| { | |
| "paper_id": "2020", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T12:53:20.277354Z" | |
| }, | |
| "title": "Joint Learning Chinese Dependency Parsing and Semantic Composition based on Graph Neural Network", | |
| "authors": [ | |
| { | |
| "first": "Kai", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Beijing Jiaotong University", | |
| "location": { | |
| "postCode": "10004", | |
| "settlement": "Beijing" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Mingtong", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Beijing Jiaotong University", | |
| "location": { | |
| "postCode": "10004", | |
| "settlement": "Beijing" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "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": "The semantics of a sentence is composed of the meaning of its constituent components and the combination method. Therefore, syntax-based semantic composition has always been an important research direction in NLP. The semantic composition method using tree structure has became the most representative method(Tai et al., 2015). However, such methods are difficult to be applied to large-scale data. The main problem is that the order of its semantic composition depends on the structure of the specific tree, and parallel computation cannot be supported. In this paper, we present a joint framework for graph-based dependency parsing and semantic composition. The model does not need to rely on an external syntax parser for providing structural information, and the semantic composition method based on graph neural network can support parallel computation, which greatly reduces the computation time. Moreover, the joint learning of two tasks enables the model to learn the syntactic structure and semantic contextual information. Experimental results on LCQMC(Liu et al., 2018) dataset show that the \u56fd\u5bb6\u81ea\u7136\u79d1\u5b66\u57fa\u91d1(61876198,61976015,61976016)\u8d44\u52a9 \u8ba1\u7b97\u8bed\u8a00\u5b66 accuracy is close to the tree-based semantics composition method, reaching 79.54%, and the prediction speed is increased by up to 30 times.", | |
| "pdf_parse": { | |
| "paper_id": "2020", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "The semantics of a sentence is composed of the meaning of its constituent components and the combination method. Therefore, syntax-based semantic composition has always been an important research direction in NLP. The semantic composition method using tree structure has became the most representative method(Tai et al., 2015). However, such methods are difficult to be applied to large-scale data. The main problem is that the order of its semantic composition depends on the structure of the specific tree, and parallel computation cannot be supported. In this paper, we present a joint framework for graph-based dependency parsing and semantic composition. The model does not need to rely on an external syntax parser for providing structural information, and the semantic composition method based on graph neural network can support parallel computation, which greatly reduces the computation time. Moreover, the joint learning of two tasks enables the model to learn the syntactic structure and semantic contextual information. Experimental results on LCQMC(Liu et al., 2018) dataset show that the \u56fd\u5bb6\u81ea\u7136\u79d1\u5b66\u57fa\u91d1(61876198,61976015,61976016)\u8d44\u52a9 \u8ba1\u7b97\u8bed\u8a00\u5b66 accuracy is close to the tree-based semantics composition method, reaching 79.54%, and the prediction speed is increased by up to 30 times.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "1 \u5f15 \u5f15 \u5f15\u8a00 \u8a00 \u8a00 \u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u6280\u672f\u4e3a\u81ea\u7136\u8bed\u8a00\u5904\u7406\u53d1\u5c55\u5e26\u6765\u5d2d\u65b0\u5efa\u6a21\u65b9\u5f0f\u548c\u6027\u80fd\u4e0a\u7684\u5de8\u5927\u63d0\u5347\uff0c\u6210\u4e3a\u4e3b \u6d41\u7684\u7814\u7a76\u65b9\u6cd5\uff0c\u5176\u4e2d\u8bed\u4e49\u8868\u793a\u662f\u7814\u7a76\u70ed\u70b9\u4e4b\u4e00\u3002\u5df2\u6709\u7814\u7a76\u8868\u660e\u6709\u6548\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u6a21\u578b\uff0c \u5982LSTM\uff0cCNN\uff0cTree-LSTM\u7b49\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\uff0c\u53ef\u4ee5\u63d0\u5347\u81ea\u7136\u8bed\u8a00\u5904\u7406\u5e94\u7528\u7684\u6027\u80fd\uff0c\u5982\uff1a\u673a\u5668 \u7ffb\u8bd1 (Callison-Burch et al., 2006) \u3001\u60c5\u611f\u5206\u6790 (Tai et al., 2015 )\u3001\u590d\u8ff0\u8bc6\u522b(Fan et al., 2018 )\u3001\u81ea \u7136\u8bed\u8a00\u63a8\u7406 (Mou et al., 2016) \u7b49\u3002 \u57fa\u4e8e\u5e8f\u5217\u5316\u7ed3\u6784\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\u7b80\u5355\u6709\u6548\uff0c\u88ab\u5e7f\u6cdb\u91c7\u7528 (Mueller and Thyagarajan, 2016) \uff0c\u4f46\u662f\uff0c\u8fd9\u79cd\u65b9\u6cd5\u6ca1\u6709\u8003\u8651\u53e5\u6cd5\u7ed3\u6784\u4fe1\u606f\uff0c\u96be\u4ee5\u6355\u83b7\u8bcd\u5e8f\u5b8c\u5168\u76f8\u540c\u53e5\u6cd5\u7ed3\u6784\u4e0d\u540c\u7684\u53e5 \u5b50\u4e4b\u95f4\u7684\u5dee\u5f02\u3002\u6bd4\u5982\u53e5\u5b50\"\u653e\u5f03\u7f8e\u4e3d\u7684\u5973\u4eba\u8ba9\u4eba\u5fc3\u788e\u3002\"\uff0c\u53ef\u4ee5\u6709\u4e24\u79cd\u53e5\u6cd5\u7ed3\u6784\uff0c\u5982\u56fe1\u6240\u793a\u3002 \u5728\u56fe1(a)\u7684\u53e5\u6cd5\u7ed3\u6784\u4e2d\uff0c\"\u7f8e\u4e3d\"\u4f5c\u4e3a\u5f62\u5bb9\u8bcd\u4fee\u9970\"\u5973\u4eba\"\uff0c\"\u653e\u5f03\"\u7684\u5bf9\u8c61\u662f\"\u7f8e\u4e3d\u7684\u5973\u4eba\"\uff1b\u5728 \u56fe1(b)\u7684\u53e5\u6cd5\u7ed3\u6784\u4e2d\uff0c\"\u7f8e\u4e3d\"\u4f5c\u4e3a\u540d\u8bcd\uff0c\"\u653e\u5f03\"\u7684\u5bf9\u8c61\u662f\"\u7f8e\u4e3d\"\u3002\u7531\u6b64\u53ef\u89c1\uff0c\u53e5\u6cd5\u7ed3\u6784\u51b3\u5b9a\u53e5 \u5b50\u8bed\u4e49\uff0c\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a\u5e94\u8be5\u8003\u8651\u5176\u53e5\u6cd5\u7ed3\u6784\u3002 \u968f\u540e\uff0c\u7814\u7a76\u4eba\u5458\u5f00\u59cb\u5173\u6ce8\u57fa\u4e8e\u53e5\u6cd5\u7ed3\u6784\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5 (Tai et al., 2015; Chen et al., 2017; Mou et al., 2016) (Tang et al., 2016; Mueller and Thyagarajan, 2016) \u3002\u8fd9\u4e9b\u65b9\u6cd5\u7684\u4f18\u70b9\u662f\u53ef\u4ee5 \u8868\u793a\u4efb\u610f\u957f\u5ea6\u53e5\u5b50\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\uff0c\u5176\u8bed\u4e49\u4fe1\u606f\u4ece\u5de6\u5411\u53f3\u79ef\u7d2f\uff0c\u6700\u7ec8\u5c06\u6700\u540e\u65f6\u523b\u7684\u9690\u72b6\u6001\u5411\u91cf \u89c6\u4e3a\u6574\u4e2a\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a\u3002Kim (2014)\u5229\u7528\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u83b7\u5f97\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a\uff0c\u5177\u4f53\u505a\u6cd5\u662f \u4f7f\u7528\u5377\u79ef\u6838\u5728\u8f93\u5165\u53e5\u4e0a\u4ece\u5de6\u5230\u53f3\u6ed1\u52a8\uff0c\u6bcf\u6b21\u6ed1\u52a8\u6355\u6349\u53e5\u5b50\u5c40\u90e8\u533a\u57df\u7684\u7279\u5f81\uff0c\u4f7f\u5f97CNN\u66f4\u80fd\u6355 \u83b7n-gram\u7279\u5f81\uff0c\u6700\u540e\u901a\u8fc7\u6700\u5927\u6c60\u5316\u83b7\u5f97\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a\u3002\u57fa\u4e8e\u5e8f\u5217\u5316\u65b9\u6cd5\u4e3a\u4e86\u8ffd\u6c42\u8fd0\u7b97\u6548\u7387\uff0c \u76f4\u63a5\u5bf9\u53e5\u5b50\u7684\u6587\u672c\u5e8f\u5217\u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u672a\u5bf9\u7ed3\u6784\u4fe1\u606f\u52a0\u4ee5\u5229\u7528\uff0c\u96be\u4ee5\u5bf9\u7ed3\u6784\u4e0d\u540c\u5e26\u6765\u7684\u8bed \u4e49\u5dee\u5f02\u52a0\u4ee5\u533a\u5206\u3002 \u8fd1\u5e74\u6765\u6709\u8bb8\u591a\u5de5\u4f5c (Socher et al., 2012; Li et al., 2015; Tai et al., 2015 )\u8bd5\u56fe\u5f15\u5165\u53e5\u6cd5\u7ed3\u6784 \u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u5e76\u5728\u60c5\u611f\u5206\u7c7b (Tai et al., 2015) ", | |
| "cite_spans": [ | |
| { | |
| "start": 143, | |
| "end": 172, | |
| "text": "(Callison-Burch et al., 2006)", | |
| "ref_id": "BIBREF2" | |
| }, | |
| { | |
| "start": 179, | |
| "end": 196, | |
| "text": "(Tai et al., 2015", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 197, | |
| "end": 220, | |
| "text": ")\u3001\u590d\u8ff0\u8bc6\u522b(Fan et al., 2018", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 231, | |
| "end": 249, | |
| "text": "(Mou et al., 2016)", | |
| "ref_id": "BIBREF15" | |
| }, | |
| { | |
| "start": 280, | |
| "end": 311, | |
| "text": "(Mueller and Thyagarajan, 2016)", | |
| "ref_id": "BIBREF16" | |
| }, | |
| { | |
| "start": 536, | |
| "end": 554, | |
| "text": "(Tai et al., 2015;", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 555, | |
| "end": 573, | |
| "text": "Chen et al., 2017;", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 574, | |
| "end": 591, | |
| "text": "Mou et al., 2016)", | |
| "ref_id": "BIBREF15" | |
| }, | |
| { | |
| "start": 592, | |
| "end": 611, | |
| "text": "(Tang et al., 2016;", | |
| "ref_id": "BIBREF20" | |
| }, | |
| { | |
| "start": 612, | |
| "end": 642, | |
| "text": "Mueller and Thyagarajan, 2016)", | |
| "ref_id": "BIBREF16" | |
| }, | |
| { | |
| "start": 889, | |
| "end": 910, | |
| "text": "(Socher et al., 2012;", | |
| "ref_id": "BIBREF17" | |
| }, | |
| { | |
| "start": 911, | |
| "end": 927, | |
| "text": "Li et al., 2015;", | |
| "ref_id": "BIBREF12" | |
| }, | |
| { | |
| "start": 928, | |
| "end": 944, | |
| "text": "Tai et al., 2015", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 971, | |
| "end": 989, | |
| "text": "(Tai et al., 2015)", | |
| "ref_id": "BIBREF19" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "w 2 w 3 w 4 w (a) \u5e8f\u5217\u5316\u7684\u8bed\u4e49\u7ec4\u5408 2 w 4 w 3 w 1 w (b) \u6811\u7ed3\u6784\u7684\u8bed\u4e49\u7ec4\u5408 2 w 4 w 3 w 1 w 2 w 4 w 3 w 1 w (c) \u56fe\u7ed3\u6784\u7684\u8bed\u4e49\u7ec4\u5408 \u56fe 2: \u4e0d\u540c\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5 \u5355\u8bcd\u95f4\u8bed\u4e49\u4fee\u9970\u5173\u7cfb\u7684\u4f9d\u5b58\u5173\u7cfb\uff0c\u672c\u6587\u91c7\u7528\u4f9d\u5b58\u7ed3\u6784\u4fe1\u606f\u6307\u5bfc\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u3002\u6211\u4eec\u8bbe\u8ba1\u57fa\u4e8e\u56fe \u7684\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u65b9\u6cd5\uff0c\u7528\u4e8e\u751f\u6210\u5e26\u6709\u6982\u7387\u7684\u4f9d\u5b58\u5173\u7cfb\u7ed3\u6784\u56fe\uff0c\u5305\u542b\u6240\u6709\u5355\u8bcd\u5bf9\u95f4\u7684\u6709\u5411\u4f9d\u5b58\u5f27 \u7684\u6982\u7387\u3002\u7136\u540e\uff0c\u6211\u4eec\u4f7f\u7528\u4f9d\u5b58\u5173\u7cfb\u6982\u7387\u4f5c\u4e3a\u56fe\u795e\u7ecf\u7f51\u7edc\u4e2d\u4fe1\u606f\u4f20\u9012\u7684\u6743\u503c\uff0c\u4e00\u65b9\u9762\u4f7f\u5f97\u8bed\u4e49\u7ec4 \u5408\u8ba1\u7b97\u53ef\u4ee5\u6309\u7167\u5355\u8bcd\u95f4\u4fee\u9970\u5173\u7cfb\u6982\u7387\u7ed3\u5408\u6240\u6709\u5355\u8bcd\u7684\u8bed\u4e49\uff1b\u53e6\u4e00\u65b9\u9762\uff0c\u57fa\u4e8e\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u8bed\u4e49 \u7ec4\u5408\u8ba1\u7b97\u53ef\u4ee5\u5b9e\u73b0\u5e76\u884c\u5904\u7406\uff0c\u652f\u6301\u8bad\u7ec3\u548c\u9884\u6d4b\u9636\u6bb5\u7684\u6279\u5904\u7406\uff0c\u63d0\u5347\u8ba1\u7b97\u6548\u7387\u3002\u672c\u6587\u6a21\u578b\u4e0e\u5df2\u6709 \u65b9\u6cd5\u5bf9\u6bd4\u793a\u610f\u56fe\u5982\u56fe2\u6240\u793a\u3002\u5176\u4e2d\uff0c\u56fe2(a)\u662f\u57fa\u4e8e\u5e8f\u5217\u5316\u7ed3\u6784\u7684\u7ec4\u5408\u65b9\u6cd5\uff0c\u56fe2(b)\u662f\u57fa\u4e8e\u6811\u7ed3\u6784 \u7684\u8bed\u4e49\u7ec4\u5408\u65b9\u6cd5\uff0c\u56fe2(c)\u662f\u672c\u6587\u63d0\u51fa\u7684\u57fa\u4e8e\u56fe\u7ed3\u6784\u7684\u8bed\u4e49\u7ec4\u5408\u65b9\u6cd5\u3002\u6211\u4eec\u4ee5\u590d\u8ff0\u8bc6\u522b\u4f5c\u4e3a\u8bed\u4e49 \u7ec4\u5408\u8ba1\u7b97\u7684\u76ee\u6807\u4efb\u52a1\uff0c\u5728\u516c\u5f00\u6c49\u8bed\u590d\u8ff0\u8bc6\u522b\u6570\u636e\u96c6LCQMC(Liu et al., 2018) \u4e0a\u7684\u5b9e\u9a8c\u7ed3\u679c\u8868 \u660e\uff0c\u672c\u6587\u63d0\u51fa\u7684\u6a21\u578b\u4f18\u4e8e\u5df2\u6709\u5e8f\u5217\u7ed3\u6784\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\uff0c\u53ef\u4ee5\u6709\u6548\u6539\u8fdb\u590d\u8ff0\u8bc6\u522b\u6027\u80fd\u3002\u540c \u65f6\uff0c\u672c\u6587\u63d0\u51fa\u7684\u8bed\u4e49\u7ec4\u5408\u65b9\u5f0f\u652f\u6301\u6279\u5904\u7406\u64cd\u4f5c\uff0c\u5728\u9884\u6d4b\u9636\u6bb5\u901f\u5ea6\u662fTree-LSTM\u768430\u500d\uff0c\u540c\u65f6\u80fd \u4fdd\u6301\u8f83\u9ad8\u7684\u7cbe\u5ea6\u3002 \u672c\u6587\u7684\u4e3b\u8981\u8d21\u732e\u5982\u4e0b\uff1a(1)\u63d0\u51fa\u4e00\u79cd\u57fa\u4e8e\u56fe\u7684\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u548c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u7684\u8054\u5408\u6846 \u67b6\uff1b(2)\u63d0\u51fa\u4e00\u79cd\u57fa\u4e8e\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u8bed\u4e49\u7ec4\u5408\u65b9\u6cd5\uff0c\u53ef\u4ee5\u5b9e\u73b0\u5e76\u884c\u5904\u7406\uff0c\u652f\u6301\u8bad\u7ec3\u548c\u9884\u6d4b\u9636 \u6bb5\u7684\u6279\u5904\u7406\uff0c\u63d0\u5347\u8ba1\u7b97\u6548\u7387\u3002 2 \u76f8 \u76f8 \u76f8\u5173 \u5173 \u5173\u5de5 \u5de5 \u5de5\u4f5c \u4f5c \u4f5c \u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\u4e3b\u8981\u5206\u4e3a\u57fa\u4e8e\u5e8f\u5217\u5316\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\u548c\u57fa\u4e8e\u53e5\u6cd5\u7ed3\u6784\u7684\u8bed\u4e49\u7ec4 \u5408\u8ba1\u7b97\u65b9\u6cd5\u3002\u5728\u5e8f\u5217\u5316\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u4e2d\uff0c\u5982\u56fe2(a)\u6240\u793a\uff0c\u6a21\u578b\u4ece\u5de6\u5230\u53f3\u4f9d\u6b21\u8bfb\u5165\u5355\u8bcd\uff0c \u5982LSTM,RNN\u7b49", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "p 1 p 2 p 3 p p v \u56fe\u795e\u7ecf\u7f51\u7edc \u8bed\u4e49\u7ec4\u5408 \u6a21\u5757 \u57fa\u4e8e\u56fe \u4f9d\u5b58\u53e5\u6cd5 \u5206\u6790\u6a21\u5757 0 q 1 q 2 q 3 q <root> \u5982\u4f55 \u9009\u62e9 \u624b\u673a 0 p 1 p 2 p 3 p <root> \u600e\u4e48 \u9009\u62e9 \u624b\u673a 0 q 1 q 2 q 3 q q v \u53e5\u5bf9P\u548cQ \u53e5\u6cd5\u5206\u6790\u4e0e\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97 \u53e5\u6cd5\u6811\u4e0e\u53e5\u5b50\u8bed\u4e49\u8868\u793a \u53e5\u5bf9 \u5206\u7c7b\u5668 \u56fe 3: \u672c\u6587\u63d0\u51fa\u7684\u8054\u5408\u6a21\u578b\u6574\u4f53\u6846\u67b6 3 \u57fa \u57fa \u57fa\u4e8e \u4e8e \u4e8e\u56fe \u56fe \u56fe\u795e \u795e \u795e\u7ecf \u7ecf \u7ecf\u7f51 \u7f51 \u7f51\u7edc \u7edc \u7edc\u7684 \u7684 \u7684\u4f9d \u4f9d \u4f9d\u5b58 \u5b58 \u5b58\u5206 \u5206 \u5206\u6790 \u6790 \u6790\u548c \u548c \u548c\u8bed \u8bed \u8bed\u4e49 \u4e49 \u4e49\u7ec4 \u7ec4 \u7ec4\u5408 \u5408 \u5408\u8ba1 \u8ba1 \u8ba1\u7b97 \u7b97 \u7b97\u8054 \u8054 \u8054\u5408 \u5408 \u5408\u6a21 \u6a21 \u6a21\u578b \u578b \u578b \u6211\u4eec\u91c7\u7528\u4f9d\u5b58\u53e5\u6cd5\u6811\u4f5c\u4e3a\u53e5\u6cd5\u7ed3\u6784\u6307\u5bfc\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u63d0\u51fa\u4e86\u57fa\u4e8e\u56fe\u7684\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790 \u548c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u7684\u8054\u5408\u6846\u67b6\uff0c\u6a21\u578b\u67b6\u6784\u5982\u56fe3\u6240\u793a\u3002\u6a21\u578b\u63a5\u6536\u53e5\u5bf9P = {p 1 , . . . , p N }\u548cQ = {q 1 , . . . , q M }\u3002\u9996\u5148\u7ecf\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u5206\u522b\u5f97\u5230\u5e26\u6709\u6982\u7387\u7684\u4f9d\u5b58\u5173\u7cfb\u7ed3\u6784\u56fe\uff0c\u5e76\u4ece\u4e2d\u5f97\u5230\u4f9d\u5b58\u6811\uff0c \u7136\u540e\u7ecf\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u5229\u7528\u8be5\u56fe\u5f97\u5230\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a\uff0c\u5e76\u9001\u5165\u53e5\u5bf9\u5206\u7c7b\u5668\u8fdb\u884c\u5224\u65ad\u3002\u5728\u6a21\u578b\u8bad\u7ec3 \u9636\u6bb5\uff0c\u6211\u4eec\u8054\u5408\u4f9d\u5b58\u5206\u6790\u548c\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e24\u4e2a\u76ee\u6807\u5171\u540c\u5b66\u4e60\u6a21\u578b\u53c2\u6570\u3002 3.1 \u4f9d \u4f9d \u4f9d\u5b58 \u5b58 \u5b58\u53e5 \u53e5 \u53e5\u6cd5 \u6cd5 \u6cd5\u5206 \u5206 \u5206\u6790 \u6790 \u6790 \u672c\u6587\u91c7\u7528\u57fa\u4e8e\u56fe\u7684\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u65b9\u6cd5(Dozat and Manning, 2017)\uff0c\u8be5\u65b9\u6cd5\u53ef\u4ee5\u8003\u8651\u5168\u5c40\u4fe1 \u606f\u8fdb\u884c\u4f9d\u5b58\u5206\u6790\u51b3\u7b56\uff0c\u6700\u8fd1\u7814\u7a76\u663e\u793a\u8be5\u65b9\u6cd5\u5728\u6027\u80fd\u4e0a\u8d85\u8fc7\u4e86\u57fa\u4e8e\u8f6c\u79fb\u7684\u4f9d\u5b58\u5206\u6790\u65b9\u6cd5(Ji et al., 2019)\u3002\u4e0b\u9762\uff0c\u6211\u4eec\u4ee5\u53e5\u5b50P = {p 0 , p 1 , . . . , p N }\u4e3a\u4f8b\uff0c\u8be6\u7ec6\u4ecb\u7ecd\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u6a21\u5757\u3002\u6309\u7167\u901a\u5e38\u505a \u6cd5\uff0c\u6211\u4eec\u5728\u6bcf\u4e2a\u53e5\u5b50\u7684\u5f00\u5934\u52a0\u5165\u6839\u8282\u70b9\u7684\u6807\u8bc6\"<root>\"\u4f5c\u4e3ap 0 \u3002 \u9996\u5148\u5c06\u8f93\u5165\u7684\u5355\u8bcd\u5e8f\u5217\u8f6c\u5316\u4e3a\u6570\u503c\u5411\u91cf\u8868\u793a\uff0c\u6211\u4eec\u91c7\u7528\u9884\u8bad\u7ec3\u8bcd\u5411\u91cf\u3001\u968f\u673a\u521d\u59cb\u5316\u8bcd\u5411\u91cf \u548c\u8bcd\u6027\u6807\u7b7e\u5411\u91cf\u4e09\u90e8\u5206\u6784\u6210\u8f93\u5165\u8bcd\u5411\u91cf\u3002\u6211\u4eec\u7528e(p i ) \u2208 R d \u8868\u793a\u9884\u8bad\u7ec3\u8bcd\u5411\u91cf\uff0ce (p i ) \u2208 R d \u8868\u793a \u968f\u673a\u521d\u59cb\u5316\u8bcd\u5411\u91cf\uff0ce(pos i ) \u2208 R dpos \u8868\u793a\u8bcd\u6027\u6807\u7b7e\u5411\u91cf\uff0cd pos \u4e3a\u8bcd\u6027\u7684\u5d4c\u5165\u7ef4\u5ea6\uff0c\u4e09\u90e8\u5206\u7684\u8868\u793a \u5728\u8bad\u7ec3\u4e2d\u88ab\u66f4\u65b0\u3002\u6700\u7ec8\uff0c\u6bcf\u4e2a\u5355\u8bcd\u7684\u8868\u793a\u7531\u516c\u5f0f1\u8ba1\u7b97\u5f97\u51fa\uff0c\u5176\u4e2d\u2295\u4e3a\u62fc\u63a5\u64cd\u4f5c\u3002 x i = (e(p i ) + e (p i )) \u2295 e(pos i ) (1) \u4e3a\u4e86\u6355\u6349\u53e5\u5b50\u957f\u8ddd\u79bb\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\uff0c\u6211\u4eec\u91c7\u7528\u6df1\u5c42\u53cc\u5411LSTM(BiLSTM)\u5b66\u4e60\u53e5\u5b50\u4e2d\u7684\u8bcd\u8868 \u793a\u3002\u5176\u4e2d\uff0c\u7b2ci\u65f6\u523b(\u5bf9\u5e94\u7b2ci\u4e2a\u5355\u8bcd)\u7684\u9690\u85cf\u72b6\u6001\u8868\u793a\u5982\u516c\u5f0f2\u6240\u793a\u3002 h i = BiLST M (x i , \u2190 \u2212 h i+1 , \u2212 \u2192 h i\u22121 , \u03b8) (2) \u5176\u4e2d\uff0c \u2190 \u2212 h i \u548c \u2212 \u2192 h i \u662f\u5728\u65f6\u523bi \u524d\u5411\u548c\u9006\u5411LSTM\u7684\u9690\u85cf\u8868\u793a\uff1b\u03b8 \u4e3aBiLSTM\u4e2d\u7684\u53c2\u6570\u3002 \u672c\u6587\u4f7f\u7528\u56feG = (V, E)\u8868\u793a\u53e5\u5b50P \u7684\u4f9d\u5b58\u5173\u7cfb\u56fe\uff0c\u5176\u4e2dV = {p 0 , p 1 , . . . , p N }\u662f\u53e5\u5b50\u4e2d\u5355\u8bcd \u8282\u70b9\u96c6\u5408\uff0cE\u662f\u4f9d\u5b58\u5173\u7cfb\u8fb9\u96c6\u5408\u3002\u5e8f\u5217P \u4e2d\u6bcf\u4e2a\u8bcd\u4e0e\u56fe\u4e0a\u7684\u8282\u70b9\u5bf9\u5e94\uff0c\u4f7f\u7528p j \u2192 p i \u8868\u793a\u6838\u5fc3 \u8bcd(head)p j \u4e0e\u4f9d\u5b58\u8bcd(dep)p i \u4e4b\u95f4\u5b58\u5728\u4f9d\u5b58\u5173\u7cfb\u3002\u7531\u4e8e\u53e5\u5b50\u4e2d\u4efb\u610f\u4e24\u4e2a\u5355\u8bcd\u4e4b\u95f4\u5b58\u5728\u4e24\u79cd\u4f9d\u5b58\u5173 \u7cfbp j \u2192 p i \u548cp i \u2192 p j \uff0c\u9700\u8981\u4e3a\u6bcf\u4e2a\u5355\u8bcd\u8ba1\u7b97\u5176\u4f5c\u4e3a\u6838\u5fc3\u8bcd\u6216\u4f9d\u5b58\u8bcd\u7684\u5411\u91cf\u8868\u793a\u3002\u4e3a\u6b64\uff0c\u6211\u4eec \u4e3a\u6bcf\u4e2a\u5355\u8bcd\u8bbe\u7f6e\u4e24\u4e2a\u5411\u91cf\u8868\u793a\uff0c\u4e00\u4e2a\u662f\u5355\u8bcd\u4f5c\u4e3a\u4f9d\u5b58\u8bcd\u7684\u8868\u793a\uff0c\u53e6\u4e00\u4e2a\u662f\u5355\u8bcd\u4f5c\u4e3a\u6838\u5fc3\u8bcd\u7684 \u8868\u793a\u3002\u5bf9\u4e8e\u8fd9\u4e24\u79cd\u8868\u793a\u7684\u8ba1\u7b97\uff0c\u6211\u4eec\u5206\u522b\u91c7\u7528\u591a\u5c42\u611f\u77e5\u5668\u5bf9BiLSTM\u7684\u8f93\u51fah i \u8fdb\u884c\u8ba1\u7b97\uff0c\u5982\u516c \u5f0f3\u548c4\u6240\u793a(Dozat", | |
| "cite_spans": [], | |
| "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": "\u56fe 4: \u57fa\u4e8e\u56fe\u795e\u7ecf\u7f51\u7edc\u4f7f\u7528\u7ed3\u6784\u4fe1\u606f\u7684\u4e24\u79cd\u65b9\u5f0f \u793ap j \u2192 p i \u7684\u5f97\u5206\uff0c\u5f97\u5206\u8d8a\u5927\u8868\u793a\u6784\u6210p j \u2192 p i \u7684\u53ef\u80fd\u6027\u8d8a\u5927\u3002 r dep i = M LP (dep) (h i )", | |
| "eq_num": "(3)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "r head j = M LP (head) (h j )", | |
| "eq_num": "(4)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "s ij = r dep i U r head j + r head j u (5) \u5176\u4e2d\uff0cU \u8868\u793a\u6743\u91cd\u77e9\u9635\uff0cu\u8868\u793a\u504f\u7f6e\u9879\u3002 s i = [s i0 , . . . , s ij , . . . , s iN ](j \u2208 {0, 1, . . . , N })\uff0cs ij \u662fp j \u2192 p i \u4f9d\u5b58\u5173\u7cfb\u7684\u5f97\u5206\uff0c\u5176\u4e2ds i0 \u7528 \u4e8e\u8861\u91cf\u7b2ci\u4e2a\u5355\u8bcd\u6210\u4e3a\u6839ROOT \u7684\u53ef\u80fd\u6027\u3002\u968f\u540e\u91c7\u7528\u516c\u5f0f6\u8fdb\u884c\u5f52\u4e00\u5316\u64cd\u4f5c\u5f97\u5230\u6982\u7387\u5206\u5e03\u03b1 i \uff0c \u7531\u03b1 i (i \u2208 {0, 1, . . . , N })\u6784\u6210\u4f9d\u5b58\u5173\u7cfb\u6982\u7387\u77e9\u9635\u03b1\u3002\u6700\u540e\u91c7\u7528\u6700\u5927\u751f\u6210\u6811\u7b97\u6cd5\u89e3\u7801\u83b7\u5f97\u53e5\u5b50\u7684\u4f9d\u5b58 \u7ed3\u6784\u3002\u5728\u8bad\u7ec3\u9636\u6bb5\uff0c\u6211\u4eec\u4f7f\u7528\u4ea4\u53c9\u71b5\u4f5c\u4e3a\u635f\u5931\u51fd\u6570\uff0c\u5982\u516c\u5f0f7\u6240\u793a\u3002 \u03b1 i = sof tmax(s i )", | |
| "eq_num": "(6)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "L 0 = \u2212 Ns k=1 N k i=1 \u03b2 k i log(\u03b1 k i ) (7) \u76ee\u6807\u51fd\u6570L 0 \u8868\u793a\u4ea4\u53c9\u71b5\u635f\u5931\uff0cN s \u8868\u793a\u4e00\u4e2a\u6279\u6b21\u4e2d\u53e5\u5bf9\u4e2a\u6570\uff0cN k \u8868\u793a\u7b2ck\u4e2aP \u53e5\u7684\u5355\u8bcd\u4e2a \u6570\uff0c\u03b2 k i \u662f\u7b2ck\u4e2aP \u53e5\u4e2d\u7b2ci\u4e2a\u5355\u8bcd\u771f\u5b9e\u6838\u5fc3\u8bcd\u7684\u72ec\u70ed\u7801\u8868\u793a\u3002\u53c2\u7167(Dozat and Manning, 2017) \uff0c \u6211\u4eec\u8bbe\u8ba1\u9884\u6d4b\u4f9d\u5b58\u5173\u7cfb\u7c7b\u578b\u7684\u9884\u6d4b\u6a21\u578b\uff0c\u6b64\u90e8\u5206\u6784\u6210\u7684\u635f\u5931\u51fd\u6570\u4e3aL 1 \u3002\u6211\u4eec\u5c06\u9884\u6d4b\u7ed3\u6784\u7684\u635f \u5931L 0 \u4e0e\u9884\u6d4b\u4f9d\u5b58\u5173\u7cfb\u7c7b\u578b\u7684\u635f\u5931L 1 \u76f8\u52a0\u6784\u6210L p \u3002\u540c\u7406\uff0c\u5bf9\u4e8e\u53e5\u5b50Q\u6211\u4eec\u53ef\u4ee5\u5f97\u5230\u76f8\u5e94\u7684\u76ee\u6807\u51fd \u6570L q \u3002\u6700\u540e\uff0c\u6211\u4eec\u5c06L p \u4e0eL q \u76f8\u52a0\u4f5c\u4e3a\u4f9d\u5b58\u5206\u6790\u6a21\u578b\u7684\u635f\u5931\u51fd\u6570L dep \u3002 3.2 \u8bed \u8bed \u8bed\u4e49 \u4e49 \u4e49\u7ec4 \u7ec4 \u7ec4\u5408 \u5408 \u5408\u8ba1 \u8ba1 \u8ba1\u7b97 \u7b97 \u7b97 \u6211\u4eec\u63d0\u51fa\u4e00\u79cd\u57fa\u4e8e\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\uff0c\u901a\u8fc7\u5229\u7528\u4e0a\u4e00\u8282\u7684\u4f9d\u5b58\u5206\u6790\u63d0\u4f9b\u7684\u4f9d \u5b58\u5173\u7cfb\u7684\u6982\u7387\u77e9\u9635\u03b1\u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u4ee5\u652f\u6301\u6279\u5904\u7406\u5927\u5e45\u63d0\u5347\u8ba1\u7b97\u901f\u5ea6\u3002\u6839\u636e\u53e5\u6cd5\u5206\u6790\uff0c\u03b1 ij \u8868\u793a\u5355\u8bcdp j \u662fp i \u6838\u5fc3\u8bcd\u7684\u6982\u7387\uff0c\u6211\u4eec\u5c06\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u5b66\u4e60\u5230\u7684\u6743\u91cd\u03b1 ij \u89c6\u4e3a\u4f9d\u5b58\u5173\u7cfbp j \u2192 p i \u7684\u8bed \u4e49\u76f8\u5173\u6027\u6743\u91cd\uff0c\u540c\u65f6\u5c06h i \u89c6\u4e3a\u56fe\u4e0a\u8282\u70b9p i \u7684\u8bed\u4e49\u8868\u793a\uff0c\u7136\u540e\u5728\u6b64\u56fe\u57fa\u7840\u4e0a\u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u3002", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u672c\u6587\u91c7\u7528\u56fe\u4fe1\u606f\u4f20\u9012\u673a\u5236 (Veli\u010dkovi\u0107 et al., 2018; Huang et al., 2019 ) \u5efa\u6a21\u56fe\u4e2d\u6bcf\u4e2a\u8282\u70b9\u7684 \u8bed\u4e49\u4fe1\u606f\uff0c\u9996\u5148\u8282\u70b9p i \u4ece\u90bb\u8282\u70b9\u6536\u96c6\u8bed\u4e49\u4fe1\u606f\uff0c\u6211\u4eec\u8bbe\u8ba1\u4e86\u4e24\u79cd\u6536\u96c6\u7684\u65b9\u5f0f\u3002\u7b2c\u4e00\u79cd\u6536\u96c6\u65b9\u5f0f \u5229\u7528\u4f9d\u5b58\u5173\u7cfb\u6982\u7387\u77e9\u9635\u03b1\u76f4\u63a5\u4f5c\u4e3a\u6743\u91cd\u7ed3\u5408\u90bb\u8282\u70b9\u7684\u8bed\u4e49\u8868\u793a\uff0c\u8ba1\u7b97\u516c\u5f0f\u59828\u6240\u793a\uff0c\u6211\u4eec\u79f0\u8fd9 \u79cd\u65b9\u5f0f\u4e3a\u8f6f\u7ed3\u6784\u4fe1\u606f\uff0c\u793a\u610f\u56fe\u5982\u56fe4(a)\u6240\u793a\u3002\u7b2c\u4e8c\u79cd\u6536\u96c6\u65b9\u5f0f\u4f9d\u636e\u4f9d\u5b58\u7ed3\u6784\u7ed3\u5408\u5177\u6709\u4f9d\u5b58\u5173\u7cfb \u8282\u70b9\u7684\u8bed\u4e49\u4fe1\u606f (Huang et al., 2019; Yao et al., 2018) ", | |
| "cite_spans": [ | |
| { | |
| "start": 12, | |
| "end": 37, | |
| "text": "(Veli\u010dkovi\u0107 et al., 2018;", | |
| "ref_id": "BIBREF22" | |
| }, | |
| { | |
| "start": 38, | |
| "end": 56, | |
| "text": "Huang et al., 2019", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 208, | |
| "end": 228, | |
| "text": "(Huang et al., 2019;", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 229, | |
| "end": 246, | |
| "text": "Yao et al., 2018)", | |
| "ref_id": "BIBREF23" | |
| } | |
| ], | |
| "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": "\uff0c\u9996\u5148\u4fee\u6539\u4f9d\u5b58\u5173\u7cfb\u6982\u7387\u77e9\u9635\u03b1\uff0c\u5bf9\u4e8e\u5355 \u8bcdp i \u8bbe\u7f6e\u6982\u7387\u6700\u5927\u7684\u6838\u5fc3\u8bcd\u7684\u6982\u7387\u4e3a1\uff0c\u5176\u4ed6\u5355\u8bcd\u7684\u6982\u7387\u8bbe\u7f6e\u4e3a0\uff0c\u5177\u4f53\u4fee\u6539\u65b9\u5f0f\u5982\u516c\u5f0f9\u6240\u793a\uff0c \u7136\u540e\u518d\u6309\u516c\u5f0f8\u8fdb\u884c\u8bed\u4e49\u4fe1\u606f\u7684\u6536\u96c6\uff0c\u6211\u4eec\u79f0\u8fd9\u79cd\u65b9\u5f0f\u4e3a\u786c\u7ed3\u6784\u4fe1\u606f\uff0c\u793a\u610f\u56fe\u5982\u56fe4(b)\u6240\u793a\u3002\u5f97 \u5230\u90bb\u8282\u70b9\u8bed\u4e49\u4fe1\u606fM i \u540e\uff0c\u6839\u636e\u516c\u5f0f10\u66f4\u65b0\u5f53\u524d\u8282\u70b9\u7684\u8bed\u4e49\u8868\u793a\u3002 M i = N j=1 \u03b1 ij h j (8) \u03b1 ik = 1 ,k = argmax\u03b1 i 0 ,else (9) h i = (1 \u2212 \u03b7 p i )LeakyReLU (M i ) + \u03b7 p i h i (10) \u5176\u4e2d\uff0cM i \u2208 R d \u662f\u8282\u70b9p i \u4ece\u90bb\u8282\u70b9\u83b7\u5f97\u7684\u8bed\u4e49\u4fe1\u606f\uff0ch i \u2208 R d \u8868\u793a\u8282\u70b9p i \u539f\u59cb\u7684\u8bed\u4e49\u8868 \u793a\uff0c\u03b7 p i \u2208 R\u662f\u8282\u70b9p i \u7684\u8bed\u4e49\u66f4\u65b0\u6743\u91cd\uff0c\u63a7\u5236\u5e94\u4fdd\u7559p i \u591a\u5c11\u539f\u6765\u7684\u8bed\u4e49\u4fe1\u606f\uff0c1 \u2212 \u03b7 p i \u7528\u4e8e\u63a7\u5236\u8282 \u70b9p i \u63a5\u6536\u5230\u591a\u5c11\u90bb\u8282\u70b9\u7684\u8bed\u4e49\u4fe1\u606f\u3002\u6700\u540e\uff0c\u4f7f\u7528\u5e73\u5747\u6c60\u5316\u83b7\u5f97\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a\u3002\u53e5\u5b50\u8bed\u4e49\u8868\u793a \u5b9a\u4e49\u4e3a\uff1a v p = 1 |N p | i\u2208Np h i (11) \u5176\u4e2d\uff0cN p \u662f\u53e5\u5b50P \u4e2d\u5355\u8bcd\u8282\u70b9\u4e0b\u6807\u7684\u96c6\u5408\uff0c|N p |\u662f\u53e5\u5b50P \u4e2d\u5355\u8bcd\u7684\u4e2a\u6570\u3002v p \u5373\u4e3a\u53e5\u5b50P \u7684\u8bed\u4e49\u8868 \u793a\uff0c\u540c\u7406\uff0c\u5bf9\u4e8e\u53e5\u5b50Q\uff0c\u6211\u4eec\u53ef\u4ee5\u83b7\u5f97\u5176\u8bed\u4e49\u8868\u793av q \u3002 \u4e3a\u4e86\u68c0\u9a8c\u672c\u6587\u57fa\u4e8e\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\u80fd\u66f4\u597d\u7684\u5b66\u4e60\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a\uff0c\u6211\u4eec \u8054\u5408\u4e86\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u3002\u7ed9\u5b9a\u53e5\u5bf9P \u548cQ\uff0c\u9884\u6d4b\u4e24\u4e2a\u53e5\u5b50\u662f\u5426\u5177\u6709\u76f8\u540c\u7684\u8bed\u4e49\u3002\u9996\u5148\u57fa\u4e8e\u8bed\u4e49 \u7ec4\u5408\u8ba1\u7b97\u6a21\u5757\uff0c\u4e3a\u53e5\u5bf9\u4e2d\u7684\u6bcf\u4e2a\u53e5\u5b50\u751f\u6210\u8bed\u4e49\u8868\u793av p \u548cv q \u3002\u7136\u540e\uff0c\u4f7f\u7528\u8fd9\u4e24\u4e2a\u53e5\u5b50\u7684\u8bed\u4e49\u8868 \u793a(v p \u548cv q )\u6784\u9020\u7279\u5f81\u5411\u91cfd (Mou et al., 2016)\uff0c\u5982\u516c\u5f0f12\u6240\u793a\u3002\u7136\u540e\u5c06\u6b64\u7279\u5f81\u5411\u91cfd\u9001\u5165\u53e5\u5bf9\u5206 \u7c7b\u5668\u3002 d = v p \u2295 v q \u2295 (v p \u2212 v q ) \u2295 (v p v q )", | |
| "eq_num": "(12)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u5176\u4e2d\uff0c \u8868\u793a\u6309\u5143\u7d20\u4e58\u79ef\u64cd\u4f5c\uff0c\u2295\u8868\u793a\u5411\u91cf\u62fc\u63a5\u64cd\u4f5c\uff0cd \u2208 R 4d \u662f\u6784\u9020\u7684\u7279\u5f81\u5411\u91cf\uff0c\u53e5\u5bf9\u5206\u7c7b\u5668 \u6211\u4eec\u91c7\u7528\u591a\u5c42\u611f\u77e5\u673a\u7684\u65b9\u5f0f\uff0c\u5982\u516c\u5f0f13\u6240\u793a\u3002", | |
| "cite_spans": [], | |
| "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": "y = sof tmax(M LP (clf ) (d))", | |
| "eq_num": "(13)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u5728\u8bad\u7ec3\u9636\u6bb5\u6211\u4eec\u4f7f\u7528\u4ea4\u53c9\u71b5\u4f5c\u4e3a\u635f\u5931\u51fd\u6570\uff0c\u5b9a\u4e49\u4e3a\uff1a (Mueller and Thyagarajan, 2016; Tomar et al., 2017; Liu et al., 2018) ", | |
| "cite_spans": [ | |
| { | |
| "start": 24, | |
| "end": 55, | |
| "text": "(Mueller and Thyagarajan, 2016;", | |
| "ref_id": "BIBREF16" | |
| }, | |
| { | |
| "start": 56, | |
| "end": 75, | |
| "text": "Tomar et al., 2017;", | |
| "ref_id": "BIBREF21" | |
| }, | |
| { | |
| "start": 76, | |
| "end": 93, | |
| "text": "Liu et al., 2018)", | |
| "ref_id": "BIBREF13" | |
| } | |
| ], | |
| "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": "L pair = \u2212 Ns i=1 g i log(\u0177 i )", | |
| "eq_num": "(14)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u5176\u4e2d\uff0cN s \u4e3a\u4e00\u4e2a\u6279\u6b21\u4e2d\u53e5\u5bf9\u7684\u4e2a\u6570\uff0cg i \u8868\u793a\u7b2ci\u4e2a\u53e5\u5bf9\u662f\u5426\u4e3a\u590d\u8ff0\uff0c\u5982\u679c\u4e3a\u590d\u8ff0\u5173\u7cfb\uff0cg i = [1, 0]\uff0c\u5982\u679c\u4e3a\u975e\u590d\u8ff0\u5173\u7cfb\uff0c\u5219g i = [0, 1]\uff0c\u0177 i \u662f\u7b2ci\u4e2a\u53e5\u5bf9\u5404\u7c7b\u522b\u7684\u4f30\u8ba1\u6982\u7387\uff0c\u5982\u516c\u5f0f13\u6240\u793a\u3002 3.3 \u8054 \u8054 \u8054\u5408 \u5408 \u5408\u5b66 \u5b66 \u5b66\u4e60 \u4e60 \u4e60 \u672c\u6587\u63d0\u51fa\u7684\u8054\u5408\u6a21\u578b\u6d89\u53ca\u5230\u4e24\u4e2a\u4efb\u52a1\uff0c\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u548c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u6211\u4eec\u91c7\u7528\u590d\u8ff0\u8bc6\u522b \u9a8c\u8bc1\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u3002\u7531\u6b64\uff0c\u6a21\u578b\u9700\u8981\u540c\u65f6\u5b66\u4e60\u548c\u4f18\u5316\u591a\u4e2a\u5b66\u4e60\u76ee\u6807\u3002\u5728\u4f20\u7edf\u7684\u8054\u5408\u5b66\u4e60\u4e2d\uff0c\u901a \u5e38\u5bf9\u5404\u4e2a\u4efb\u52a1\u7684\u635f\u5931\u8fdb\u884c\u7ebf\u6027\u52a0\u6743\u6c42\u548c\uff0c\u5982\u516c\u5f0f15\uff0c\u8be5\u65b9\u6cd5\u6743\u91cd\u8f83\u96be\u8bbe\u5b9a\u3002\u4e3a\u4e86\u89e3\u51b3\u591a\u76ee\u6807\u8054 \u5408\u5b66\u4e60\u95ee\u9898\uff0c\u6211\u4eec\u91c7\u7528Kendall et al. (2018) \u8bbe\u8ba1\u7684\u81ea\u5b66\u4e60\u591a\u76ee\u6807\u6743\u91cd\u65b9\u6cd5\u3002\u8be5\u65b9\u6cd5\u6839\u636e\u566a\u58f0\u65b9 \u5dee\u4f5c\u4e3a\u6a21\u578b\u6536\u655b\u7a0b\u5ea6\u7684\u8bc4\u4f30\uff0c\u8fdb\u884c\u6bd4\u91cd\u8c03\u6574\u3002\u5176\u76ee\u6807\u51fd\u6570\u8bbe\u8ba1\u5982\u516c\u5f0f16\u3002 L = (1 \u2212 w)L pair + wL dep (15) L = 1 2\u03c3 2 1 L pair + log\u03c3 2 1 + 1 2\u03c3 2 2 wL dep + log\u03c3 2 2", | |
| "eq_num": "(16)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u8868 1: \u5b9e\u9a8c\u6570\u636e\u96c6 \u5b9e \u9a8c \u4e2d \u91c7 \u7528 \u9884 \u8bad \u7ec3 \u7684Word2Vec \u8bcd \u5411 \u91cf(Mikolov et al., 2013)\uff0c \u9884 \u8bad \u7ec3 \u8bcd \u5411 \u91cf \u4e3a200\u7ef4 \u3002 \u8bcd \u6027 \u6807 \u7b7e \u5411 \u91cf \u8bbe \u7f6e100\u7ef4 \uff0c \u8bbe \u7f6e \u6240 \u6709LSTM\u7ed3 \u6784 \u7684 \u9690 \u85cf \u5c42 \u4e3a400\u7ef4 \uff0c \u5c42 \u6570 \u4e3a3\u3002 \u5bf9\u4e0eM LP (dep) \u548cM LP (head) \u8bbe\u7f6e\u5c42\u6570\u90fd\u4e3a1\u5c42\u9690\u85cf\u5c42\u7ef4\u5ea6\u5206\u522b\u4e3a100\u548c500\uff0c\u91c7\u7528leakyrelu\u6fc0 \u6d3b \u51fd \u6570 \uff0c\u03b1\u8bbe \u7f6e \u4e3a0.1\u3002 \u5bf9 \u4e8eM LP (clf ) \u8bbe \u7f6e \u5c42 \u6570 \u4e3a2\uff0c \u9690 \u85cf \u5c42 \u7ef4 \u5ea6 \u4e3a800\u548c400\uff0c \u91c7 \u7528 \u76f8 \u540c \u7684 \u6fc0 \u6d3b \u51fd \u6570 \u3002 \u6211 \u4eec \u91c7 \u7528Adam (Kingma and Ba, )\u4f18 \u5316 \u7b97 \u6cd5 \uff0c \u8bbe \u7f6e \u521d \u59cb \u5b66 \u4e60 \u7387 \u5927 \u5c0f \u4e3a2e-3\uff0c \u4e3a\u03b2 1 \u4e3a0.9\uff0c\u03b2 2 \u4e3a0.9\u3002\u5728\u6bcf\u4e00\u8f6e\u8fed\u4ee3\u4e2d\uff0c\u5b66\u4e60\u7387\u4ee50.95\u7684\u9891\u7387\u8870\u51cf\u3002\u8bad\u7ec3batch\u7684\u5927\u5c0f\u4e3a128\u3002 \u4e3a\u4e86\u9632\u6b62\u8fc7\u62df\u5408\uff0c\u6211\u4eec\u4f7f\u7528\u4e86dropout\u3002\u8bbe\u7f6e\u8bcd\u5411\u91cf\u8f93\u5165\u5c42\u7684drop\u7387\u4e3a0.33\uff0cleakyrelu\u5c42\u8f93\u51fa \u5c42\u7684drop\u7387\u4e3a0.33\u3002\u4e0e\u5df2\u6709\u5de5\u4f5c\u4e00\u81f4\uff0c\u6211\u4eec\u91c7\u7528\u65e0\u6807\u8bb0\u4f9d\u5b58\u6b63\u786e\u7387UAS\u548c\u5e26\u6807\u8bb0\u4f9d\u5b58\u6b63\u786e \u7387LAS\u4f5c\u4e3a\u4f9d\u5b58\u5206\u6790\u8bc4\u4ef7\u6307\u6807\uff0c\u91c7\u7528Accuracy\u548c\u878d\u5408Precision\u548cRecall\u7684\u7efc\u5408\u6307\u6807F1 \u503c\u4f5c\u4e3a\u590d \u8ff0\u8bc6\u522b\u7684\u8bc4\u4ef7\u6307\u6807\u3002 4.2 \u57fa \u57fa \u57fa\u4e8e \u4e8e \u4e8e\u81ea \u81ea \u81ea\u5b66 \u5b66 \u5b66\u4e60 \u4e60 \u4e60\u591a \u591a \u591a\u76ee \u76ee \u76ee\u6807 \u6807 \u6807\u6743 \u6743 \u6743\u91cd \u91cd \u91cd\u7684 \u7684 \u7684\u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u7ed3 \u7ed3 \u7ed3\u679c \u679c \u679c \u5982\u679c\u6309\u7167\u516c\u5f0f15\u8ba1\u7b97\u635f\u5931\u51fd\u6570\uff0c\u4e3a\u4e86\u627e\u5230\u5408\u7406\u7684w\u9700\u8981\u591a\u6b21\u5b9e\u9a8c\uff0c\u5b9e\u9a8c\u7ed3\u679c\u5982\u88682\u6240\u793a\u3002 \u88682\u663e\u793a\u4e86\u4e0d\u540c\u6743\u91cdw\u5bf9\u4f9d\u5b58\u5206\u6790\u548c\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u6027\u80fd\u7684\u5f71\u54cd\u7ed3\u679c\u3002\u5f53w\u8f83\u5c0f\u65f6\uff0c\u590d\u8ff0\u8bc6\u522b\u6027\u80fd \u8f83\u597d\uff0c\u4f46\u662f\u4f9d\u5b58\u5206\u6790\u7cbe\u5ea6\u8f83\u4f4e\uff1b\u5f53w\u8f83\u5927\u65f6\uff0c\u4f9d\u5b58\u5206\u6790\u7cbe\u5ea6\u8f83\u597d\u4f46\u662f\u590d\u8ff0\u8bc6\u522b\u6027\u80fd\u8f83\u4f4e\u3002\u5f53w\u8bbe \u7f6e\u4e3a0.9\u65f6\uff0c\u4f9d\u5b58\u5206\u6790\u7684\u7ed3\u679c\u8fbe\u5230\u6700\u597d\uff0c\u5e26\u6807\u8bb0\u6b63\u786e\u7387\u8fbe\u523094.37%\uff0c\u4f46\u590d\u8ff0\u8bc6\u522b\u7684Accuracy\u53ea \u670973.37%\u3002\u5f53w\u8bbe\u7f6e\u4e3a0.5\u65f6\uff0c\u80fd\u5171\u540c\u5f97\u5230\u8f83\u597d\u7684\u6027\u80fd\uff0c\u590d\u8ff0\u8bc6\u522bAccuracy\u4e3a76.31%\uff0c\u4f9d\u5b58\u5206 \u6790LAS\u4e3a93.99%\u3002 \u5982 \u679c \u6309 \u7167 \u516c \u5f0f16\uff0c \u91c7 \u7528Srivastava et al. (2014)\u8bbe \u8ba1 \u7684 \u591a \u76ee \u6807 \u635f \u5931 \u51fd \u6570 \uff0c \u590d \u8ff0 \u8bc6", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\uff1a \uff1a \u4f7f \u7528 \u524d \u5411LSTM\u548c \u540e \u5411LSTM\u6700 \u540e \u65f6 \u523b \u7684 \u9690 \u72b6 \u6001 \u5411 \u91cf \u62fc \u63a5 \u4f5c \u4e3a \u53e5 \u5b50 \u8868 \u793a", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "A comparison of vector-based representations for semantic composition", | |
| "authors": [ | |
| { | |
| "first": "William", | |
| "middle": [], | |
| "last": "Blacoe", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mirella", | |
| "middle": [], | |
| "last": "Lapata", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "546--556", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "William Blacoe and Mirella Lapata. 2012. A comparison of vector-based representations for semantic composition. In Jun'ichi Tsujii, James Henderson, and Marius Pasca, editors, Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, EMNLP-CoNLL 2012, July 12-14, 2012, Jeju Island, Korea, pages 546-556. ACL.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "A fast unified model for parsing and sentence understanding", | |
| "authors": [ | |
| { | |
| "first": "R", | |
| "middle": [], | |
| "last": "Samuel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jon", | |
| "middle": [], | |
| "last": "Bowman", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Abhinav", | |
| "middle": [], | |
| "last": "Gauthier", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Raghav", | |
| "middle": [], | |
| "last": "Rastogi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Christopher", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Gupta", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Christopher", | |
| "middle": [], | |
| "last": "Manning", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Potts", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "1466--1477", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Samuel R. Bowman, Jon Gauthier, Abhinav Rastogi, Raghav Gupta, Christopher D. Manning, and Christopher Potts. 2016. A fast unified model for parsing and sentence understanding. In Proceed- ings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1466-1477, Berlin, Germany, August. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Improved statistical machine translation using paraphrases", | |
| "authors": [ | |
| { | |
| "first": "Chris", | |
| "middle": [], | |
| "last": "Callison-Burch", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Philipp", | |
| "middle": [], | |
| "last": "Koehn", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Miles", | |
| "middle": [], | |
| "last": "Osborne", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2006, | |
| "venue": "Human Language Technology Conference of the North American Chapter of the Association of Computational Linguistics, Proceedings", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Chris Callison-Burch, Philipp Koehn, and Miles Osborne. 2006. Improved statistical machine translation using paraphrases. In Robert C. Moore, Jeff A. Bilmes, Jennifer Chu-Carroll, and Mark Sanderson, editors, Human Language Technology Conference of the North American Chapter of the Associa- tion of Computational Linguistics, Proceedings, June 4-9, 2006, New York, New York, USA. The Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "LTP: A Chinese language technology platform", | |
| "authors": [ | |
| { | |
| "first": "Wanxiang", | |
| "middle": [], | |
| "last": "Che", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Zhenghua", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ting", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Coling 2010: Demonstrations", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "13--16", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Wanxiang Che, Zhenghua Li, and Ting Liu. 2010. LTP: A Chinese language technology platform. In Coling 2010: Demonstrations, pages 13-16, Beijing, China, August. Coling 2010 Organizing Committee.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Enhanced LSTM for natural language inference", | |
| "authors": [ | |
| { | |
| "first": "Qian", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xiaodan", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Zhen-Hua", | |
| "middle": [], | |
| "last": "Ling", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Si", | |
| "middle": [], | |
| "last": "Wei", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Hui", | |
| "middle": [], | |
| "last": "Jiang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Diana", | |
| "middle": [], | |
| "last": "Inkpen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "1657--1668", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Qian Chen, Xiaodan Zhu, Zhen-Hua Ling, Si Wei, Hui Jiang, and Diana Inkpen. 2017. Enhanced LSTM for natural language inference. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1657-1668, Vancouver, Canada, July. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Deep biaffine attention for neural dependency parsing", | |
| "authors": [ | |
| { | |
| "first": "Timothy", | |
| "middle": [], | |
| "last": "Dozat", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Christopher", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Manning", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "5th International Conference on Learning Representations", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Timothy Dozat and Christopher D. Manning. 2017. Deep biaffine attention for neural dependency parsing. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings. OpenReview.net.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "A globalization-semantic matching neural network for paraphrase identification", | |
| "authors": [ | |
| { | |
| "first": "Wutao", | |
| "middle": [], | |
| "last": "Miao Fan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yue", | |
| "middle": [], | |
| "last": "Lin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mingming", | |
| "middle": [], | |
| "last": "Feng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ping", | |
| "middle": [], | |
| "last": "Sun", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the 27th ACM International Conference on Information and Knowledge Management, CIKM '18", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2067--2075", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Miao Fan, Wutao Lin, Yue Feng, Mingming Sun, and Ping Li. 2018. A globalization-semantic matching neural network for paraphrase identification. In Proceedings of the 27th ACM International Con- ference on Information and Knowledge Management, CIKM '18, page 2067-2075, New York, NY, USA. Association for Computing Machinery.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Text level graph neural network for text classification", | |
| "authors": [ | |
| { | |
| "first": "Lianzhe", | |
| "middle": [], | |
| "last": "Huang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Dehong", | |
| "middle": [], | |
| "last": "Ma", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sujian", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xiaodong", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Houfeng", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "3444--3450", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lianzhe Huang, Dehong Ma, Sujian Li, Xiaodong Zhang, and Houfeng Wang. 2019. Text level graph neural network for text classification. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3444-3450, Hong Kong, China, November. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Graph-based dependency parsing with graph neural networks", | |
| "authors": [ | |
| { | |
| "first": "Tao", | |
| "middle": [], | |
| "last": "Ji", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yuanbin", | |
| "middle": [], | |
| "last": "Wu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Man", | |
| "middle": [], | |
| "last": "Lan", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the 57th Conference of the Association for Computational Linguistics, ACL 2019", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "2475--2485", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tao Ji, Yuanbin Wu, and Man Lan. 2019. Graph-based dependency parsing with graph neural networks. In Anna Korhonen, David R. Traum, and Llu\u00eds M\u00e0rquez, editors, Proceedings of the 57th Conference of the Association for Computational Linguistics, ACL 2019, Florence, Italy, July 28-August 2, 2019, Volume 1: Papers, pages 2475-2485. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Multi-task learning using uncertainty to weigh losses for scene geometry and semantics", | |
| "authors": [ | |
| { | |
| "first": "Alex", | |
| "middle": [], | |
| "last": "Kendall", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yarin", | |
| "middle": [], | |
| "last": "Gal", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Roberto", | |
| "middle": [], | |
| "last": "Cipolla", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "2018 IEEE Conference on Computer Vision and Pattern Recognition", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "7482--7491", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Alex Kendall, Yarin Gal, and Roberto Cipolla. 2018. Multi-task learning using uncertainty to weigh losses for scene geometry and semantics. In 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018, pages 7482-7491. IEEE Computer Society.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Convolutional neural networks for sentence classification", | |
| "authors": [ | |
| { | |
| "first": "Yoon", | |
| "middle": [], | |
| "last": "Kim", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1746--1751", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Yoon Kim. 2014. Convolutional neural networks for sentence classification. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1746-1751, Doha, Qatar, October. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Adam: A method for stochastic optimization", | |
| "authors": [ | |
| { | |
| "first": "Diederik", | |
| "middle": [], | |
| "last": "Kingma", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jimmy", | |
| "middle": [], | |
| "last": "Ba", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "When are tree structures necessary for deep learning of representations?", | |
| "authors": [ | |
| { | |
| "first": "Jiwei", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Thang", | |
| "middle": [], | |
| "last": "Luong", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Dan", | |
| "middle": [], | |
| "last": "Jurafsky", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Eduard", | |
| "middle": [], | |
| "last": "Hovy", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2304--2314", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Jiwei Li, Thang Luong, Dan Jurafsky, and Eduard Hovy. 2015. When are tree structures necessary for deep learning of representations? In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2304-2314, Lisbon, Portugal, September. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "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 | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Efficient estimation of word representations in vector space", | |
| "authors": [ | |
| { | |
| "first": "Tomas", | |
| "middle": [], | |
| "last": "Mikolov", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Greg", | |
| "middle": [], | |
| "last": "Corrado", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chen", | |
| "middle": [], | |
| "last": "Kai", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jeffrey", | |
| "middle": [], | |
| "last": "Dean", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "Proceedings of the International Conference on Learning Representations", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tomas Mikolov, Greg Corrado, Chen Kai, and Jeffrey Dean. 2013. Efficient estimation of word represen- tations in vector space. In Proceedings of the International Conference on Learning Representations (ICLR 2013).", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "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", | |
| "volume": "2", | |
| "issue": "", | |
| "pages": "130--136", | |
| "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 (Volume 2: Short Papers), pages 130-136, Berlin, Germany, August. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Siamese recurrent architectures for learning sentence similarity", | |
| "authors": [ | |
| { | |
| "first": "Jonas", | |
| "middle": [], | |
| "last": "Mueller", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Aditya", | |
| "middle": [], | |
| "last": "Thyagarajan", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2786--2792", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Jonas Mueller and Aditya Thyagarajan. 2016. Siamese recurrent architectures for learning sentence similarity. In Dale Schuurmans and Michael P. Wellman, editors, Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence, February 12-17, 2016, Phoenix, Arizona, USA, pages 2786- 2792. AAAI Press.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "Semantic compositionality through recursive matrix-vector spaces", | |
| "authors": [ | |
| { | |
| "first": "Richard", | |
| "middle": [], | |
| "last": "Socher", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Brody", | |
| "middle": [], | |
| "last": "Huval", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Christopher", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Manning", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Andrew", | |
| "middle": [ | |
| "Y" | |
| ], | |
| "last": "Ng", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Richard Socher, Brody Huval, Christopher D. Manning, and Andrew Y. Ng. 2012. Semantic composi- tionality through recursive matrix-vector spaces. In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Dropout: A simple way to prevent neural networks from overfitting", | |
| "authors": [ | |
| { | |
| "first": "Nitish", | |
| "middle": [], | |
| "last": "Srivastava", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Geoffrey", | |
| "middle": [], | |
| "last": "Hinton", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Alex", | |
| "middle": [], | |
| "last": "Krizhevsky", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ilya", | |
| "middle": [], | |
| "last": "Sutskever", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ruslan", | |
| "middle": [], | |
| "last": "Salakhutdinov", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Journal of Machine Learning Research", | |
| "volume": "15", | |
| "issue": "56", | |
| "pages": "1929--1958", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. 2014. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(56):1929-1958.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "Improved semantic representations from tree-structured long short-term memory networks", | |
| "authors": [ | |
| { | |
| "first": "Kai Sheng", | |
| "middle": [], | |
| "last": "Tai", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Richard", | |
| "middle": [], | |
| "last": "Socher", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Christopher", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Manning", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "1556--1566", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kai Sheng Tai, Richard Socher, and Christopher D. Manning. 2015. Improved semantic representations from tree-structured long short-term memory networks. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 1556-1566, Beijing, China, July. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF20": { | |
| "ref_id": "b20", | |
| "title": "Effective LSTMs for target-dependent sentiment classification", | |
| "authors": [ | |
| { | |
| "first": "Duyu", | |
| "middle": [], | |
| "last": "Tang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bing", | |
| "middle": [], | |
| "last": "Qin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xiaocheng", | |
| "middle": [], | |
| "last": "Feng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ting", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "3298--3307", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Duyu Tang, Bing Qin, Xiaocheng Feng, and Ting Liu. 2016. Effective LSTMs for target-dependent sen- timent classification. In Proceedings of COLING 2016, the 26th International Conference on Com- putational Linguistics: Technical Papers, pages 3298-3307, Osaka, Japan, December. The COLING 2016 Organizing Committee.", | |
| "links": null | |
| }, | |
| "BIBREF21": { | |
| "ref_id": "b21", | |
| "title": "Neural paraphrase identification of questions with noisy pretraining", | |
| "authors": [ | |
| { | |
| "first": "Thyago", | |
| "middle": [], | |
| "last": "Gaurav Singh Tomar", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Oscar", | |
| "middle": [], | |
| "last": "Duque", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jakob", | |
| "middle": [], | |
| "last": "T\u00e4ckstr\u00f6m", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Dipanjan", | |
| "middle": [], | |
| "last": "Uszkoreit", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Das", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of the First Workshop on Subword and Character Level Models in NLP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "142--147", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Gaurav Singh Tomar, Thyago Duque, Oscar T\u00e4ckstr\u00f6m, Jakob Uszkoreit, and Dipanjan Das. 2017. Neural paraphrase identification of questions with noisy pretraining. In Manaal Faruqui, Hinrich Sch\u00fctze, Isabel Trancoso, and Yadollah Yaghoobzadeh, editors, Proceedings of the First Workshop on Subword and Character Level Models in NLP, Copenhagen, Denmark, September 7, 2017, pages 142-147. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF22": { | |
| "ref_id": "b22", | |
| "title": "Graph Attention Networks. International Conference on Learning Representations", | |
| "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": "Li\u00f2", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yoshua", | |
| "middle": [], | |
| "last": "Bengio", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Petar Veli\u010dkovi\u0107, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Li\u00f2, and Yoshua Bengio. 2018. Graph Attention Networks. International Conference on Learning Representations. accepted as poster.", | |
| "links": null | |
| }, | |
| "BIBREF23": { | |
| "ref_id": "b23", | |
| "title": "Graph convolutional networks for text classification", | |
| "authors": [ | |
| { | |
| "first": "Liang", | |
| "middle": [], | |
| "last": "Yao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chengsheng", | |
| "middle": [], | |
| "last": "Mao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yuan", | |
| "middle": [], | |
| "last": "Luo", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Liang Yao, Chengsheng Mao, and Yuan Luo. 2018. Graph convolutional networks for text classification. CoRR, abs/1809.05679.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "num": null, | |
| "text": "and Manning, 2017)\u3002\u5728\u6b64\u57fa\u7840\u4e0a\uff0c\u53ef\u4ee5\u4e3a\u6240\u6709\u5355\u8bcd\u5bf9\u4e2d\u7684\u4e24\u79cd\u4f9d\u5b58\u5173\u7cfb \u8ba1\u7b97\u5f97\u5206\uff0c\u5177\u4f53\u7684\u6211\u4eec\u91c7\u7528\u53cc\u4eff\u5c04\u6ce8\u610f\u529b\u673a\u5236\u8fdb\u884c\u8ba1\u7b97\uff0c\u8ba1\u7b97\u8fc7\u7a0b\u5982\u516c\u5f0f5\u6240\u793a\u3002\u5176\u4e2d\uff0cs ij \u8868", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "TABREF2": { | |
| "content": "<table><tr><td colspan=\"2\">4 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c</td><td/><td/><td/><td/></tr><tr><td>4.1</td><td>\u6570 \u6570 \u6570\u636e\u96c6</td><td>\u5212\u5206</td><td colspan=\"4\">\u53e5\u5bf9\u6570 \u6b63\u4f8b\u6570 \u8d1f\u4f8b\u6570 \u8bcd\u6570</td></tr><tr><td/><td/><td colspan=\"5\">\u8bad\u7ec3\u96c6 238,766 138,574 100,192 3279k</td></tr><tr><td/><td>LCQMC</td><td>\u5f00\u53d1\u96c6</td><td>8,802</td><td>4,402</td><td>4,400</td><td>138k</td></tr><tr><td/><td/><td colspan=\"2\">\u6d4b\u8bd5\u96c6 12,500</td><td>6,250</td><td>6,250</td><td>152k</td></tr></table>", | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "\u6570\u636e \u636e \u636e\u96c6 \u96c6 \u96c6\u4ecb \u4ecb \u4ecb\u7ecd \u7ecd \u7ecd\u548c \u548c \u548c\u8d85 \u8d85 \u8d85\u53c2 \u53c2 \u53c2\u6570 \u6570 \u6570\u8bbe \u8bbe \u8bbe\u7f6e \u7f6e \u7f6e \u672c\u6587\u4f7f\u7528\u516c\u5f00\u6c49\u8bed\u590d\u8ff0\u8bc6\u522b\u6570\u636e\u96c6LCQMC (Liu et al., 2018)\u4f5c\u4e3a\u5b9e\u9a8c\u6570\u636e\u3002\u6211\u4eec\u91c7\u7528\u9ad8\u7cbe \u5ea6\u7684\u54c8\u5de5\u5927\u8bed\u8a00\u6280\u672f\u5e73\u53f0ltp3.4.0 -1(Che et al., 2010)\u83b7\u53d6\u5206\u8bcd\u3001\u8bcd\u6027\u548c\u4f9d\u5b58\u53e5\u6cd5\u6807\u6ce8\uff0c\u6211\u4eec\u5c06\u4f9d \u5b58\u53e5\u6cd5\u6807\u6ce8\u89c6\u4e3aground truth\u3002\u88681\u7ed9\u51fa\u4e86LCQMC\u6570\u636e\u96c6\u7684\u7edf\u8ba1\u4fe1\u606f\u3002" | |
| }, | |
| "TABREF3": { | |
| "content": "<table><tr><td/><td/><td colspan=\"2\">\u76ee\u6807\u51fd\u6570\u6743\u91cdw</td><td colspan=\"3\">\u590d\u8ff0\u8bc6\u522b \u4f9d\u5b58\u53e5\u6cd5</td><td/></tr><tr><td/><td colspan=\"3\">\u590d\u8ff0\u8bc6\u522b \u4f9d\u5b58\u53e5\u6cd5</td><td colspan=\"2\">Acc[%]</td><td>LAS[%]</td><td/></tr><tr><td/><td/><td>0.975</td><td>0.025</td><td colspan=\"2\">74.97</td><td>86.37</td><td/></tr><tr><td/><td/><td>0.95</td><td>0.05</td><td colspan=\"2\">75.68</td><td>89.09</td><td/></tr><tr><td/><td/><td>0.9</td><td>0.1</td><td colspan=\"2\">75.61</td><td>90.75</td><td/></tr><tr><td/><td/><td>0.85</td><td>0.15</td><td colspan=\"2\">76.68</td><td>91.61</td><td/></tr><tr><td/><td/><td>0.8</td><td>0.2</td><td colspan=\"2\">76.47</td><td>92.33</td><td/></tr><tr><td/><td/><td>0.7</td><td>0.3</td><td colspan=\"2\">76.75</td><td>93.31</td><td/></tr><tr><td/><td/><td>0.5</td><td>0.5</td><td colspan=\"2\">76.31</td><td>93.99</td><td/></tr><tr><td/><td/><td>0.1</td><td>0.9</td><td colspan=\"2\">73.37</td><td>94.37</td><td/></tr><tr><td/><td colspan=\"3\">Kendall et al. (2018)</td><td colspan=\"2\">76.77</td><td>92.70</td><td/></tr><tr><td colspan=\"8\">\u8868 2: \u4e0d\u540cw\u4e0b\uff0c\u8054\u5408\u6a21\u578b\u5728\u5f00\u53d1\u96c6\u4e2d\u4e24\u4e2a\u4efb\u52a1\u4e0a\u7684\u6027\u80fd</td></tr><tr><td/><td/><td colspan=\"2\">\u5f00\u53d1\u96c6</td><td/><td/><td colspan=\"2\">\u6d4b\u8bd5\u96c6</td></tr><tr><td>\u7ec4\u5408\u6b21\u6570</td><td colspan=\"2\">\u4f9d\u5b58\u5206\u6790</td><td colspan=\"2\">\u590d\u8ff0\u8bc6\u522b</td><td colspan=\"2\">\u4f9d\u5b58\u5206\u6790</td><td colspan=\"2\">\u590d\u8ff0\u8bc6\u522b</td></tr><tr><td/><td>UAS</td><td>LAS</td><td>F 1</td><td>Acc</td><td>UAS</td><td>LAS</td><td>F 1</td><td>Acc</td></tr><tr><td>n=0</td><td colspan=\"8\">93.77 92.67 73.93 74.07 95.36 94.25 79.88 77.57</td></tr><tr><td>n=1</td><td colspan=\"8\">93.92 92.70 77.23 76.77 95.32 94.19 81.84 79.54</td></tr><tr><td>n=2</td><td colspan=\"8\">93.82 92.60 76.94 76.02 95.27 94.16 80.76 78.24</td></tr><tr><td>n=3</td><td colspan=\"8\">93.77 92.57 76.74 75.47 95.23 94.08 79.92 77.00</td></tr><tr><td colspan=\"9\">\u8868 3: \u8bed\u4e49\u7ec4\u5408\u6b21\u6570\u5728\u4e0d\u540c\u4efb\u52a1\u4e0a\u7684\u6027\u80fd\uff0cn=0\u8868\u793a\u6ca1\u6709\u5229\u7528\u53e5\u6cd5\u7ed3\u6784\u4fe1\u606f</td></tr><tr><td>4.4 \u6a21 \u6a21 \u6a21\u578b \u578b \u578b\u5bf9 \u5bf9 \u5bf9\u6bd4 \u6bd4 \u6bd4\u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"9\">\u6211\u4eec\u4e0e\u57fa\u4e8e\u5e8f\u5217\u5316\u548c\u6811\u7ed3\u6784\u76845\u79cd\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\u8fdb\u884c\u6bd4\u8f83\uff0c\u5bf9\u6bd4\u6a21\u578b\u5206\u4e3a\u4ee5\u4e0b\u51e0\u7c7b\uff1a</td></tr><tr><td colspan=\"7\">Baseline: \u4e0a\u4e00\u8282\u4e2dn=0\u7684\u6a21\u578b\uff0c\u5373\u5305\u542b\u5e8f\u5217\u4fe1\u606f\u65e0\u7ed3\u6784\u4fe1\u606f\u3002</td><td/></tr><tr><td colspan=\"9\">MeanVector: \u5c06\u8bcd\u8868\u793a\u7684\u5e73\u5747\u6c60\u5316\u4f5c\u4e3a\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a(Blacoe and Lapata, 2012)\uff0c\u5176</td></tr><tr><td colspan=\"7\">\u4e2d\u8bcd\u8868\u793a\u7684\u8ba1\u7b97\u65b9\u6cd5\u5982\u516c\u5f0f1\uff0c\u8be5\u65b9\u5f0f\u65e0\u5e8f\u5217\u4fe1\u606f\u4e5f\u65e0\u53e5\u6cd5\u7ed3\u6784\u4fe1\u606f\u3002</td><td/></tr><tr><td colspan=\"9\">CNN\uff1a \uff1a \uff1a \u57fa\u4e8e\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5Kim (2014)\u548cLiu et al. (2018)\uff0c\u8be5\u65b9\u5f0f\u5305</td></tr><tr><td colspan=\"9\">\u542b\u5e8f\u5217\u4fe1\u606f\u65e0\u7ed3\u6784\u4fe1\u606f\u3002 \u522bAccuracy\u8fbe\u523076.77%\uff0c\u4f9d\u5b58\u5206\u6790LAS 92.70 %\uff0c\u4e0e\u516c\u5f0f15\u4e2dw = 0.5\u65f6\u7684\u6700\u597d\u7ed3\u679c\u76f8\u6bd4\uff0c\u5176 \u590d\u8ff0\u8bc6\u522b\u7684\u51c6\u786e\u7387\u63d0\u9ad80.46\u4e2a\u70b9\uff0c\u663e\u793a\u8be5\u65b9\u6cd5\u4f18\u4e8e\u7ebf\u6027\u52a0\u6743\u7684\u635f\u5931\u51fd\u6570\u3002\u968f\u540e\u5b9e\u9a8c\u4e2d\u6211\u4eec\u91c7 BiLSTM\uff1a</td></tr><tr><td colspan=\"5\">\u7528Kendall et al. (2018) \u8bbe\u8ba1\u7684\u591a\u76ee\u6807\u51fd\u6570\u65b9\u6cd5\u3002</td><td/><td/><td/></tr><tr><td colspan=\"9\">4.3 3.2\u8282\u4ecb\u7ecd\u4e86\u6bcf\u4e2a\u8282\u70b9\u7ed3\u5408\u90bb\u8fd1\u8282\u70b9\u8bed\u4e49\u4fe1\u606f\u66f4\u65b0\u81ea\u8eab\u8bed\u4e49\u8868\u793a\u7684\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\uff0c\u4f7f\u5f97</td></tr><tr><td colspan=\"9\">\u6bcf\u4e2a\u8282\u70b9\u5305\u542b\u4e86\u76f4\u63a5\u6838\u5fc3\u8bcd\u7684\u8bed\u4e49\u4fe1\u606f\u3002\u5982\u679c\u5728\u6b64\u57fa\u7840\u4e0a\u518d\u8fdb\u884c\u4e00\u6b21\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u5c06\u4f7f</td></tr><tr><td colspan=\"9\">\u6bcf\u4e2a\u8282\u70b9\u83b7\u5f97\u95f4\u63a5\u6838\u5fc3\u8bcd\u7684\u8bed\u4e49\u4fe1\u606f\u3002\u4e3a\u4e86\u5206\u6790\u8bed\u4e49\u7ec4\u5408\u6b21\u6570\u7684\u5f71\u54cd\uff0c\u6211\u4eec\u5206\u522b\u8fdb\u884c\u4e86\u57fa</td></tr><tr><td colspan=\"9\">\u4e8e0\u6b21\u30011\u6b21\u30012\u6b21\u548c3\u6b21\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u7684\u8bc4\u6d4b\uff0c\u5b9e\u9a8c\u7ed3\u679c\u5982\u88683\u6240\u793a\u3002n=0\u8868\u793a\u6ca1\u6709\u5229\u7528\u7ed3\u6784\u4fe1</td></tr><tr><td colspan=\"9\">\u606f\uff0cn=1,2,3\u8868\u793a\u4ee5\u4e0d\u540c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u6b21\u6570\u5229\u7528\u7ed3\u6784\u4fe1\u606f\u3002\u4e0en=0\u76f8\u6bd4\uff0cn=1\u7684\u6a21\u578b\u5728\u6d4b\u8bd5\u96c6</td></tr><tr><td colspan=\"9\">\u4e0a\uff0c\u590d\u8ff0\u8bc6\u522b\u5728F 1 \u548cAccuracy\u5206\u522b\u63d0\u9ad8\u4e861.96\u548c1.97\u4e2a\u70b9, \u8bf4\u660e\u53e5\u6cd5\u7ed3\u6784\u6307\u5bfc\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u4e0a\u7684</td></tr><tr><td>\u6709\u6548\u6027\u3002</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"9\">\u4e0en=2,3\u76f8\u6bd4\uff0cn=1\u7684\u6a21\u578b\u5728\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e0a\u5747\u4f18\u4e8en=2,3\u7684\u6a21\u578b\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\u7ee7\u7eed\u589e\u52a0</td></tr><tr><td colspan=\"9\">\u7ec4\u5408\u6b21\u6570\u5e76\u6ca1\u6709\u63d0\u5347\u6548\u679c\uff0c\u540c\u65f6\uff0c\u968f\u7740\u7ec4\u5408\u6b21\u6570\u7684\u589e\u52a0\uff0c\u6a21\u578b\u7684\u590d\u6742\u5ea6\u4e5f\u4f1a\u589e\u52a0\uff0c\u968f\u540e\u5b9e\u9a8c\u4e2d</td></tr><tr><td colspan=\"9\">\u6211\u4eec\u9009\u62e9\u4e00\u6b21\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u3002\u53e6\u5916\u6211\u4eec\u6ce8\u610f\u5230\u8054\u5408\u6a21\u578b\u5e76\u672a\u7ed9\u4f9d\u5b58\u5206\u6790\u5e26\u6765\u6027\u80fd\u4e0a\u7684\u63d0\u5347\uff0c\u4e00</td></tr><tr><td colspan=\"9\">\u65b9\u9762\u7531\u4e8e\u672c\u6587\u7684\u91cd\u70b9\u653e\u5728\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u4e0a\uff0c\u8fd8\u6ca1\u6709\u627e\u5230\u540c\u65f6\u63d0\u5347\u4f9d\u5b58\u5206\u6790\u7cbe\u5ea6\u7684\u6709\u6548\u8054\u5408\u65b9</td></tr><tr><td colspan=\"9\">\u6cd5\uff1b\u53e6\u4e00\u65b9\u9762\u672c\u6587\u4f7f\u7528\u7684\u4f9d\u5b58\u7ed3\u6784\u6807\u6ce8\u5e76\u975e\u4eba\u5de5\u6807\u6ce8\uff0c\u6211\u4eec\u5206\u6790\u5b58\u5728\u4e00\u5b9a\u9519\u8bef\u96be\u4ee5\u7ed9\u51fa\u4f9d\u5b58\u5206</td></tr><tr><td colspan=\"2\">\u6790\u6a21\u578b\u7684\u6b63\u786e\u8bc4\u6d4b\u7ed3\u679c\u3002</td><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"2\">-1 http://ltp.ai/download.html</td><td/><td/><td/><td/><td/><td/></tr></table>", | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "\u8bed \u8bed \u8bed\u4e49 \u4e49 \u4e49\u7ec4 \u7ec4 \u7ec4\u5408 \u5408 \u5408\u6b21 \u6b21 \u6b21\u6570 \u6570 \u6570\u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u7ed3 \u7ed3 \u7ed3\u679c \u679c \u679c" | |
| }, | |
| "TABREF4": { | |
| "content": "<table><tr><td>\u8ba1\u7b97\u8bed\u8a00\u5b66</td><td/><td/><td/><td/><td/><td/></tr><tr><td>ID</td><td>\u53e5\u5b501(P )</td><td>3516</td><td/><td>\u53e5\u5b502(Q)</td><td/><td colspan=\"2\">Bleu \u771f\u5b9e\u6807\u7b7e Our Baseline</td></tr><tr><td colspan=\"7\">2731 A \u7f51\u7ad9\u6392\u540d\u63a8\u5e7f\uff0c\u4e3b\u8981\u6709\u54ea\u4e9b\u63a8\u5e7f\u65b9\u5f0f\uff0c\u6548\u679c\u597d\u70b9\u7684. B \u6211\u60f3\u77e5\u9053\u5973\u751f\u5404\u79cd\u53d1\u578b\u7684\u540d\u5b57\uff0c\u52a0\u4e0a\u914d\u56fe C \u4eb2\u7231\u7684\u97e9\u8bed\u600e\u4e48\u8bf4 /s D \u5c0f\u858f\u7c73\u548c\u5927\u858f\u7c73\u6709\u4ec0\u4e48\u533a\u522b \u858f\u4ec1\u7c89\u548c\u858f\u7c73\u7c89\u6709\u4ec0\u4e48\u533a\u522b 0.61 \u76ee\u524d\u7f51\u4e0a\u6709\u54ea\u4e9b\u63a8\u5e7f\u65b9\u5f0f 0.19 2XU \u5973\u751f\u5404\u79cd\u53d1\u578b\u540d\u79f0\u56fe\u7247 0.24 7UHH/670 \u4eb2\u7231\u7684\u97e9\u8bed\u600e\u4e48\u5199\uff1f 0.71</td><td>T T F F</td><td>T T F F</td><td>F F T T</td></tr><tr><td>E</td><td colspan=\"2\">\u5982\u4f55\u7f16\u7ec7\u5c0f\u72d7\u72d7\u7684\u8863\u670d\u8981\u65b9\u6cd5\u53ca\u56fe\u89e3</td><td colspan=\"3\">\u5982\u4f55\u7ed9\u5c0f\u72d7\u505a\u8863\u670d\u56fe\u7247</td><td>0.24</td><td>T</td><td>F</td><td>F</td></tr><tr><td>F</td><td>\u5c0f\u5b66\u4e8c\u5e74\u7ea7\u8bed\u6587</td><td/><td/><td colspan=\"2\">1125 \u5c0f\u5b66\u4e8c\u5e74\u7ea7\u8bed\u6587\u9898</td><td>0.80</td><td>F</td><td>T</td><td>T</td></tr><tr><td colspan=\"8\">\u8868 6: \u4e00\u4e9b\u590d\u6742\u7684\u4f8b\u5b50\u5728\u672c\u6587\u6a21\u578b\u548cBaseline\u4e0a\u7684\u8868\u73b0\uff0cT \u8868\u793a\u662f\u590d\u8ff0\u5173\u7cfb\uff0cF \u8868\u793a\u975e\u590d\u8ff0\u5173\u7cfb\u3002</td></tr><tr><td/><td/><td>112</td><td>75</td><td>51</td><td>72</td><td>27</td></tr><tr><td colspan=\"2\">4.6 \u5b9e \u5b9e \u5b9e\u4f8b \u4f8b \u4f8b\u5206 \u5206 \u5206\u6790 \u6790 \u6790</td><td/><td/><td/><td/><td/></tr><tr><td colspan=\"8\">\u6211\u4eec\u5728LCQMC\u7684\u6d4b\u8bd5\u96c6\u4e2d\u6311\u9009\u4e86\u4e00\u4e9b\u53e5\u5bf9\u8fdb\u884c\u8fdb\u4e00\u6b65\u5206\u6790\u3002\u4f7f\u75281-gram\u8ba1\u7b97\u53e5\u5b50P \u4e0e\u53e5</td></tr><tr><td colspan=\"8\">\u56fe 5: \u53e5\u957f\u5bf9\u6a21\u578b\u9884\u6d4b\u901f\u5ea6\u7684\u5f71\u54cd \u5b50Q\u7684Bleu\u503c\uff0c\u5bf9\u4e8e\u590d\u8ff0\u8bc6\u522b\u6765\u8bf4\uff0cBleu\u5f88\u9ad8\u7684\u975e\u590d\u8ff0\u53e5\u5bf9\u548cBleu\u5f88\u4f4e\u7684\u590d\u8ff0\u53e5\u5bf9\uff0c\u90fd\u662f\u5f88\u96be</td></tr><tr><td colspan=\"8\">\u7684\u4efb\u52a1\uff0c\u57fa\u4e8e\u6d45\u5c42\u4fe1\u606f\u7684\u65b9\u6cd5\u5f88\u96be\u6b63\u786e\u8bc6\u522b\uff0c\u9700\u8981\u6df1\u5c42\u8bed\u4e49\u7406\u89e3\u624d\u53ef\u89e3\u51b3\u3002\u6211\u4eec\u7279\u5730\u9009\u62e9\u8fd9\u6837</td></tr><tr><td colspan=\"8\">\u7cfb\u7edf \u7684\u53e5\u5bf9\u8bc4\u6d4b\u6211\u4eec\u6a21\u578b\u7684\u6548\u679c\uff0c\u5206\u6790\u7ed3\u679c\u5982\u88686\u6240\u793a\u3002 Dev F 1 Acc \u793a\u4f8bA-B\u4e3aBleu\u8f83\u4f4e\u7684\u590d\u8ff0\u53e5\u5bf9\uff0c\u56e0\u6b64\uff0c\u5bb9\u6613\u8bc6\u522b\u4e3a\u975e\u590d\u8ff0\u5173\u7cfb\u3002\u4f46\u662f\u672c\u6587\u6a21\u578b\u80fd\u591f\u6b63\u786e Test F 1 Acc \u8bc6\u522b\u4e3a\u590d\u8ff0\u5173\u7cfb\uff0c\u800cBaseline\u9519\u8bef\u7684\u8bc6\u522b\u4e3a\u975e\u590d\u8ff0\u5173\u7cfb\u3002\u8fd9\u4e00\u5bf9\u6bd4\u7ed3\u679c\u8868\u660e\u672c\u6587\u5229\u7528\u53e5\u6cd5\u7ed3\u6784 (1)\u65e0\u7ed3\u6784\u4fe1\u606f 73.93 74.07 79.88 77.57 \u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u7684\u65b9\u6cd5\u53ef\u4ee5\u6355\u6349\u53e5\u5bf9\u4e4b\u95f4\u6df1\u5c42\u7684\u8bed\u4e49\u76f8\u5173\u6027\uff0c\u5b9e\u73b0\u6b63\u786e\u5224\u65ad\u3002 (2)\u786c\u7ed3\u6784\u4fe1\u606f 77.01 76.51 81.52 79.15 \u793a\u4f8bC-D\u4e3aBleu\u8f83\u9ad8\u7684\u975e\u590d\u8ff0\u53e5\u5bf9\uff0c\u56e0\u6b64\uff0c\u5bb9\u6613\u8bc6\u522b\u4e3a\u590d\u8ff0\u5173\u7cfb\u3002\u4f46\u662f\u672c\u6587\u6a21\u578b\u80fd\u591f\u6b63\u786e (3)\u8f6f\u7ed3\u6784\u4fe1\u606f 77.23 76.78 81.84 79.54 \u8bc6\u522b\u4e3a\u975e\u590d\u8ff0\u5173\u7cfb\uff0c\u800cBaseline\u9519\u8bef\u7684\u8bc6\u522b\u4e3a\u590d\u8ff0\u5173\u7cfb\u3002\u8fd9\u4e00\u5bf9\u6bd4\u7ed3\u679c\u8868\u660e\u53e5\u6cd5\u7ed3\u6784\u66f4\u6613\u4e8e\u89e3</td></tr><tr><td colspan=\"8\">\u51b3\u6d89\u53ca\u7ed3\u6784\u590d\u6742\u8868\u8fbe\u7684\u8bed\u4e49\u7406\u89e3\u3002 \u8868 5: \u6a21\u578b\u878d\u5165\u4f9d\u5b58\u7ed3\u6784\u4fe1\u606f\u6709\u6548\u6027\u5206\u6790\u7ed3\u679c \u793a\u4f8bE-F\u662fBaseline\u548c\u672c\u6587\u6a21\u578b\u90fd\u4ea7\u751f\u9519\u8bef\u7684\u60c5\u51b5\u3002E\u4e3aBleu\u8f83\u4f4e\u7684\u590d\u8ff0\u53e5\u5bf9\u3002\u6211\u4eec\u5206\u6790\u9884</td></tr><tr><td colspan=\"8\">\u6d4b\u9519\u8bef\u7684\u539f\u56e0\u662f\u53e5\u5b50\u7684\u8868\u8fbe\u53e3\u8bed\u5316\uff0c\u53e5\u6cd5\u5206\u6790\u5f88\u96be\u8fdb\u884c\u6b63\u786e\u5206\u6790\u3002F\u4e3aBleu\u8f83\u9ad8\u7684\u975e\u590d\u8ff0\u53e5</td></tr><tr><td colspan=\"8\">\u5bf9\uff0c\u5176\u4e2d\u542b\u6709\u76f8\u4f3c\u7684\u8bcd\u8bed\"\u8bed\u6587\"\u548c\"\u8bed\u6587\u9898\"\uff0c\u6211\u4eec\u5206\u6790\u9884\u6d4b\u9519\u8bef\u7684\u539f\u56e0\u662f\u8bcd\u7684\u8bed\u4e49\u8868\u793a\u4e0d\u80fd\u6709 \u679c\uff0c\u5728F 1 \u548cAccuracy\u8fbe\u523079.88%\u548c77.57%\u3002\u6211\u4eec\u5206\u6790\u539f\u56e0\u662f\u5728\u6211\u4eec\u7684\u6a21\u578b\u4e2d\u4f7f\u7528\u4e86\u56fe\u795e\u7ecf\u7f51 \u6548\u7684\u533a\u5206\u4e8c\u8005\uff0c\u8fd9\u4f7f\u6a21\u578b\u9519\u8bef\u7684\u8ba4\u4e3a\u5b83\u4eec\u662f\u590d\u8ff0\u7684\u5173\u7cfb\u3002\u5bf9\u4e8e\u66f4\u590d\u6742\u7684\u60c5\u51b5\uff0c\u53e5\u5b50\u7684\u8bed\u4e49\u8868\u793a</td></tr><tr><td colspan=\"8\">\u7edc\uff0c\u5f53\u5904\u7406\u8f83\u957f\u53e5\u5b50\u65f6\uff0c\u6bcf\u4e2a\u8282\u70b9\u80fd\u4ece\u8f83\u8fdc\u8282\u70b9\u6536\u96c6\u8bed\u4e49\u4fe1\u606f\u66f4\u65b0\u81ea\u8eab\u8868\u793a\uff0c\u80fd\u6355\u6349\u8f83\u957f\u53e5\u5b50 \u4efb\u7136\u9762\u4e34\u5f88\u591a\u7684\u95ee\u9898\uff0c\u4f8b\u5982\u6b67\u4e49\u6027\u4ee5\u53ca\u53e3\u8bed\u8868\u8fbe\u3002\u6a21\u578b\u53ef\u80fd\u9700\u8981\u66f4\u591a\u7684\u63a8\u7406\u4fe1\u606f\u6765\u533a\u5206\u8fd9\u4e9b\u5173</td></tr><tr><td colspan=\"8\">\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\u3002\u4e0e\u672a\u5229\u7528\u7ed3\u6784\u4fe1\u606f\u7684\u6a21\u578b\u4e2d\u6700\u597d\u7cbe\u5ea6\u7684Baseline\u76f8\u6bd4\uff0c\u6211\u4eec\u7684\u6a21\u578b\u5728Baseline\u7684 \u7cfb\u5e76\u505a\u51fa\u6b63\u786e\u7684\u51b3\u5b9a\uff0c\u4f8b\u5982\u7ed3\u5408\u5916\u90e8\u77e5\u8bc6\u7528\u4e8e\u5e2e\u52a9\u6a21\u578b\u66f4\u597d\u5730\u7406\u89e3\u8bcd\u6c47\u548c\u77ed\u8bed\u8bed\u4e49\u3002 \u57fa\u7840\u4e0a\u52a0\u5165\u4f9d\u5b58\u7ed3\u6784\u4f18\u5316\u76ee\u6807\uff0c\u5728F 1 \u548cAccuracy\u4e0a\u5206\u522b\u63d0\u9ad8\u4e861.96\u548c1.97\u4e2a\u70b9\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e \u5229\u7528\u53e5\u6cd5\u7ed3\u6784\u4fe1\u606f\u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u7684\u6709\u6548\u6027\u3002\u4e0e\u5229\u7528\u7ed3\u6784\u4fe1\u606f\u7684Tree-LSTM\u76f8\u6bd4\uff0c\u6211\u4eec\u6a21\u578b 5 \u603b \u603b \u603b\u7ed3 \u7ed3 \u7ed3\u4e0e \u4e0e \u4e0e\u5c55 \u5c55 \u5c55\u671b \u671b \u671b</td></tr><tr><td colspan=\"8\">\uff0c\u8be5\u65b9\u5f0f\u5305\u542b\u5e8f\u5217 \u5728F 1 \u548cAccuracy\u7565\u4f4e\u4e8eTree-LSTM 0.18\u548c0.68\u4e2a\u70b9\u3002\u6211\u4eec\u5206\u6790\u539f\u56e0\u662fTree-LSTM\u76f4\u63a5\u4f7f\u7528\u4e86\u6211 \u4fe1\u606f\u65e0\u7ed3\u6784\u4fe1\u606f\u3002 \u672c\u6587\u63d0\u51fa\u4e00\u79cd\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u548c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u7684\u8054\u5408\u6846\u67b6\uff0c\u8bbe\u8ba1\u4e86\u57fa\u4e8e\u56fe\u7684\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u6a21 \u4eec\u89c6\u4e3aground truth\u7684\u4f9d\u5b58\u6807\u7b7e\uff0c\u800c\u6211\u4eec\u7684\u6a21\u578b\u4f7f\u7528\u7684\u662f\u4f9d\u5b58\u6807\u7b7e\u8bad\u7ec3\u4e4b\u540e\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u6a21\u5757 \u578b\u548c\u57fa\u4e8e\u56fe\u795e\u7ecf\u7f51\u7edc\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u6a21\u578b\uff0c\u5229\u7528\u4f9d\u5b58\u5206\u6790\u7ed9\u51fa\u7684\u5e26\u6709\u6982\u7387\u7684\u4f9d\u5b58\u5173\u7cfb\u7ed3\u6784\u56fe\uff0c\u5b9e \u4ea7\u751f\u7684\u4f9d\u5b58\u7ed3\u6784\uff0c\u5176\u4f9d\u5b58\u5206\u6790\u7cbe\u5ea6\u6ca1\u6709ground truth\u9ad8\u3002 \u73b0\u8f6f\u7ed3\u6784\u7684\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u65b9\u6cd5\u3002\u4e00\u65b9\u9762\u56fe\u6a21\u578b\u4e2d\u7684\u5e76\u884c\u8ba1\u7b97\u80fd\u591f\u652f\u6301\u8bad\u7ec3\u548c\u9884\u6d4b\u9636\u6bb5\u7684\u6279\u5904 \u5df2\u6709\u7684\u57fa\u4e8e\u7ed3\u6784\u7684Tree-LSTM\u6bcf\u6b21\u53ea\u80fd\u5904\u7406\u4e00\u4e2a\u53e5\u5bf9\uff0c\u672c\u6587\u91c7\u7528\u57fa\u4e8e\u56fe\u7684\u4f9d\u5b58\u5206\u6790\u548c\u56fe\u7f51 \u7406\uff0c\u6781\u5927\u63d0\u9ad8\u8ba1\u7b97\u901f\u5ea6\uff1b\u53e6\u4e00\u65b9\u9762\u4e24\u4e2a\u4efb\u52a1\u7684\u8054\u5408\u5b66\u4e60\u53ef\u4f7f\u8bed\u4e49\u8868\u793a\u540c\u65f6\u5b66\u4e60\u53e5\u6cd5\u7ed3\u6784\u548c\u8bed\u4e49 \u7edc\u8bed\u4e49\u7ec4\u5408\u65b9\u6cd5\uff0c\u53ef\u4ee5\u5b9e\u73b0\u5bf9\u591a\u4e2a\u53e5\u5bf9\u7684\u6279\u5904\u7406\uff0c\u4ece\u800c\u89e3\u51b3\u5df2\u6709\u6a21\u578b\u9884\u6d4b\u901f\u5ea6\u6162\u7684\u95ee\u9898\u3002\u6211 \u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\uff0c\u6295\u9ad8\u590d\u8ff0\u8bc6\u522b\u7cbe\u5ea6\u3002 \u4eec\u5728\u4e0d\u540c\u957f\u5ea6\u7684\u53e5\u5b50\u4e0a\u5bf9\u6a21\u578b\u7684\u9884\u6d4b\u901f\u5ea6\u8fdb\u884c\u8bc4\u6d4b\uff0c\u5e76\u4e0eTree-LSTM\u8fdb\u884c\u5bf9\u6bd4\uff0c\u8bc4\u6d4b\u7ed3\u679c\u5982 \u4eca\u540e\uff0c\u6211\u4eec\u8003\u8651\u7ed3\u5408\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u5982ELMO\uff0cBERT\uff0c\u4ee5\u6539\u8fdb\u6a21\u578b\u6027\u80fd\u3002\u540c\u65f6\uff0c\u63a2\u7d22\u8054\u5408 \u56fe5\u6240\u793a\u3002\u53e5\u957f\u57281-5\u548c6-10\u4e2d\u6211\u4eec\u6a21\u578b\u9884\u6d4b\u901f\u5ea6\u662fTree-LSTM \u768430\u500d\uff1b\u5728\u53e5\u957f\u4e3a11-15\u4e2d\uff0c\u901f\u5ea6 \u662fTree-LSTM\u768420\u500d\u3002\u8fd9\u4e9b\u7ed3\u679c\u663e\u793a\u672c\u6587\u63d0\u51fa\u7684\u6a21\u578b\u5728\u9884\u6d4b\u901f\u5ea6\u4e0a\u8f83Tree-LSTM\u6709\u663e\u8457\u4f18\u52bf\u3002 \u6a21\u578b\u4e2d\u63d0\u5347\u4f9d\u5b58\u5206\u6790\u7cbe\u5ea6\u7684\u65b9\u6cd5\uff0c\u4ece\u800c\u8fdb\u4e00\u6b65\u63d0\u5347\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\u7684\u7cbe\u5ea6\u3002</td></tr><tr><td colspan=\"8\">TreeLSTM: \u4f7f\u7528Tai et al. (2015)\u63d0\u51fa\u7684Child-Sum Tree-LSTM\uff0c\u5229\u7528\u4f9d\u5b58\u7ed3\u6784\u6811\u8fdb\u884c\u8bed \u4ee5\u4e0a\u5206\u6790\u7ed3\u679c\u663e\u793a\uff0c\u672c\u6587\u63d0\u51fa\u7684\u57fa\u4e8e\u4f9d\u5b58\u53e5\u6cd5\u5206\u6790\u548c\u590d\u8ff0\u8bc6\u522b\u7684\u8054\u5408\u6a21\u578b\uff0c\u91c7\u7528\u57fa\u4e8e\u56fe\u795e \u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u5c06\u6839\u8282\u70b9\u83b7\u5f97\u7684\u9690\u72b6\u6001\u5411\u91cf\u89c6\u4e3a\u53e5\u5b50\u7684\u8868\u793a\uff0c\u8be5\u65b9\u5f0f\u5305\u542b\u7ed3\u6784\u4fe1\u606f\u3002 \u7ecf\u7f51\u7edc\u7684\u8bed\u4e49\u7ec4\u5408\u65b9\u6cd5\uff0c\u53ef\u4ee5\u6709\u6548\u5229\u7528\u53e5\u6cd5\u7ed3\u6784\u4fe1\u606f\u6539\u8fdb\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u63d0\u9ad8\u590d\u8ff0\u8bc6\u522b\u7cfb\u7edf\u7684 \u5728\u590d\u8ff0\u8bc6\u522b\u4efb\u52a1\u4e0a\uff0c\u6211\u4eec\u7684\u6a21\u578b\u4e0e5\u79cd\u6a21\u578b\u5728\u6d4b\u8bd5\u96c6\u4e0a\u7684\u8bc4\u6d4b\u7ed3\u679c\u5982\u88684\u6240\u793a\u3002 \u4ece\u88684\u7684\u7ed3\u679c\u53ef\u4ee5\u770b\u51fa\uff0c\u5728\u65e0\u7ed3\u6784\u4fe1\u606f\u76844\u79cd\u65b9\u6cd5\u4e2d\uff0c\u6211\u4eec\u8bbe\u8ba1\u7684Baseline\u53d6\u5f97\u4e86\u6700\u597d\u7684\u7ed3 \u7cbe\u5ea6\u548c\u8ba1\u7b97\u901f\u5ea6\u3002</td></tr><tr><td colspan=\"8\">\u662f\u5426\u5229\u7528\u7ed3\u6784\u4fe1\u606f \u65b9\u6cd5 \u5426 Baseline 4.5 \u6211\u4eec\u5206\u6790\u4e86\u6a21\u578b\u4e2d\u7ed3\u6784\u4fe1\u606f\u5bf9\u6700\u7ec8\u590d\u8ff0\u8bc6\u522b\u7cbe\u5ea6\u7684\u5f71\u54cd\uff0c\u5b9e\u9a8c\u7ed3\u679c\u5c55\u793a\u5728\u88685\u4e2d\u3002\u4ece\u88685\u7684 F 1 Acc 79.88 77.57 \u5426 \u5b9e\u9a8c\u7ed3\u679c\u6765\u770b\uff0c\u57fa\u4e8e\u56fe\u795e\u7ecf\u7f51\u7edc\u5f15\u5165\u4f9d\u5b58\u7ed3\u6784\u4fe1\u606f\uff0c\u6709\u6548\u6539\u8fdb\u4e86\u590d\u8ff0\u8bc6\u522b\u7684\u6027\u80fd\u3002\u6a21\u578b(1) Mean vectors 78.68 75.08 \u5426 CNN \u6ca1\u6709\u4f7f\u7528\u7ed3\u6784\u4fe1\u606f\uff0c\u4ec5\u4f7f\u7528\u4e86\u590d\u8ff0\u8bc6\u522b\u7684\u76ee\u6807\u51fd\u6570\u8fdb\u884c\u4f18\u5316\uff0c\u672a\u8003\u8651\u53e5\u5b50\u7684\u53e5\u6cd5\u7ed3\u6784\uff0c\u590d\u8ff0\u8bc6 75.70 72.80 \u5426 BiLSTM \u522b\u7684Accuracy\u8fbe\u523077.57%\u3002\u6a21\u578b(2)\u5f15\u5165\u4e86\u53e5\u6cd5\u76ee\u6807\u8bad\u7ec3\u6a21\u578b\u53c2\u6570\uff0c\u91c7\u7528\u4e86\u672c\u6587\u63d0\u51fa\u7684\u786c\u7ed3 78.92 76.10 \u662f Tree-LSTM \u6784\u4fe1\u606f\uff0c\u590d\u8ff0\u8bc6\u522bAccuracy\u8fbe\u523079.15%\uff0c\u5bf9\u6bd4\u6ca1\u6709\u7ed3\u6784\u4fe1\u606f\u63d0\u9ad8\u4e861.58\u4e2a\u767e\u5206\u70b9\uff0c\u8fd9\u8868\u660e\u5f15\u5165 82.02 80.22 \u662f Our \u53e5\u6cd5\u7ed3\u6784\u5bf9\u8bed\u4e49\u7ec4\u5408\u7684\u6709\u6548\u6027\u3002\u6a21\u578b(3)\u91c7\u7528\u4e86\u8f6f\u7ed3\u6784\u4fe1\u606f\uff0cAccuracy\u8fbe\u523079.54%\uff0c\u8fdb\u4e00\u6b65 81.84 79.54 \u6539\u8fdb\u4e86\u590d\u8ff0\u8bc6\u522b\u7684\u6027\u80fd\uff0c\u540c\u65f6\uff0c\u5b9e\u9a8c\u8868\u660e\u672c\u6587\u63d0\u51fa\u7684\u8f6f\u7ed3\u6784\u4f9d\u5b58\u4fe1\u606f\u5728\u6027\u80fd\u4e0a\u4f18\u4e8e\u786c\u7ed3\u6784\u7684\u65b9</td></tr><tr><td colspan=\"8\">\u6cd5\u3002\u6700\u7ec8\uff0c\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u672c\u6587\u63d0\u51fa\u7684\u57fa\u4e8e\u53e5\u6cd5\u7ed3\u6784\u8fdb\u884c\u8bed\u4e49\u7ec4\u5408\u8ba1\u7b97\uff0c\u53ef\u4ee5\u6709\u6548\u5b66\u4e60\u53e5\u5b50\u7684 \u8868 4: \u5728\u590d\u8ff0\u8bc6\u522b\u4e0a\u548c\u5df2\u6709\u5e8f\u5217\u5316\u548c\u6811\u7ed3\u6784\u8bed\u4e49\u7ec4\u5408\u65b9\u5f0f\u7684\u6bd4\u8f83\u7ed3\u679c \u8bed\u4e49\u8868\u793a\uff0c\u63d0\u9ad8\u4e86\u590d\u8ff0\u8bc6\u522b\u7cfb\u7edf\u7684\u7cbe\u5ea6\u3002</td></tr></table>", | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "\u7ed3 \u7ed3 \u7ed3\u6784 \u6784 \u6784\u4fe1 \u4fe1 \u4fe1\u606f \u606f \u606f\u6709 \u6709 \u6709\u6548 \u6548 \u6548\u6027 \u6027 \u6027\u5206 \u5206 \u5206\u6790 \u6790 \u6790" | |
| } | |
| } | |
| } | |
| } |