f(Hu et al., 2015)\uff0c\u5bf9\u6bcf\u4e2a\u53e5\u5b50\u72ec\u7acb\u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u7136\u540e\u4f7f\u7528\u5206\u7c7b\u5668\u8fdb\u884c\u590d\u8ff0\u7684\u5224\u522b\u3002 \u6211\u4eec\u7684\u590d\u8ff0\u8bc6\u522b\u6a21\u578b\u5982\u56fe5\u6240\u793a\uff0c\u4f7f\u7528\u672c\u7ae0\u63d0\u51fa\u7684\u8bed\u4e49\u7ec4\u5408\u6a21\u578b(\u4f9d\u5b58\u5206\u6790\u90e8\u5206\u672a\u5728\u56fe\u4e2d\u663e \u793a)\u5bf9\u4e24\u4e2a\u8f93\u5165\u53e5\u5206\u522b\u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u5f97\u5230\u5b83\u4eec\u7684\u8bed\u4e49\u8868\u793a\u3002\u7136\u540e\u5c06\u4e24\u4e2a\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a \u62fc\u63a5\u540e\u8f93\u5165\u4e00\u4e2a\u5206\u7c7b\u5668\uff0c\u5224\u522b\u5b83\u4eec\u662f\u5426\u4e92\u4e3a\u590d\u8ff0\u3002 3.3.2 \u8054 \u8054 \u8054\u5408 \u5408 \u5408\u6a21 \u6a21 \u6a21\u578b \u578b \u578b\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u65b9 \u65b9 \u65b9\u5f0f \u5f0f \u5f0f \u5728\u8fdb\u884c\u8054\u5408\u6a21\u578b\u7684\u8bad\u7ec3\u65f6\uff0c\u6211\u4eec\u8003\u8651\u4e24\u79cd\u8bad\u7ec3\u65b9\u5f0f\uff1a\u4e00\u79cd\u662f\u5c06\u4f9d\u5b58\u5206\u6790\u6a21\u578b\u8fdb\u884c\u9884\u8bad\u7ec3\uff0c \u7136\u540e\u5728\u8bad\u7ec3\u8bed\u4e49\u7ec4\u5408\u6a21\u578b\u65f6\u5bf9\u4f9d\u5b58\u5206\u6790\u6a21\u578b\u7684\u53c2\u6570\u8fdb\u884c\u5fae\u8c03\uff1b\u53e6\u4e00\u79cd\u662f\u76f4\u63a5\u5c06\u4e24\u4e2a\u4efb\u52a1\u8fdb\u884c\u8fed \u4ee3\u8bad\u7ec3\u3002 \u9884 \u9884 \u9884\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u65b9 \u65b9 \u65b9\u5f0f \u5f0f \u5f0f: \u6211\u4eec\u9996\u5148\u5bf9\u8054\u5408\u6a21\u578b\u4e2d\u7684\u4f9d\u5b58\u5206\u6790\u90e8\u5206\u7684\u6a21\u578b\u53c2\u6570\u8fdb\u884c\u9884\u8bad\u7ec3\uff0c\u7136\u540e\u8fdb\u884c\u590d \u8ff0\u8bc6\u522b\u4efb\u52a1\u7684\u8bad\u7ec3\uff0c\u5e76\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5bf9\u4f9d\u5b58\u5206\u6790\u90e8\u5206\u7684\u6a21\u578b\u53c2\u6570\u8fdb\u884c\u5fae\u8c03\u3002\u6a21\u578b\u7684\u76ee\u6807\u51fd\u6570\u901a \u8fc7\u6700\u5c0f\u5316\u4ea4\u53c9\u71b5\u635f\u5931\u5b9a\u4e49\uff1a L(\u03b8) = L par (\u03b8 com , \u03b8 dep ) = \u2212logP par (y|x, \u03b8 dep , \u03b8 com )", | |
| "eq_num": "(3)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u5176\u4e2d\u03b8 com \u548c\u03b8 dep \u5206\u522b\u8868\u793a\u4f9d\u5b58\u5206\u6790\u548c\u8bed\u4e49\u7ec4\u5408\u6a21\u578b\u7684\u53c2\u6570\uff0c\u8868\u793a\u590d\u8ff0\u5173\u7cfb\u6807\u7b7e\u3002 \u8fed \u8fed \u8fed\u4ee3 \u4ee3 \u4ee3\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u65b9 \u65b9 \u65b9\u5f0f \u5f0f \u5f0f: \u5728\u6a21\u578b\u8bad\u7ec3\u65f6\uff0c\u6211\u4eec\u6bcf\u6b21\u968f\u673a\u4ece\u4e24\u4e2a\u4efb\u52a1\u4e2d\u9009\u62e9\u4e00\u4e2a\u4efb\u52a1\uff0c\u5e76\u4ece\u5bf9\u5e94\u7684\u8bad \u7ec3\u96c6\u4e2d\u968f\u673a\u9009\u53d6\u4e00\u6279\u6570\u636e\u8fdb\u884c\u6a21\u578b\u7684\u8bad\u7ec3\u3002\u5176\u4e2d\uff0c\u5728\u8fdb\u884c\u4f9d\u5b58\u5206\u6790\u4efb\u52a1\u7684\u8bad\u7ec3\u65f6\uff0c\u4ec5\u5bf9\u4f9d\u5b58\u5206 \u6790\u90e8\u5206\u7684\u6a21\u578b\u53c2\u6570\u8fdb\u884c\u5b66\u4e60\uff1b\u5728\u8fdb\u884c\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u7684\u8bad\u7ec3\u65f6\uff0c\u540c\u65f6\u5bf9\u8054\u5408\u6a21\u578b\u4e2d\u4e24\u4e2a\u90e8\u5206\u7684\u53c2 \u6570\u8fdb\u884c\u5b66\u4e60\u3002\u6a21\u578b\u7684\u76ee\u6807\u51fd\u6570\u901a\u8fc7\u6700\u5c0f\u5316\u4ea4\u53c9\u71b5\u635f\u5931\u5b9a\u4e49\uff1a L(\u03b8) = L dep (\u03b8 dep ) + L par (\u03b8 com , \u03b8 dep ) = \u2212logP dep (A|x, \u03b8 dep ) \u2212 logP par (y|x, \u03b8 dep , \u03b8 com ) (4) \u5176\u4e2d\u03b8 com \u548c\u03b8 dep \u5206\u522b\u8868\u793a\u4f9d\u5b58\u5206\u6790\u548c\u8bed\u4e49\u7ec4\u5408\u6a21\u578b\u7684\u53c2\u6570\uff0cA\u8868\u793a\u5b57\u8282\u522b\u4f9d\u5b58\u53e5\u6cd5\u6811\u7684 \u8fb9\u96c6\uff0cy\u8868\u793a\u590d\u8ff0\u5173\u7cfb\u6807\u7b7e\u3002\u4e3a\u4e86\u4f7f\u6a21\u578b\u8bad\u7ec3\u8fc7\u7a0b\u66f4\u4e3a\u7a33\u5b9a\uff0c\u6211\u4eec\u9996\u5148\u5bf9\u4f9d\u5b58\u5206\u6790\u6a21\u578b\u8fdb\u884c \u4e86100\u6b21\u8bad\u7ec3\u3002 4 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c 4.1 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u8bbe \u8bbe \u8bbe\u7f6e \u7f6e \u7f6e \u672c\u6587\u4f7f\u7528\u7684\u4f9d\u5b58\u5206\u6790\u5b9e\u9a8c\u6570\u636e\u4e3a\u5bbe\u5dde\u6c49\u8bed\u6811\u5e93CTB5\uff0c\u590d\u8ff0\u8bc6\u522b\u5b9e\u9a8c\u6570\u636e\u4e3a\u8bed\u4e49\u76f8\u4f3c\u5ea6\u6570 \u636e\u96c6LCQMC(", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "A meeting of SIGDAT, a Special Interest Group of the ACL", | |
| "authors": [], | |
| "year": 2014, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1724--1734", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": ", October 25-29, 2014, Doha, Qatar, A meeting of SIGDAT, a Special Interest Group of the ACL, pages 1724-1734. ACL.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "BERT: pre-training of deep bidirectional transformers for language understanding", | |
| "authors": [ | |
| { | |
| "first": "Jacob", | |
| "middle": [], | |
| "last": "Devlin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ming-Wei", | |
| "middle": [], | |
| "last": "Chang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kenton", | |
| "middle": [], | |
| "last": "Lee", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kristina", | |
| "middle": [], | |
| "last": "Toutanova", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL-HLT 2019", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "4171--4186", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: pre-training of deep bidirectional transformers for language understanding. In Jill Burstein, Christy Doran, and Thamar Solorio, editors, Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL-HLT 2019, Minneapolis, MN, USA, June 2-7, 2019, Volume 1 (Long and Short Papers), pages 4171-4186. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Long short-term memory", | |
| "authors": [ | |
| { | |
| "first": "Sepp", | |
| "middle": [], | |
| "last": "Hochreiter", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J\u00fcrgen", | |
| "middle": [], | |
| "last": "Schmidhuber", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1997, | |
| "venue": "Neural Computation", | |
| "volume": "9", | |
| "issue": "8", | |
| "pages": "1735--1780", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Sepp Hochreiter and J\u00fcrgen Schmidhuber. 1997. Long short-term memory. Neural Computation, 9(8):1735-1780.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Convolutional neural network architectures for matching natural language sentences", | |
| "authors": [ | |
| { | |
| "first": "Baotian", | |
| "middle": [], | |
| "last": "Hu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Zhengdong", | |
| "middle": [], | |
| "last": "Lu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Hang", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Qingcai", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Baotian Hu, Zhengdong Lu, Hang Li, and Qingcai Chen. 2015. Convolutional neural network architec- tures for matching natural language sentences. CoRR, abs/1503.03244.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "LCQMC:a large-scale Chinese question matching corpus", | |
| "authors": [ | |
| { | |
| "first": "Xin", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Qingcai", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chong", | |
| "middle": [], | |
| "last": "Deng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Huajun", | |
| "middle": [], | |
| "last": "Zeng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jing", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Dongfang", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Buzhou", | |
| "middle": [], | |
| "last": "Tang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the 27th International Conference on Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1952--1962", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Xin Liu, Qingcai Chen, Chong Deng, Huajun Zeng, Jing Chen, Dongfang Li, and Buzhou Tang. 2018. LCQMC:a large-scale Chinese question matching corpus. In Proceedings of the 27th International Conference on Computational Linguistics, pages 1952-1962, Santa Fe, New Mexico, USA, August. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Stack-pointer networks for dependency parsing", | |
| "authors": [ | |
| { | |
| "first": "Xuezhe", | |
| "middle": [], | |
| "last": "Ma", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Zecong", | |
| "middle": [], | |
| "last": "Hu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jingzhou", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Nanyun", | |
| "middle": [], | |
| "last": "Peng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Graham", | |
| "middle": [], | |
| "last": "Neubig", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Eduard", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Hovy", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "1403--1414", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Xuezhe Ma, Zecong Hu, Jingzhou Liu, Nanyun Peng, Graham Neubig, and Eduard H. Hovy. 2018. Stack-pointer networks for dependency parsing. In Iryna Gurevych and Yusuke Miyao, editors, Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, ACL 2018, Melbourne, Australia, July 15-20, 2018, Volume 1: Long Papers, pages 1403-1414. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Distributed representations of words and phrases and their compositionality", | |
| "authors": [ | |
| { | |
| "first": "Tomas", | |
| "middle": [], | |
| "last": "Mikolov", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ilya", | |
| "middle": [], | |
| "last": "Sutskever", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kai", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Gregory", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Corrado", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jeffrey", | |
| "middle": [], | |
| "last": "Dean", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "Advances in Neural Information Processing Systems 26: 27th Annual Conference on Neural Information Processing Systems 2013. Proceedings of a meeting held", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "3111--3119", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tomas Mikolov, Ilya Sutskever, Kai Chen, Gregory S. Corrado, and Jeffrey Dean. 2013. Distributed representations of words and phrases and their compositionality. In Christopher J. C. Burges, L\u00e9on Bottou, Zoubin Ghahramani, and Kilian Q. Weinberger, editors, Advances in Neural Information Processing Systems 26: 27th Annual Conference on Neural Information Processing Systems 2013. Proceedings of a meeting held December 5-8, 2013, Lake Tahoe, Nevada, United States, pages 3111- 3119.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Natural language inference by tree-based convolution and heuristic matching", | |
| "authors": [ | |
| { | |
| "first": "Lili", | |
| "middle": [], | |
| "last": "Mou", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Rui", | |
| "middle": [], | |
| "last": "Men", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ge", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yan", | |
| "middle": [], | |
| "last": "Xu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Lu", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Rui", | |
| "middle": [], | |
| "last": "Yan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Zhi", | |
| "middle": [], | |
| "last": "Jin", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, ACL 2016", | |
| "volume": "2", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lili Mou, Rui Men, Ge Li, Yan Xu, Lu Zhang, Rui Yan, and Zhi Jin. 2016. Natural language inference by tree-based convolution and heuristic matching. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, ACL 2016, August 7-12, 2016, Berlin, Germany, Volume 2: Short Papers. The Association for Computer Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Sequence to sequence learning with neural networks", | |
| "authors": [ | |
| { | |
| "first": "Ilya", | |
| "middle": [], | |
| "last": "Sutskever", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Oriol", | |
| "middle": [], | |
| "last": "Vinyals", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "V", | |
| "middle": [], | |
| "last": "Quoc", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Le", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Advances in Neural Information Processing Systems 27: Annual Conference on Neural Information Processing Systems", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "3104--3112", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Ilya Sutskever, Oriol Vinyals, and Quoc V. Le. 2014. Sequence to sequence learning with neural networks. In Zoubin Ghahramani, Max Welling, Corinna Cortes, Neil D. Lawrence, and Kilian Q. Weinberger, editors, Advances in Neural Information Processing Systems 27: Annual Conference on Neural Information Processing Systems 2014, December 8-13 2014, Montreal, Quebec, Canada, pages 3104- 3112.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Graph attention networks", | |
| "authors": [ | |
| { | |
| "first": "Petar", | |
| "middle": [], | |
| "last": "Veli\u010dkovi\u0107", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Guillem", | |
| "middle": [], | |
| "last": "Cucurull", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Arantxa", | |
| "middle": [], | |
| "last": "Casanova", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Adriana", | |
| "middle": [], | |
| "last": "Romero", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Pietro", | |
| "middle": [], | |
| "last": "Lio", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yoshua", | |
| "middle": [], | |
| "last": "Bengio", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": { | |
| "arXiv": [ | |
| "arXiv:1710.10903" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Petar Veli\u010dkovi\u0107, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. 2017. Graph attention networks. arXiv preprint arXiv:1710.10903.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Bilateral multi-perspective matching for natural language sentences", | |
| "authors": [ | |
| { | |
| "first": "Zhiguo", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Wael", | |
| "middle": [], | |
| "last": "Hamza", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Radu", | |
| "middle": [], | |
| "last": "Florian", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "4144--4150", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Zhiguo Wang, Wael Hamza, and Radu Florian. 2017. Bilateral multi-perspective matching for natural language sentences. In Carles Sierra, editor, Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, IJCAI 2017, Melbourne, Australia, August 19-25, 2017, pages 4144-4150. ijcai.org.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Long short-term memory over recursive structures", | |
| "authors": [ | |
| { | |
| "first": "Xiao-Dan", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Parinaz", | |
| "middle": [], | |
| "last": "Sobhani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Hongyu", | |
| "middle": [], | |
| "last": "Guo", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proceedings of the 32nd International Conference on Machine Learning", | |
| "volume": "37", | |
| "issue": "", | |
| "pages": "1604--1612", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Xiao-Dan Zhu, Parinaz Sobhani, and Hongyu Guo. 2015. Long short-term memory over recursive structures. In Francis R. Bach and David M. Blei, editors, Proceedings of the 32nd International Conference on Machine Learning, ICML 2015, Lille, France, 6-11 July 2015, volume 37 of JMLR Workshop and Conference Proceedings, pages 1604-1612. JMLR.org.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF2": { | |
| "content": "<table><tr><td>\u6a21\u578b \u6a21\u578b</td><td colspan=\"4\">\u5206\u8bcd(%) \u8bcd\u6027\u6807\u6ce8(%) \u4f9d\u5b58\u5206\u6790(%) \u7c7b\u578b Acc(%) F1(%)</td></tr><tr><td>Ours (pipeline) Baseline (Mean)</td><td>98.25</td><td>95.13</td><td>73.21</td><td>85.44 74.72</td></tr><tr><td>Ours (pre-train) CNN</td><td>97.85 \u5e8f\u5217\u5316</td><td>94.35</td><td>74.84</td><td>83.04 76.27</td></tr><tr><td>Ours (alternate) LSTM</td><td>97.93</td><td>94.22</td><td>75.53</td><td>82.13 76.31</td></tr><tr><td colspan=\"5\">\u8868 4: \u8054\u5408\u6a21\u578b\u8bad\u7ec3\u65b9\u5f0f\u5728\u4f9d\u5b58\u5206\u6790\u4e0a\u7684\u5bf9\u6bd4\u7ed3\u679c Tree-CNN 72.93 75.46 \u7ed3\u6784\u5316(\u7ba1\u9053) Ours (pipeline) 72.64 75.74</td></tr><tr><td colspan=\"5\">Ours (alternate) \u7ed3\u6784\u5316(\u8054\u5408) 76.37 78.03</td></tr><tr><td colspan=\"5\">Liu et al., 2018)\u3002\u4e24\u4e2a\u6570\u636e\u96c6\u7684\u5212\u5206\u53ca\u7edf\u8ba1\u6570\u636e\u5982\u88681\u548c2\u6240\u793a\u3002 \u603b \u603b \u603b\u7ed3 \u7ed3 \u7ed3\u5206 \u5206 \u5206\u6790 \u6790 \u6790: \u603b\u7684\u6765\u8bf4\uff0c\u867d\u7136\u6211\u4eec\u7684\u8054\u5408\u6a21\u578b\u5728\u4e00\u4f53\u5316\u4f9d\u5b58\u5206\u6790\u4efb\u52a1\u4e0a\u7684\u7cbe\u5ea6\u6709\u6240\u964d\u4f4e\uff0c \u8868 6: \u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\u5bf9\u6bd4\u7ed3\u679c</td></tr><tr><td colspan=\"5\">\u4f46\u5728\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e0a\u7684\u7cbe\u5ea6\u6709\u6240\u63d0\u5347\u3002\u6211\u4eec\u6839\u636e\u88681\u548c\u88682\u4e2d\u4e24\u79cd\u6570\u636e\u5e73\u5747\u53e5\u957f\u7684\u5bf9\u6bd4\uff0c\u4ee5\u53ca \u6570\u636e\u96c6 \u53e5\u5b50\u6570 \u5e73\u5747\u53e5\u957f \u8bcd\u6570 \u5b57\u6570 \u8bad\u7ec3\u96c6 18104 44.4 \u8fdb\u884c\u6570\u636e\u5206\u6790\u540e\u53d1\u73b0\uff1a\u6211\u4eec\u6240\u4f7f\u7528\u7684\u4f9d\u5b58\u5206\u6790\u6570\u636e(CTB5)\u4e3a\u65b0\u95fb\u9886\u57df\u7684\u6587\u672c\uff0c\u53e5\u5b50\u8f83\u957f\u4e14 494k 805k \u5f00\u53d1\u96c6 352 32.8 7k 12k \u6d4b\u8bd5\u96c6 348 39.5 8k 14k \u8868 1: CTB5\u6570\u636e\u5212\u5206 \u6570\u636e\u96c6 \u53e5\u5bf9\u6570 \u5e73\u5747\u53e5\u957f \u5b57\u6570 \u6b63\u4f8b\u5360\u6bd4 \u8868\u8fbe\u5f62\u5f0f\u8f83\u4e3a\u4e66\u9762\u5316\uff1b\u590d\u8ff0\u8bc6\u522b\u6570\u636e(LCQMC)\u4e3a\u641c\u7d22\u5f15\u64ce\u4e0a\u6536\u96c6\u7684\u95ee\u53e5\uff0c\u53e5\u5b50\u8f83\u77ed\u4e14\u8868\u8fbe \u5f62\u5f0f\u8f83\u4e3a\u53e3\u8bed\u5316\u3002\u4e24\u79cd\u6570\u636e\u5728\u9886\u57df\u548c\u8bed\u8a00\u73b0\u8c61\u4e0a\u5b58\u5728\u8f83\u5927\u7684\u5dee\u5f02\uff0c\u56e0\u6b64\u6211\u4eec\u505a\u51fa\u5982\u4e0b\u63a8\u65ad\uff1a (1)\u4f7f\u7528CTB5\u4e0a\u8bad\u7ec3\u7684\u4f9d\u5b58\u5206\u6790\u6a21\u578b\uff0c\u5728\u5bf9LCQMC\u4e2d\u7684\u53e5\u5b50\u8fdb\u884c\u7684\u4f9d\u5b58\u5206\u6790\u7cbe\u5ea6\u4f1a\u6709\u660e\u663e \u7684\u964d\u4f4e\uff0c\u5e76\u56e0\u6b64\u5bfc\u81f4\u8bed\u4e49\u7ec4\u5408\u6a21\u578b\u5728\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u7684\u7cbe\u5ea6\u8f83\u4f4e\uff1b(2)\u6211\u4eec\u7684\u8054\u5408\u6a21\u578b\u80fd\u591f\u9488 \u5bf9LCQMC\u7684\u6570\u636e\u7279\u70b9\uff0c\u5bf9\u4f9d\u5b58\u5206\u6790\u90e8\u5206\u7684\u53c2\u6570\u8fdb\u884c\u9002\u5f53\u5730\u8c03\u6574\uff0c\u867d\u7136\u4f7f\u5176\u5728CTB5\u4e0a\u7684\u4e00\u4f53\u5316 4.3.4 \u6a21\u578b ACC(%) F1(%) \u4f9d\u5b58\u5206\u6790\u7cbe\u5ea6\u6709\u6240\u964d\u4f4e\uff0c\u4f46\u80fd\u591f\u9690\u5f0f\u5730\u63d0\u5347\u5176\u5728LCQMC\u6570\u636e\u4e0a\u7684\u4f9d\u5b58\u5206\u6790\u7cbe\u5ea6\uff0c\u8fdb\u800c\u63d0\u5347\u5728 BiMPM 83.4 85.0 \u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e0a\u7684\u7cbe\u5ea6\u3002 Bert-large 87.3 -</td></tr><tr><td colspan=\"5\">\u8bad\u7ec3\u96c6 \u5f00\u53d1\u96c6 Ours (alternate) 239k 10.9 9k 12.5 4.3.2 \u4f9d \u4f9d \u4f9d\u5b58 \u5b58 \u5b58\u4fe1 \u4fe1 \u4fe1\u606f \u606f \u606f\u5229 \u5229 \u5229\u7528 \u7528 \u7528\u7684 \u7684 \u7684\u5bf9 \u5bf9 \u5bf9\u6bd4 \u6bd4 \u6bd4 \u6d4b\u8bd5\u96c6 13k 9.7 \u6211\u4eec\u5bf9\u8054\u5408\u6a21\u578b\u4e2d\u7684\u8bed\u4e49\u7ec4\u5408\u90e8\u5206\u4f7f\u7528\u5230\u7684\u4f9d\u5b58\u4fe1\u606f\u8fdb\u884c\u6d88\u878d\u5b9e\u9a8c\u3002\u5206\u522b\u5bf9\u6bd4\u4e86\u4e0d\u4f7f\u7528\u4f9d 5.2m 0.58 0.2m 76.37 78.03 0.50 0.2m 0.50 \u5b58\u7ed3\u6784\u4fe1\u606f(without-structure)\u548c\u4e0d\u4f7f\u7528\u4f9d\u5b58\u4e2d\u95f4\u4fe1\u606f(without-intermediate)\uff0c\u5b9e\u9a8c\u7ed3\u679c \u8868 7: \u590d\u8ff0\u8bc6\u522b\u6027\u80fd\u5bf9\u6bd4</td></tr><tr><td colspan=\"5\">\u5982\u88685\u6240\u793a\u3002 \u5bf9\u6bd4\u7ed3\u679c\u8868\u660e\uff0c\u6211\u4eec\u7684\u6a21\u578b\u8f83\u73b0\u6709\u590d\u8ff0\u8bc6\u522b\u65b9\u6cd5\u6709\u5341\u5206\u660e\u663e\u7684\u5dee\u8ddd\u3002\u7ecf\u8fc7\u5206\u6790\uff0c\u6211\u4eec \u8868 2: LCQMC\u6570\u636e\u8be6\u60c5</td></tr><tr><td colspan=\"5\">\u6a21\u578b \u8ba4\u4e3a\u8fd9\u4e3b\u8981\u7531\u4ee5\u4e0b\u539f\u56e0\u5bfc\u81f4\uff1a1)\u73b0\u5728\u4e3b\u6d41\u590d\u8ff0\u8bc6\u522b\u65b9\u6cd5\u4e3b\u8981\u4f7f\u7528\u4ea4\u53c9\u6ce8\u610f\u529b\u673a\u5236(Cross-ACC(%) F1(%)</td></tr><tr><td colspan=\"5\">Ours (alternate) Attention)\u8fdb\u884c\u5b57\u4e49\u8868\u793a\u7684\u5b66\u4e60\uff0c\u5373\u5bf9\u4e24\u4e2a\u53e5\u5b50\u4e2d\u7684\u5b57\u5411\u91cf\u8fdb\u884c\u6ce8\u610f\u529b\u673a\u5236\u7684\u8ba1\u7b97\uff0c\u8fd9\u6837\u80fd\u591f 76.37 78.03 4.2 \u53c2 \u53c2 \u53c2\u6570 \u6570 \u6570\u8bbe \u8bbe \u8bbe\u7f6e \u7f6e \u7f6e Ours (without-structure) 75.70 76.78 \u5bf9\u4e24\u4e2a\u53e5\u5b50\u7684\u76f8\u5173\u4fe1\u606f\u8fdb\u884c\u5229\u7528\uff0c\u5df2\u6709\u5de5\u4f5c(Liu et al., 2018)\u8868\u660e\u4ea4\u53c9\u6ce8\u610f\u529b\u673a\u5236\u5bf9\u590d\u8ff0\u8bc6\u522b\u4efb \u6211\u4eec\u4f7f\u7528word2vec\u5de5\u5177\u5728gigaword\u751f\u8bed\u6599\u4e0a\u9884\u8bad\u7ec3\u5b57\u5411\u91cf\uff0c\u5b57\u5411\u91cf\u7ef4\u5ea6\u4e3a100\u7ef4\uff1bLSTM\u9690 \u85cf\u5c42\u7ef4\u5ea6\u4e3a400\uff0cDropout\u7387\u4e3a0.33\u3002\u6a21\u578b\u8bad\u7ec3\u4f7f\u7528Adam(Adaptive Moment Estimation)\u4f18\u5316 Ours (without-intermediate) 75.86 \u52a1\u7684\u6027\u80fd\u80fd\u591f\u5e26\u6765\u663e\u8457\u63d0\u5347\u3002\u672c\u7ae0\u4e2d\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u7684\u4e3b\u8981\u76ee\u7684\u5728\u4e8e\u5bf9\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u6a21\u578b\u7684\u6027\u80fd 77.07 \u8fdb\u884c\u8bc4\u4ef7\uff0c\u56e0\u6b64\u672a\u4f7f\u7528\u4ea4\u53c9\u6ce8\u610f\u529b\u673a\u5236\uff0c\u4ec5\u5bf9\u5355\u53e5\u4fe1\u606f\u8fdb\u884c\u5229\u7528\u30022)\u4ee5Bert\u4e3a\u4e3b\u7684\u9884\u8bad\u7ec3\u8bed\u8a00 \u7b97\u6cd5\uff0c\u4f9d\u5b58\u5206\u6790\u6a21\u578b\u521d\u59cb\u5b66\u4e60\u7387\u8bbe\u7f6e\u4e3a0.002\uff0c\u590d\u8ff0\u8bc6\u522b\u6a21\u578b\u521d\u59cb\u5b66\u4e60\u7387\u8bbe\u7f6e\u4e3a0.0001\u3002\u5bf9\u4e8e\u9884 \u8868 5: \u4f9d\u5b58\u4fe1\u606f\u5229\u7528\u5bf9\u6bd4\u7ed3\u679c \u6a21\u578b\u80fd\u591f\u663e\u8457\u63d0\u5347\u5b57\u5411\u91cf\u5728\u5177\u4f53\u53e5\u5b50\u4e2d\u8bed\u4e49\u8868\u793a\u7684\u7cbe\u5ea6\uff0c\u5e76\u4e14\u80fd\u591f\u5728\u5927\u89c4\u6a21\u8bad\u7ec3\u96c6\u4e2d\u63d0\u53d6\u4e30\u5bcc \u8bad\u7ec3\u7684\u6a21\u578b\u8bad\u7ec3\u65b9\u6cd5\uff0c\u4e3a\u4e86\u5bf9\u4f9d\u5b58\u5206\u6790\u90e8\u5206\u7684\u6a21\u578b\u53c2\u6570\u8fdb\u884c\u5fae\u8c03\uff0c\u6211\u4eec\u5bf9\u5176\u8bbe\u7f6e\u4e86\u4e00\u4e2a\u8f83\u5c0f\u7684 \u7684\u8bed\u8a00\u5b66\u77e5\u8bc6\uff0c\u4ee5\u6b64\u63d0\u5347\u4e0b\u6e38\u4efb\u52a1\u7684\u6027\u80fd\u3002\u672c\u7ae0\u4e3b\u8981\u9488\u5bf9\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u6a21\u578b\u4e2d\u7684\u7ed3\u6784\u5316\u4fe1\u606f\u5229 \u5b66\u4e60\u73870.00001\u3002 \u7528\u548c\u6a21\u578b\u8054\u5408\u65b9\u6cd5\u8fdb\u884c\u6539\u8fdb\uff0c\u4ec5\u4f7f\u7528word2vec\u9884\u8bad\u7ec3\u7684\u9759\u6001\u5b57\u5411\u91cf\uff0c\u4e14\u8bad\u7ec3\u6570\u636e\u8f83\u5c0f\u3002\u4eca\u540e\u53ef</td></tr><tr><td colspan=\"5\">\u4ece\u8868\u4e2d\u7ed3\u679c\u53ef\u4ee5\u770b\u51fa\uff0c\u53bb\u6389\u4f9d\u5b58\u7ed3\u6784\u4fe1\u606f\u548c\u53bb\u6389\u4f9d\u5b58\u4e2d\u95f4\u4fe1\u606f\u90fd\u4f1a\u5e26\u6765\u590d\u8ff0\u8bc6\u522b\u7cbe\u5ea6\u7684\u660e \u663e\u4e0b\u964d\u3002\u5176\u4e2d\u53bb\u9664\u4f9d\u5b58\u7ed3\u6784\u4fe1\u606f\u5e26\u6765\u7684\u6027\u80fd\u964d\u4f4e\u8f83\u4e3a\u660e\u663e\uff0c\u8868\u660e\u4f9d\u5b58\u53e5\u6cd5\u4fe1\u606f\u80fd\u591f\u63d0\u5347\u8bed\u4e49\u7ec4 \u5408\u8ba1\u7b97\u6a21\u578b\u7684\u6027\u80fd\uff1b\u53bb\u9664\u4f9d\u5b58\u4e2d\u95f4\u4fe1\u606f\u5e26\u6765\u7684\u6027\u80fd\u964d\u4f4e\u8868\u660e\uff0c\u6211\u4eec\u7684\u8bed\u4e49\u7ec4\u5408\u6a21\u578b\u80fd\u591f\u6709\u6548\u5229 \u4ee5\u5c1d\u8bd5\u5f15\u5165\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\uff0c\u8fdb\u4e00\u6b65\u9a8c\u8bc1\u6211\u4eec\u6240\u63d0\u65b9\u6cd5\u7684\u6709\u6548\u6027\u3002 5 \u603b \u603b \u603b\u7ed3 \u7ed3 \u7ed3 4.3 \u6211\u4eec\u5206\u522b\u5bf9\u7ba1\u9053\u6a21\u578b(pipeline)\u3001\u9884\u8bad\u7ec3\u65b9\u6cd5(pre-train)\u548c\u8fed\u4ee3\u8bad\u7ec3\u65b9\u6cd5(alternate) \u8fdb\u884c\u5b9e\u9a8c\uff0c\u5e76\u5728\u590d\u8ff0\u8bc6\u522b\u548c\u4e00\u4f53\u5316\u4f9d\u5b58\u5206\u6790\u4efb\u52a1\u4e0a\u8fdb\u884c\u4e86\u6bd4\u8f83\u3002\u5176\u4e2d\uff0c\u6211\u4eec\u5c06\u7ba1\u9053\u6a21\u578b\u4e2d\u4f9d\u5b58 \u7528\u4f9d\u5b58\u5206\u6790\u8fc7\u7a0b\u4e2d\u4ea7\u751f\u7684\u8bed\u4e49\u8868\u793a\uff0c\u5bf9\u6c49\u5b57\u8bed\u4e49\u8fdb\u884c\u9002\u5f53\u7684\u8865\u5145\uff0c\u63d0\u5347\u6c49\u5b57\u8868\u793a\u5305\u542b\u3002 \u672c\u6587\u9488\u5bf9\u73b0\u6709\u7ed3\u6784\u5316\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\u7684\u4e0d\u8db3\uff0c\u63d0\u51fa\u8054\u5408\u4f9d\u5b58\u5206\u6790\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\uff0c</td></tr><tr><td colspan=\"5\">4.3.3 \u8bed \u8bed \u8bed\u4e49 \u4e49 \u4e49\u7ec4 \u7ec4 \u7ec4\u5408 \u5408 \u5408\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5\u5bf9 \u5bf9 \u5bf9\u6bd4 \u6bd4 \u6bd4 \u5728\u73b0\u6709\u4f9d\u5b58\u5206\u6790\u6a21\u578b\u7684\u57fa\u7840\u4e0a\uff0c\u4f7f\u7528\u56fe\u6ce8\u610f\u529b\u7f51\u7edc\u6839\u636e\u4f9d\u5b58\u53e5\u6cd5\u6811\u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u5e76\u5229\u7528 \u5206\u6790\u90e8\u5206\u7684\u53c2\u6570\u5b66\u4e60\u7387\u8bbe\u7f6e\u4e3a0\uff0c\u7528\u4ee5\u6a21\u62df\u4f7f\u7528\u5916\u90e8\u4f9d\u5b58\u5206\u6790\u5668\u63d0\u4f9b\u7ed3\u6784\u4fe1\u606f\u7684\u4f20\u7edf\u7ed3\u6784\u5316\u65b9 \u6cd5\u3002 \u590d \u590d \u590d\u8ff0 \u8ff0 \u8ff0\u8bc6 \u8bc6 \u8bc6\u522b \u522b \u522b\u4efb \u4efb \u4efb\u52a1 \u52a1 \u52a1: \u4e09\u4e2a\u6a21\u578b\u5728\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e0a\u7684\u5bf9\u6bd4\u7ed3\u679c\u5982\u88683\u6240\u793a\uff0c\u5176\u4e2d\u9884\u8bad\u7ec3\u548c\u8fed\u4ee3\u8bad\u7ec3 \u4e24\u79cd\u65b9\u5f0f\u76f8\u8f83\u4e8e\u7ba1\u9053\u6a21\u578b\u5747\u6709\u660e\u663e\u7684\u63d0\u5347\uff0c\u4e14\u4f7f\u7528\u8fed\u4ee3\u8bad\u7ec3\u65b9\u5f0f\u7684\u6a21\u578b\u53d6\u5f97\u4e86\u6700\u597d\u7684\u7ed3\u679c\u3002 \u4f9d\u5b58\u5206\u6790\u4e2d\u95f4\u4fe1\u606f\u5bf9\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u7684\u5b57\u4e49\u8868\u793a\u8fdb\u884c\u8865\u5145\uff0c\u63d0\u5347\u6a21\u578b\u7684\u9c81\u68d2\u6027\u3002\u4eca\u540e\u6211\u4eec\u5c06\u5728\u73b0 \u6211\u4eec\u5728\u672c\u7ae0\u6240\u63d0\u7684\u57fa\u4e8e\u6ce8\u610f\u529b\u7684\u8bed\u4e49\u7ec4\u5408\u6a21\u578b\u4e2d\uff0c\u5bf9\u5b57\u4e49\u7ec4\u5408\u5c42\u548c\u53e5\u4e49\u8f93\u51fa\u5c42\u8fdb\u884c\u66ff\u6362\uff0c \u6709\u6a21\u578b\u4e2d\u52a0\u5165\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\uff0c\u63d0\u5347\u5b57\u4e49\u8868\u793a\u7684\u6027\u80fd\uff0c\u4ee5\u6b64\u6765\u8fdb\u4e00\u6b65\u63d0\u5347\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u6a21\u578b\u7684 \u5b9e\u73b0\u4e86\u5e38\u89c1\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\uff0c\u5e76\u4e0e\u672c\u7ae0\u6240\u63d0\u65b9\u6cd5\u5728\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e0a\u8fdb\u884c\u5bf9\u6bd4\u3002\u5bf9\u6bd4\u7684\u5e8f\u5217 \u5316\u65b9\u6cd5\u5305\u62ec\u5e73\u5747\u6c60\u5316(Mean)\u3001\u57fa\u4e8eCNN\u7684\u65b9\u6cd5(CNN)(Hu et al., 2015)\u548c\u57fa\u4e8eLSTM\u7684\u65b9 \u6027\u80fd\u3002</td></tr><tr><td colspan=\"5\">\u6cd5(LSTM)(Sutskever et al., 2014; Cho et al., 2014)\uff0c\u7ed3\u6784\u5316\u65b9\u6cd5\u5305\u62ec\u91c7\u7528\u5e76\u5217\u5316\u5904\u7406\u7684\u57fa\u4e8e</td></tr><tr><td colspan=\"5\">\u6a21\u578b Ours (pipeline) \u7ed3\u6784\u5316(\u7ba1\u9053) \u7c7b\u578b \u6811\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u7684\u65b9\u6cd5(Tree-CNN)(Mou et al., 2016)\u3002\u5176\u4e2d\u5e8f\u5217\u5316\u65b9\u6cd5\u5c06\u4e0d\u4f7f\u7528\u4f9d\u5b58\u5206\u6790\u6a21 Acc(%) F1(%) 72.64 75.74 \u578b\u63d0\u4f9b\u4fe1\u606f\uff0c\u7ed3\u6784\u5316\u65b9\u6cd5\u5c06\u4f9d\u5b58\u5206\u6790\u6a21\u578b\u7684\u53c2\u6570\u5b66\u4e60\u7387\u8bbe\u7f6e\u4e3a0\u3002\u5bf9\u6bd4\u7ed3\u679c\u5982\u88686\u6240\u793a\u3002 \u53c2 \u53c2 \u53c2\u8003 \u8003 \u8003\u6587 \u6587 \u6587\u732e \u732e \u732e</td></tr><tr><td colspan=\"5\">Ours (pre-train) \u7ed3\u6784\u5316(\u8054\u5408) Ours (alternate) \u4ece\u5bf9\u6bd4\u7ed3\u679c\u53ef\u4ee5\u770b\u51fa\uff0c\u6211\u4eec\u7684\u6a21\u578b\u8f83\u73b0\u6709\u5e38\u89c1\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\uff0c\u5728\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e0a\u7684 74.01 76.86 76.37 78.03 \u9884\u6d4b\u51c6\u786e\u7387\u548cF1\u503c\u5747\u6709\u8f83\u660e\u663e\u7684\u63d0\u5347\u3002\u5176\u4e2d\u8f83\u6700\u597d\u7684LSTM\u65b9\u6cd5\u5206\u522b\u63d0\u5347\u4e860.83%\u548c1.68%\u3002\u8868</td></tr><tr><td colspan=\"5\">\u660e\u6211\u4eec\u7684\u65b9\u6cd5\u80fd\u591f\u6709\u6548\u5730\u5229\u7528\u4f9d\u5b58\u53e5\u6cd5\u4fe1\u606f\u548c\u4f9d\u5b58\u4e2d\u95f4\u4fe1\u606f\uff0c\u4ece\u800c\u83b7\u53d6\u66f4\u4e3a\u51c6\u786e\u7684\u8bed\u4e49\u8868\u793a\uff0c \u8868 3: \u8054\u5408\u6a21\u578b\u8bad\u7ec3\u65b9\u5f0f\u5728\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e0a\u7684\u5bf9\u6bd4\u7ed3\u679c \u5e76\u53cd\u5e94\u5728\u4e0b\u6e38\u4efb\u52a1\u4e2d\u3002</td></tr><tr><td colspan=\"5\">\u6b64\u5916\uff0c\u6211\u4eec\u5b9e\u73b0\u7684\u7ed3\u6784\u5316\u65b9\u6cd5Tree-CNN\u548c\u6211\u4eec\u7684\u7ba1\u9053\u6a21\u578b\u5728\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e0a\u6027\u80fd\u63a5\u8fd1\uff0c\u4f46</td></tr><tr><td colspan=\"5\">\u4e00 \u4e00 \u4e00\u4f53 \u4f53 \u4f53\u5316 \u5316 \u5316\u4f9d \u4f9d \u4f9d\u5b58 \u5b58 \u5b58\u5206 \u5206 \u5206\u6790 \u6790 \u6790: \u4e09\u4e2a\u6a21\u578b\u5728\u4e00\u4f53\u5316\u4f9d\u5b58\u5206\u6790\u4efb\u52a1\u4e0a\u7684\u5bf9\u6bd4\u7ed3\u679c\u5982\u88684\u6240\u793a\uff0c\u5176\u4e2d\u4e24\u79cd\u8bad\u7ec3 \u8f83\u5e8f\u5217\u5316\u65b9\u6cd5\u7684\u6027\u80fd\u6709\u8f83\u660e\u663e\u7684\u4e0b\u964d\u3002\u6211\u4eec\u8ba4\u4e3a\u8fd9\u662f\u7531\u4e8e\u4f9d\u5b58\u5206\u6790\u548c\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u7684\u6570\u636e\u9886\u57df</td></tr><tr><td colspan=\"5\">\u65b9\u5f0f\u5f97\u5230\u7684\u6a21\u578b\u90fd\u8f83\u53c2\u6570\u8c03\u6574\u524d\u7684\u4f9d\u5b58\u5206\u6790\u6a21\u578b\u6027\u80fd\u66f4\u4f4e\uff0c\u4e14\u4f7f\u7528\u8fed\u4ee3\u8bad\u7ec3\u65b9\u5f0f\u7684\u6a21\u578b\u964d\u4f4e\u66f4 \u4e0d\u4e00\u81f4\u7684\u95ee\u9898\uff0c\u5bf9\u7ed3\u6784\u5316\u8bed\u4e49\u7ec4\u5408\u65b9\u6cd5\u5e26\u6765\u7684\u5de8\u5927\u5f71\u54cd\uff0c\u4fa7\u9762\u53cd\u6620\u4e86\u6211\u4eec\u63d0\u51fa\u7684\u8054\u5408\u6a21\u578b\u80fd\u591f</td></tr><tr><td>\u4e3a\u660e\u663e\u3002 \u6709\u6548\u964d\u4f4e\u6570\u636e\u9886\u57df\u4e0d\u4e00\u81f4\u7684\u95ee\u9898\u3002</td><td/><td/><td/><td/></tr></table>", | |
| "text": "\u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u7ed3 \u7ed3 \u7ed3\u679c \u679c \u679c\u4e0e \u4e0e \u4e0e\u5206 \u5206 \u5206\u6790 \u6790 \u6790 4.3.1 \u8054 \u8054 \u8054\u5408 \u5408 \u5408\u6a21 \u6a21 \u6a21\u578b \u578b \u578b\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u65b9 \u65b9 \u65b9\u5f0f \u5f0f \u5f0f\u5bf9 \u5bf9 \u5bf9\u6bd4 \u6bd4 \u6bd4 \u4e0e \u4e0e \u4e0e\u73b0 \u73b0 \u73b0\u6709 \u6709 \u6709\u590d \u590d \u590d\u8ff0 \u8ff0 \u8ff0\u8bc6 \u8bc6 \u8bc6\u522b \u522b \u522b\u6a21 \u6a21 \u6a21\u578b \u578b \u578b\u7684 \u7684 \u7684\u5bf9 \u5bf9 \u5bf9\u6bd4 \u6bd4 \u6bd4 \u6211\u4eec\u4e0e\u73b0\u5728\u5e38\u7528\u7684\u590d\u8ff0\u8bc6\u522b\u6a21\u578b(Wang et al., 2017;Devlin et al., 2019)\u5728LCQMC\u6570\u636e\u96c6 \u4e0a\u7684\u6700\u597d\u7ed3\u679c\u8fdb\u884c\u4e86\u6bd4\u8f83\uff0c\u7ed3\u679c\u5982\u88687\u6240\u793a\u3002 Qian Chen, Xiaodan Zhu, Zhen-Hua Ling, Si Wei, Hui Jiang, and Diana Inkpen. 2017. Enhanced LSTM for natural language inference. In Regina Barzilay and Min-Yen Kan, editors, Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, ACL 2017, Vancouver, Canada, July 30 -August 4, Volume 1: Long Papers, pages 1657-1668. Association for Computational Linguistics. Kyunghyun Cho, Bart van Merrienboer, \u00c7 aglar G\u00fcl\u00e7ehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning phrase representations using RNN encoder-decoder for statistical machine translation. In Alessandro Moschitti, Bo Pang, and Walter Daelemans, editors, Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, EMNLP", | |
| "type_str": "table", | |
| "num": null, | |
| "html": null | |
| } | |
| } | |
| } | |
| } |