| { | |
| "paper_id": "2020", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T12:52:25.748920Z" | |
| }, | |
| "title": "\u878d \u878d \u878d\u5408 \u5408 \u5408\u5168 \u5168 \u5168\u5c40 \u5c40 \u5c40\u548c \u548c \u548c\u5c40 \u5c40 \u5c40\u90e8 \u90e8 \u90e8\u4fe1 \u4fe1 \u4fe1\u606f \u606f \u606f\u7684 \u7684 \u7684\u6c49 \u6c49 \u6c49\u8bed \u8bed \u8bed\u5b8f \u5b8f \u5b8f\u89c2 \u89c2 \u89c2\u7bc7 \u7bc7 \u7bc7\u7ae0 \u7ae0 \u7ae0\u7ed3 \u7ed3 \u7ed3\u6784 \u6784 \u6784\u8bc6 \u8bc6 \u8bc6\u522b \u522b \u522b", | |
| "authors": [ | |
| { | |
| "first": "\u8303", | |
| "middle": [ | |
| "\u8303" | |
| ], | |
| "last": "\u8303\u4e9a \u4e9a \u4e9a\u946b", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uff0c", | |
| "middle": [ | |
| "\uff0c" | |
| ], | |
| "last": "\uff0c\u848b \u848b \u848b\u5cf0", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uff0c", | |
| "middle": [ | |
| "\uff0c" | |
| ], | |
| "last": "\uff0c\u891a", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\u891a", | |
| "middle": [], | |
| "last": "\u891a\u6653 \u6653 \u6653\u654f", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uff0c", | |
| "middle": [ | |
| "\uff0c" | |
| ], | |
| "last": "\uff0c\u674e", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\u674e", | |
| "middle": [], | |
| "last": "\u674e\u57f9 \u57f9 \u57f9\u5cf0", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uff0c", | |
| "middle": [ | |
| "\uff0c" | |
| ], | |
| "last": "\uff0c\u6731", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\u6731", | |
| "middle": [], | |
| "last": "\u6731\u5de7 \u5de7 \u5de7\u660e", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "\u82cf\u5dde\u5927\u5b66\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u9662\uff0c\u82cf\u5dde\uff0c\u4e2d\u56fd", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Yaxin", | |
| "middle": [], | |
| "last": "Fan", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uff0c", | |
| "middle": [ | |
| "\uff0c" | |
| ], | |
| "last": "\uff0cfeng Jiang", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uff0c", | |
| "middle": [ | |
| "\uff0c" | |
| ], | |
| "last": "\uff0cxiaomin Chu", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uff0c", | |
| "middle": [ | |
| "\uff0c" | |
| ], | |
| "last": "\uff0cpeifeng Li", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uff0c", | |
| "middle": [ | |
| "\uff0c" | |
| ], | |
| "last": "\uff0cqiaoming Zhu", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou", | |
| "country": "China" | |
| } | |
| }, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "As the fundamental task in macro discourse analysis, the discourse structure recognition task aims to identify the structure between adjacent discourse units and build a discourse structure tree hierarchically. Existing work only considers local structural and semantic information or only global information. Therefore, this paper proposes a pointer network model that integrates global and local information. It can effectively improve the ability of macro text structure recognition by considering the global semantic information and the closeness of the semantic relationship between paragraphs. The experimental results in the Chinese macro discourse treebank show that the proposed model outperforms the state-of-the-art model.", | |
| "pdf_parse": { | |
| "paper_id": "2020", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "As the fundamental task in macro discourse analysis, the discourse structure recognition task aims to identify the structure between adjacent discourse units and build a discourse structure tree hierarchically. Existing work only considers local structural and semantic information or only global information. Therefore, this paper proposes a pointer network model that integrates global and local information. It can effectively improve the ability of macro text structure recognition by considering the global semantic information and the closeness of the semantic relationship between paragraphs. The experimental results in the Chinese macro discourse treebank show that the proposed model outperforms the state-of-the-art model.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "1 \u5f15 \u5f15 \u5f15\u8a00 \u8a00 \u8a00 \u00a92020 \u4e2d\u56fd\u8ba1\u7b97\u8bed\u8a00\u5b66\u5927\u4f1a \u6839\u636e\u300aCreative Commons Attribution 4.0 International License\u300b\u8bb8\u53ef\u51fa\u7248 \u57fa \u57fa \u57fa\u91d1 \u91d1 \u91d1\u9879 \u9879 \u9879\u76ee \u76ee \u76ee\uff1a \uff1a \uff1a\u56fd\u5bb6\u81ea\u7136\u79d1\u5b66\u57fa\u91d1 (61772354, 61836007, 61773276) ;\u6c5f\u82cf\u9ad8\u6821\u4f18\u52bf\u5b66\u79d1\u5efa\u8bbe\u5de5\u7a0b\u8d44\u52a9\u9879\u76ee \u5f53\u524d\uff0c\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7684\u7814\u7a76\u5185\u5bb9\u5df2\u7ecf\u4ece\u8bcd\u6c47\u7406\u89e3\u3001\u53e5\u6cd5\u5206\u6790\u7b49\u6d45\u5c42\u8bed\u4e49\u5206\u6790\u9886\u57df\u5ef6\u4f38\u5230\u6df1 \u5c42\u8bed\u4e49\u7406\u89e3\u7684\u7bc7\u7ae0\u5206\u6790\u9886\u57df\u3002\u7bc7\u7ae0\u5206\u6790\u662f\u81ea\u7136\u8bed\u8a00\u5904\u7406\u9886\u57df\u7684\u91cd\u70b9\u548c\u96be\u70b9\uff0c\u5176\u4e3b\u8981\u4efb\u52a1\u662f\u4ece\u6574 \u4f53\u4e0a\u5206\u6790\u4e00\u7bc7\u6587\u7ae0\u7684\u903b\u8f91\u7ed3\u6784\u548c\u7bc7\u7ae0\u5355\u5143\u4e4b\u95f4\u7684\u8bed\u4e49\u5173\u7cfb\uff0c\u8fdb\u800c\u4ece\u66f4\u6df1\u7684\u5c42\u6b21\u6316\u6398\u81ea\u7136\u8bed\u8a00 \u6587\u672c\u7684\u8bed\u4e49\u548c\u7ed3\u6784\u4fe1\u606f\u3002\u7bc7\u7ae0\u5206\u6790\u6709\u52a9\u4e8e\u7406\u89e3\u7bc7\u7ae0\u7684\u4e2d\u5fc3\u601d\u60f3\u548c\u4e3b\u8981\u5185\u5bb9\uff0c\u53ef\u4ee5\u63d0\u5347\u81ea\u7136\u8bed \u8a00\u5904\u7406\u76f8\u5173\u5e94\u7528\u7684\u6027\u80fd\uff0c\u4f8b\u5982\u95ee\u7b54\u7cfb\u7edf (Liakata et al., 2013) \u548c\u81ea\u52a8\u6587\u6458 (Cohan and Goharian, 2015) ", | |
| "cite_spans": [ | |
| { | |
| "start": 126, | |
| "end": 136, | |
| "text": "(61772354,", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 137, | |
| "end": 146, | |
| "text": "61836007,", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 147, | |
| "end": 156, | |
| "text": "61773276)", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 357, | |
| "end": 379, | |
| "text": "(Liakata et al., 2013)", | |
| "ref_id": "BIBREF12" | |
| }, | |
| { | |
| "start": 386, | |
| "end": 412, | |
| "text": "(Cohan and Goharian, 2015)", | |
| "ref_id": "BIBREF4" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "(\u5b8c) \u4f8b1.\u674e\u9e4f\u5f3a\u8c03\u8981\u52a0\u5feb\u63a8\u884c\u516c\u52a1\u5458\u5236\u5ea6 p 1 -p 5 \u6784\u6210\u7684\u7bc7\u7ae0\u7ed3\u6784\u6811\u5982\u56fe1\u6240\u793a\u3002\u56fe\u4e2d\uff0c\u53f6\u5b50\u8282\u70b9(p 1 -p 5 )\u4e3a\u6bb5\u843d\uff0c\u5373\u5b8f\u89c2\u7bc7\u7ae0\u7ed3\u6784\u4e2d \u7684\u57fa\u672c\u7bc7\u7ae0\u5355\u5143(EDUs)\uff1b\u76f8\u90bb\u53f6\u5b50\u8282\u70b9\u901a\u8fc7\u7bc7\u7ae0\u5173\u7cfb\u8054\u7cfb\u8d77\u6765\uff0c\u901a\u8fc7\u8fde\u63a5\u540e\u6784\u6210\u7684\u8282\u70b9\u662f \u7bc7\u7ae0\u5355\u5143(DUs)\uff0c\u8868\u793a\u4e24\u4e2a\u57fa\u672c\u7bc7\u7ae0\u5355\u5143\u4e4b\u95f4\u7684\u5173\u7cfb\uff1b\u7bad\u5934\u6307\u5411\u7684\u662f\u6838\u5fc3\uff0c\u5373\u91cd\u8981\u7684\u7bc7\u7ae0\u5355 \u5143\u3002\u5177\u4f53\u800c\u8a00\uff0c\u7bc7\u7ae0\u5355\u5143\u4e4b\u95f4\u901a\u8fc7\u7bc7\u7ae0\u5173\u7cfb\u76f8\u8fde\u63a5\uff0c\u6700\u7ec8\u5f62\u6210\u4e00\u68f5\u5b8c\u6574\u7684\u7bc7\u7ae0\u7ed3\u6784\u6811\u3002\u672c\u6587\u7814 \u7a76\u7684\u4e3b\u8981\u5185\u5bb9\u5c31\u662f\u8bc6\u522b\u76f8\u90bb\u7bc7\u7ae0\u5355\u5143\u4e4b\u95f4\u7684\u7ed3\u6784\uff0c\u5e76\u5c42\u6b21\u5316\u6784\u5efa\u7bc7\u7ae0\u7ed3\u6784\u6811\u3002", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u5728MCDTB\u8bed\u6599\u5e93\u4e0a\uff0c\u5df2\u6709\u7684\u7bc7\u7ae0\u7ed3\u6784\u8bc6\u522b\u7684\u7814\u7a76 (Jiang et al., 2018a; Zhou et al., 2019 ) \u90fd\u53ea\u8003\u8651\u76f8\u90bb\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\u7684\u8bed\u4e49\u5173\u7cfb\uff0c\u5982\u679c\u76f8\u90bb\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\u8bed\u4e49\u5173\u7cfb\u5f88\u63a5\u8fd1\uff0c\u90a3\u4e48\u8fd9\u4e24\u4e2a \u7bc7\u7ae0\u5355\u5143\u5c31\u4f1a\u5927\u6982\u7387\u4ee5\u67d0\u79cd\u5173\u7cfb\u8fde\u63a5\u8d77\u6765\uff0c\u5f62\u6210\u4e00\u4e2a\u66f4\u5927\u7684\u7bc7\u7ae0\u5355\u5143\uff0c\u8fdb\u800c\u5c42\u6b21\u5316\u7684\u6784\u5efa\u7bc7\u7ae0 \u7ed3\u6784\u6811\u3002\u4f46\u662f\u8fd9\u4e9b\u7814\u7a76\u90fd\u53ea\u8003\u8651\u5c40\u90e8\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\uff0c\u800c\u6ca1\u6709\u5c06\u6574\u4e2a\u6587\u7ae0\u7684\u8bed\u4e49\u4fe1\u606f(\u5168\u5c40\u4fe1 \u606f)\u6709\u6548\u7684\u8fd0\u7528\u5230\u7bc7\u7ae0\u7ed3\u6784\u8bc6\u522b\u4efb\u52a1\u4e2d\u3002 \u5728RST-DT (Carlson et al., 2007) (Vaswani et al., 2017; Xu et al., 2019) ", | |
| "cite_spans": [ | |
| { | |
| "start": 24, | |
| "end": 45, | |
| "text": "(Jiang et al., 2018a;", | |
| "ref_id": "BIBREF9" | |
| }, | |
| { | |
| "start": 46, | |
| "end": 63, | |
| "text": "Zhou et al., 2019", | |
| "ref_id": "BIBREF22" | |
| }, | |
| { | |
| "start": 218, | |
| "end": 240, | |
| "text": "(Carlson et al., 2007)", | |
| "ref_id": "BIBREF0" | |
| }, | |
| { | |
| "start": 241, | |
| "end": 263, | |
| "text": "(Vaswani et al., 2017;", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 264, | |
| "end": 280, | |
| "text": "Xu et al., 2019)", | |
| "ref_id": "BIBREF21" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u5bf9\u4e8e\u4e00\u7bc7\u6587\u7ae0\u8868\u793a\u4e3aP = {p 1 , p 2 , \u2022 \u2022 \u2022 , p m },\u5176\u4e2dp i \u662f\u6bb5\u843d\u8bcd\u8bed\u5e8f\u5217\uff0cm\u662f\u6587\u7ae0\u7684\u6bb5\u843d\u6570\u3002 \u5c06p i \u901a\u8fc7\u6bb5\u843d\u7f16\u7801\u5c42(PEL)\uff0c\u5f97\u5230\u6bb5\u843d\u7f16\u7801\u4e3aR = {r 1 , r 2 , \u2022 \u2022 \u2022 , r m }\u3002\u5c06\u76f8\u90bb\u4e24\u4e2a\u6bb5\u843d\u7684\u7f16\u7801 \u901a\u8fc7\u6bb5\u843d\u4ea4\u4e92\u5c42(PIL)\uff0c\u5f97\u5230\u8868\u793a\u76f8\u90bb\u4e24\u4e2a\u6bb5\u843d\u8bed\u4e49\u8054\u7cfb\u7684\u8868\u793aH = {h 1 , h 2 , \u2022 \u2022 \u2022 , h m\u22121 },h i \u8868 \u793a\u6bb5\u843dp i \u548cp i+1 \u4e4b\u95f4\u7684\u8bed\u4e49\u8054\u7cfb\u7684\u7d27\u5bc6\u7a0b\u5ea6\uff0c\u5373\u5f97\u5230\u4e86\u5c40\u90e8\u4fe1\u606f\u3002\u540c\u65f6\u5c06r i \u5e73\u5747\u6c60\u5316\u4e4b\u540e\u901a\u8fc7\u6307 \u9488\u7f51\u7edc\u7f16\u7801\u5c42\uff0c\u7f16\u7801\u5c42\u662f\u53cc\u5411GRU\uff0c\u6700\u540e\u4e00\u4e2a\u65f6\u95f4\u6b65\u8f93\u51fa\u4f5c\u4e3a\u6574\u4e2a\u7bc7\u7ae0\u7684\u8bed\u4e49\u8868\u793a\uff0c\u5373\u5168\u5c40\u4fe1 \u606f(\u4f8b\u5982e 5 \u8868\u793a\u6574\u4e2a\u7bc7\u7ae0\u7684\u8bed\u4e49)\u3002 \u6307\u9488\u7f51\u7edc\u89e3\u7801\u5c42\u662f\u5355\u5411GRU\uff0c\u672c\u6587\u6839\u636e\u6df1\u5ea6\u4f18\u5148\u7684\u539f\u5219\uff0c\u4f7f\u7528\u6808\u6765\u751f\u6210\u7bc7\u7ae0\u7ed3\u6784\u6811\u3002 \u5728\u7b2ct\u6b65\uff0c\u6808\u9876\u7684\u7bc7\u7ae0\u5355\u5143DU (l,r) \u51fa\u6808\u3002\u89e3\u7801\u5c42\u7684\u8f93\u5165\u4e3a\u7bc7\u7ae0\u5355\u5143DU (l,r) \u7684\u8bed\u4e49\u8868\u793ae r \uff0c\u5373\u7f16 \u7801\u5c42\u7b2cr\u6b65\u7684\u8f93\u51fa\uff1b\u89e3\u7801\u5c42\u7684\u8f93\u51fa\u4e3ad t ,d t \u548c\u5c40\u90e8\u4fe1\u606fH\u8fdb\u884c\u4ea4\u4e92\uff0c\u901a\u8fc7\u8ba1\u7b97\u6ce8\u610f\u529b\u6765\u878d\u5408\u5168 \u5c40\u4fe1\u606f\u548c\u5c40\u90e8\u4fe1\u606f\uff0c\u4ece\u800c\u4e3a\u6bcf\u4e00\u4e2ah i \u5206\u914d\u4e00\u4e2a\u6982\u7387\uff0c\u5176\u4e2dl \u2264 i \u2264 r \u2212 1\u3002\u6982\u7387\u8d8a\u5927\uff0c\u5219\u8868 \u793a\u6bb5\u843dp i \u548cp i+1 \u4e4b\u95f4\u7684\u8bed\u4e49\u8054\u7cfb\u8d8a\u677e\u6563\uff0c\u5219\u5e94\u8be5\u5728p i \u548cp i+1 \u4e4b\u95f4\u8fdb\u884c\u5207\u5206\uff0c\u5f62\u6210\u65b0\u7684\u7bc7\u7ae0\u5355 \u5143DU (l,i) \u548cDU (i+1,r) \u3002\u5207\u5206\u540e\u6bb5\u843d\u6570\u5927\u4e8e2\u7684\u7bc7\u7ae0\u5355\u5143\u5165\u6808\uff0c\u9012\u5f52\u5730\u5bf9\u6808\u9876\u7bc7\u7ae0\u5355\u5143\u8fdb\u884c\u5207\u5206\uff0c", | |
| "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": "\u53ef\u53d6 \u5f97\u4e0d\u9519\u7684\u6548\u679c\uff0c\u5176\u8ba1\u7b97\u516c\u5f0f\u5982\u5f0f(1)\u6240\u793a\u3002 Attention (Q, K, V ) = sof tmax QK T \u221a d k V", | |
| "eq_num": "(1)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u5728\u5176\u7f16\u7801\u7684\u8fc7\u7a0b\u4e2d\uff0c\u5e8f\u5217\u4e2d\u7684\u6bcf\u4e00\u4e2a\u8bcd\u8bed\u90fd\u4e0e\u5e8f\u5217\u4e2d\u7684\u5176\u4ed6\u8bcd\u8bed\u8fdb\u884c\u5339\u914d\u8ba1\u7b97\uff0c\u56e0\u800c\u66f4 \u5bb9\u6613\u6355\u83b7\u957f\u8ddd\u79bb\u8bcd\u8bed\u4e4b\u95f4\u7684\u4f9d\u8d56\u5173\u7cfb\uff0c\u672c\u8d28\u4e0a\u6ce8\u610f\u529b\u673a\u5236\u662f\u5bf9\u8f93\u5165\u5e8f\u5217\u8fdb\u884c\u52a0\u6743\u6c42\u548c\uff0c\u56e0\u800c \u6bd4LSTM\u4fdd\u7559\u4e86\u66f4\u591a\u7684\u539f\u59cb\u8f93\u5165\u7684\u4fe1\u606f\u3002\u800c\u591a\u5934\u6ce8\u610f\u529b\u673a\u5236\u5141\u8bb8\u6a21\u578b\u53ef\u4ee5\u5728\u4e0d\u540c\u7684\u8868\u793a\u5b50\u7a7a\u95f4 \u4e2d\u5b66\u4e60\u5230\u76f8\u5173\u7684\u4fe1\u606f\uff0c\u53ef\u4ee5\u4f7f\u5f97\u6a21\u578b\u66f4\u597d\u7684\u6355\u83b7\u957f\u8fdc\u8ddd\u79bb\u4f9d\u8d56\u5173\u7cfb\u3002\u56e0\u6b64\u5728PNGL\u6a21\u578b\u4e2d\uff0c\u672c \u6587\u91c7\u7528\u591a\u5934\u6ce8\u610f\u529b\u673a\u5236\u8fdb\u884c\u6bb5\u843d\u5c42\u7f16\u7801\u3002\u5982\u5f0f(2)\u6240\u793a\u3002 M ultiHAtt (Q, K, V ) = Concat (head 1 , \u2022 \u2022 \u2022 , head h ) W O head i = Attention QW Q i , KW K i , V W V i (2) \u53c2\u6570\u77e9\u9635W Q i \u2208R dm\u00d7d k ,W K i \u2208R dm\u00d7d k ,W V i \u2208R dm\u00d7dv ,W O \u2208R hdv\u00d7dm \u3002\u6bb5\u843d\u7f16\u7801\u5c42\u8f93\u5165\u8bcd\u8bed\u5e8f \u5217p = {x 1 , x 2 , \u2022 \u2022 \u2022 , x l }\uff0cl\u662f\u6bb5\u843d\u4e2d\u8bcd\u8bed\u7684\u4e2a\u6570\uff0c\u6bcf\u4e00\u4e2a\u8bcd\u8bedx i \u2208 R e \u4f7f\u7528\u5176\u5bf9\u5e94\u7684\u8bcd\u5411\u91cf\u8868\u793a\u3002 \u5f97\u5230\u6bb5\u843d\u7f16\u7801\u7ed3\u679cr i \u2208 R m\u00d7dm \uff0c\u5982\u5f0f(3)\u6240\u793a\uff0c\u5176\u4e2dW Q S , W K S , W V S \u2208 R dm \u662f\u5171\u4eab\u7684\u8f6c\u6362\u77e9 \u9635\uff0c\u4ece\u800c\u5728\u7f16\u7801\u65f6\u5c06\u6bb5\u843d\u6620\u5c04\u5230\u76f8\u540c\u7684\u7279\u5f81\u7a7a\u95f4\u3002 r i = M ultiHAttt(pW Q S , pW K S , pW V S ) (3) 3.2 \u6bb5 \u6bb5 \u6bb5\u843d \u843d \u843d\u4ea4 \u4ea4 \u4ea4\u4e92 \u4e92 \u4e92\u5c42 \u5c42 \u5c42 \u6bb5\u843d\u4ea4\u4e92\u5c42(PIL)\u7528\u6765\u6355\u83b7\u76f8\u90bb\u4e24\u4e2a\u6bb5\u843d\u4e4b\u95f4\u7684\u8bed\u4e49\u8054\u7cfb(\u5c40\u90e8\u4fe1\u606f)\u3002\u4e00\u4e9b\u7814\u7a76\u4eba\u5458 \u901a\u8fc7\u6ce8\u610f\u529b\u673a\u5236\u76f4\u63a5\u5bf9\u5e8f\u5217\u4e4b\u95f4\u7684\u4ea4\u4e92\u5efa\u6a21\uff0c\u5e76\u4e14\u63d0\u51fa\u4e86\u4e00\u4e9b\u4ea4\u4e92\u6ce8\u610f\u529b\u673a\u5236\u3002\u4f8b\u5982\uff0cGuo et al. (2018)\u63d0\u51fa\u4e00\u79cd\u6a21\u62df\u53cc\u5411\u9605\u8bfb\u7684\u4ea4\u4e92\u6ce8\u610f\u529b\u673a\u5236\uff0c\u4ed6\u4ece\u4eba\u7c7b\u9605\u8bfb\u7684\u89d2\u5ea6\u51fa\u53d1\uff0c\u53d1\u73b0\u4eba\u7c7b\u5728\u5224 \u65ad\u4e24\u4e2a\u5e8f\u5217\u4e4b\u95f4\u7684\u5173\u7cfb\u65f6\u5f80\u5f80\u9700\u8981\u6765\u56de\u9605\u8bfb\u8fd9\u4e24\u4e2a\u5e8f\u5217\uff0c\u5c24\u5176\u662f\u8003\u8651\u4e24\u4e2a\u5e8f\u5217\u4e2d\u8054\u7cfb\u6bd4\u8f83\u7d27\u5bc6 \u7684\u8bcd\u4e4b\u95f4\u7684\u8bed\u4e49\u8054\u7cfb\u3002\u53d7\u4ea4\u4e92\u6ce8\u610f\u529b\u673a\u5236\u5de5\u4f5c\u7684\u5f71\u54cd\uff0cXu et al. (2019)\u91c7\u7528\u5f0f(1)\u5bf9\u5e8f\u5217\u4e4b\u95f4 \u7684\u4ea4\u4e92\u8fdb\u884c\u5efa\u6a21\uff0c\u5e76\u5728\u7bc7\u7ae0\u5173\u7cfb\u8bc6\u522b\u4efb\u52a1\u4e2d\u53d6\u5f97\u4e86\u4e0d\u9519\u7684\u6548\u679c\uff0c\u56e0\u6b64\u672c\u6587\u5229\u7528\u591a\u5934\u4ea4\u4e92\u6ce8\u610f\u529b \u673a\u5236\u83b7\u5f97\u6bb5\u843d\u4e4b\u95f4\u4ea4\u4e92\u7684\u8bed\u4e49\u8054\u7cfb\u3002 \u5bf9\u4e8e\u4e24\u4e2a\u6bb5\u843dp 1 = {x 1 , x 2 , \u2022 \u2022 \u2022 , x m }\u548cp 2 = {x 1 , x 2 , \u2022 \u2022 \u2022 , x n }\uff0c\u4f7f\u7528\u5f0f(3)\u5f97\u5230\u6bb5\u843d\u7f16 \u7801r 1 \u548cr 2 ,\u7136\u540e\u4f7f\u7528\u5f0f(4)\u5bf9\u6bb5\u843d\u4e4b\u95f4\u7684\u4ea4\u4e92\u8fdb\u884c\u5efa\u6a21\u3002 I 1 = M ultiHAtt r 2 W Q i1 , r 1 W K i1 , r 1 W V i1 I 2 = M ultiHAtt r 1 W Q i2 , r 2 W K i2 , r 2 W V i2 (4) \u5f0f(4)\u9996\u5148\u901a\u8fc7\u8f6c\u6362\u77e9\u9635W Q i1 , W K i1 , W V i1 \u2208 R dm\u00d7d i \u548cW Q i2 , W K i2 , W V i2 \u2208 R dm\u00d7d i \u5bf9\u8f93\u5165\u5e8f\u5217\u505a \u4e86\u6620\u5c04\u3002\u5728\u591a\u5934\u6ce8\u610f\u529b\u4ea4\u4e92\u5c42\uff0c\u901a\u8fc7\u4ea4\u6362\u4e24\u4e2a\u5e8f\u5217\u7684query\u503c\uff0c\u6bcf\u4e2a\u5e8f\u5217\u7684\u8bcd\u8bed\u90fd\u6839\u636e\u4e0e\u53e6 \u4e00\u4e2a\u5e8f\u5217\u4e2d\u6240\u6709\u8bcd\u8bed\u7684\u8054\u7cfb\u8fdb\u884c\u4e86\u91cd\u65b0\u7f16\u7801\uff0c\u4ece\u800c\u5f97\u5230\u6bb5\u843dp 1 \u548cp 2 \u5f7c\u6b64\u76f8\u5173\u7684\u5411\u91cf\u8868\u793aI 1 \u2208 R m\u00d7d i \u548cI 2 \u2208 R n\u00d7d i \u3002\u6700\u540e\u901a\u8fc7\u5e73\u5747\u6c60\u5316\u64cd\u4f5c\u83b7\u5f97\u5305\u542b\u5f7c\u6b64\u4fe1\u606f\u7684\u6bb5\u843d\u8868\u793aC 1 , C 2 \u2208 R d i \u3002\u5728\u5305 \u542b\u5f7c\u6b64\u4fe1\u606f\u7684\u6bb5\u843d\u8868\u793aC 1 \uff0cC 2 \u4e0a\uff0c\u901a\u8fc7\u975e\u7ebf\u6027\u53d8\u6362\u8fdb\u4e00\u6b65\u6355\u83b7\u6bb5\u843d\u4e4b\u95f4\u7684\u4ea4\u4e92\u4fe1\u606f\uff0c\u5c06\u53d8\u6362\u5f97 \u5230\u7684\u5411\u91cfh 1 \u8868\u793a\u6bb5\u843dp 1 \u548cp 2 \u4e4b\u95f4\u8bed\u4e49\u8054\u7cfb\u7684\u7d27\u5bc6\u7a0b\u5ea6,\u5982\u5f0f(5)\u6240\u793a,\u5176\u4e2d,W h \u2208 R dm\u00d73d i \u662f\u53c2\u6570 \u77e9\u9635\u3002 h 1 = tanh(W h [C 1 , C 2 , C 1 \u2212 C 2 ])", | |
| "eq_num": "(5)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u5206\u522b\u662f\u6b63\u5411\u548c\u53cd\u5411\u7684\u8f93\u51fa\u3002\u6b64\u65f6e i \u7efc\u5408\u4e86\u524d\u9762i-1\u4e2a\u6bb5 \u843d\u7684\u8bed\u4e49\u4fe1\u606f\uff0c\u5373\u83b7\u5f97\u5168\u5c40\u4fe1\u606f\u3002\u800c\u8be5\u5168\u5c40\u4fe1\u606f\u9690\u542b\u4e86\u7bc7\u7ae0\u5355\u5143\u4e4b\u95f4\u7684\u7ed3\u6784\u4fe1\u606f\u548c\u8bed\u4e49\u8054\u7cfb\uff0c \u5bf9\u4e8e\u6700\u7ec8\u7bc7\u7ae0\u7ed3\u6784\u6811\u7684\u6784\u5efa\u8d77\u7740\u4e0d\u53ef\u5ffd\u89c6\u7684\u4f5c\u7528\u3002 3.3.2 \u89e3 \u89e3 \u89e3\u7801 \u7801 \u7801\u5c42 \u5c42 \u5c42 \u5728 \u89e3 \u7801 \u5c42 \u91c7 \u7528 \u7684 \u4e5f \u662f \u4e00 \u4e2a \u4e24 \u5c42 \u7684GRU\u3002 \u4ee5chtb 0282\u4e3a \u4f8b \uff0c \u672c \u6587 \u5c06 \u7f16 \u7801 \u5c42 \u7684 \u8f93 \u51faE = {e 1 , e 2 , e", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "3 , e 4 , e 5 }\u4f5c\u4e3aDecoder\u5c42\u7684\u8f93\u5165\u3002\u5728\u7b2ct\u6b65\u89e3\u7801\u65f6\uff0c\u7bc7\u7ae0\u5355\u5143DU (l,r) \u51fa\u6808\uff0c\u89e3\u7801\u5c42\u4f1a\u7efc \u5408\u5f53\u524d\u7bc7\u7ae0\u7684\u5168\u5c40\u4fe1\u606fe r \u548ct\u6b65\u4e4b\u524d\u751f\u6210\u7684\u7ed3\u6784\u8bed\u4e49\u4fe1\u606f\u751f\u6210\u5f53\u524d\u72b6\u6001d t \u3002d t \u548c\u6bb5\u843d\u4ea4\u4e92\u5c42\u7684\u8f93", | |
| "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": "\u51faH = {h l , h l+1 , \u2022 \u2022 \u2022 , h r\u22121 }\u8fdb\u884c\u4ea4\u4e92\uff0c\u878d\u5408\u5168\u5c40\u548c\u5c40\u90e8\u4fe1\u606f\uff0c\u901a\u8fc7\u4e00\u4e2asoftmax\u5c42\u5f97\u5230\u5173\u4e8eH\u7684 \u6982\u7387\u5206\u5e03\u3002\u5982\u5f0f(6)\u6240\uff0c\u5176\u4e2d\u03c3(., .)\u662f\u878d\u5408\u5168\u5c40\u548c\u5c40\u90e8\u4fe1\u606f\u7684\u51fd\u6570\uff0c\u5177\u4f53\u4e3a\u70b9\u79ef\u8fd0\u7b97\uff1b\u03b1 t \u4e3a\u5173 \u4e8eH\u7684\u6982\u7387\u5206\u5e03\u3002 s t,i = \u03c3(d t , h i ), i = l \u2022 \u2022 \u2022 r \u2212 1 \u03b1 t = sof tmax(s t ) = exp(s t,i ) r\u22121 i=l exp(s t,i ) (6) \u5982\u679c\u901a\u8fc7softmax\u5c42\u540eh i \u88ab\u5206\u914d\u7684\u6982\u7387\u503c\u8d8a\u5927\uff0c\u8868\u660e\u6bb5\u843dp i \u548cp i+1 \u4e4b\u95f4\u7684\u8bed\u4e49\u8054\u7cfb\u8d8a\u677e\u6563\uff0c \u56e0\u6b64\u66f4\u5e94\u8be5\u5207\u5206\u5f00\uff0c\u4ece\u800c\u5c06\u6574\u4e2a\u7bc7\u7ae0\u5206\u4e3a\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143DU (l,i) \u548cDU (i+1,r) \u3002\u6839\u636e\u6df1\u5ea6\u4f18\u5148\u7684\u539f \u5219\uff0c\u6bcf\u4e00\u6b65\u89e3\u7801\uff0c\u6bb5\u843d\u6570\u91cf\u5927\u4e8e2\u7684\u7bc7\u7ae0\u5355\u5143\u5c06\u7ee7\u7eed\u5165\u6808\uff0c\u9012\u5f52\u5730\u5bf9\u7bc7\u7ae0\u5355\u5143\u8fdb\u884c\u5207\u5206\uff0c\u76f4\u81f3\u6808 \u7a7a\uff0c\u8fc7\u7a0b\u5982\u56fe3\u6240\u793a\u3002 DU (1,5) 1 2 3 4 5 DU (1,5) DU (1,4) DU (5,5) 1 2 3 4 5 DU (1,5) DU (1,4) DU (5,5) DU (1,2) DU (3,4) 1 2 3 4 5 t=1 t=2 \u7bc7\u7ae0\u7ed3\u6784\u6811 \u56fe 3. \u89e3\u7801\u8fc7\u7a0b 3.4 \u635f \u635f \u635f\u5931 \u5931 \u5931\u51fd \u51fd \u51fd\u6570 \u6570 \u6570 \u5728PNGL\u6a21\u578b\u4e2d\uff0c\u635f\u5931\u51fd\u6570\u672c\u6587\u91c7\u7528\u8d1f\u5bf9\u6570\u4f3c\u7136\u51fd\u6570\u8fdb\u884c\u8ba1\u7b97\uff0c\u5982\u5f0f(7)\u6240\u793a\u3002y <t \u662f\u5728 \u89e3\u7801\u5c42\u7b2ct\u6b65\u4e4b\u524d\u5df2\u7ecf\u4ea7\u751f\u7684\u7bc7\u7ae0\u5355\u5143\uff0cT\u662f\u5165\u6808\u7684\u7bc7\u7ae0\u5355\u5143\u6570\u3002\u4e3a\u4e86\u9632\u6b62\u8fc7\u62df\u5408\uff0c\u672c\u6587\u5728\u6307\u9488 \u7f51\u7edc\u7684\u7f16\u7801\u548c\u89e3\u7801\u5c42\u8fdb\u884c\u4e86dropout\u64cd\u4f5c\u3002 L(\u03b8 s ) = \u2212 batch i=1 T t=1 logP \u03b8s (y t |y <t , X)", | |
| "eq_num": "(7)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "4.1 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u8bbe \u8bbe \u8bbe\u7f6e \u7f6e \u7f6e ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u672c\u6587\u5728\u5b8f\u89c2\u6c49\u8bed\u7bc7\u7ae0\u6811\u5e93(MCDTB)\u4e0a\u5bf9\u6a21\u578b\u7ed3\u6784\u8bc6\u522b\u7684\u6027\u80fd\u8fdb\u884c\u4e86\u8bc4\u4f30\u3002MCDTB\u5b9a \u4e49 \u4e86 \u4e09 \u5927 \u7c7b \u5341 \u4e94 \u5c0f \u7c7b \u7bc7 \u7ae0 \u5173 \u7cfb \uff0c \u5e76 \u6807 \u6ce8 \u4e86 \u6458 \u8981 \uff0c \u6bb5 \u843d \u4e2d \u5fc3 \u53e5 \u3001 \u7bc7 \u7ae0 \u7ed3 \u6784 \u7b49 \u5b8f \u89c2 \u7bc7 \u7ae0 \u4fe1 \u606f\u3002MCDTB\u603b\u8ba1\u6709720\u7bc7\u65b0\u95fb\u62a5\u9053\u7684\u6587\u7ae0\uff0c\u6bcf\u7bc7\u6587\u7ae0\u7684\u6bb5\u843d\u6570\u4ece2\u523022\u4e0d\u7b49\uff0c\u5176\u6bb5\u843d\u5206\u5e03\u5982", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u88684.\u52a0\u5165\u5c40\u90e8\u4fe1\u606f\u540e\u6a21\u578b\u8bc6\u522b\u6027\u80fd\u6bd4\u8f83 \u88684\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u5728\u52a0\u5165\u5c40\u90e8\u4fe1\u606f\u4e4b\u540ePNGL\u548cPN(+local)\u7684\u6700\u5e95\u4e0b\u4e24\u5c42\u5185\u90e8\u8282\u70b9 \u6b63 \u786e \u7387 \u5206 \u522b \u63d0 \u9ad8 \u4e861.01\u548c5.8\u3002PNGL\u76f8 \u8f83 \u4e8ePN(+local) \uff0c \u6027 \u80fd \u6709 \u66f4 \u591a \u7684 \u63d0 \u5347 \uff0c \u5176 \u539f \u56e0 \u5728 \u4e8ePN(+local)\u662f\u76f4\u63a5\u4f7f\u7528\u53cc\u5411GRU\u5bf9\u6bb5\u843d\u8fdb\u884c\u7f16\u7801\uff0c\u800cPNGL\u662f\u4f7f\u7528\u591a\u5934\u6ce8\u610f\u529b\u673a\u5236\u5bf9 \u6bb5\u843d\u8fdb\u884c\u7f16\u7801\uff0c\u7531\u4e8e\u591a\u5934\u6ce8\u610f\u529b\u673a\u5236\u76f8\u8f83\u4e8e\u53cc\u5411GRU\u66f4\u5bb9\u6613\u6355\u83b7\u957f\u8ddd\u79bb\u5355\u8bcd\u4e4b\u95f4\u7684\u4f9d\u8d56\u5173\u7cfb\uff0c \u80fd\u4fdd\u7559\u66f4\u591a\u539f\u59cb\u7684\u4fe1\u606f\uff0c\u5bf9\u6bb5\u843d\u7684\u7f16\u7801\u66f4\u6709\u6548\u3002\u90a3\u4e48\u76f8\u90bb\u4e24\u4e2a\u6bb5\u843d\u7684\u7f16\u7801\u8f93\u5165\u5230\u6bb5\u843d\u4ea4\u4e92\u5c42\u8fdb \u884c\u4ea4\u4e92\uff0c\u6bb5\u843d\u4ea4\u4e92\u5c42\u5c31\u80fd\u66f4\u597d\u7684\u6355\u83b7\u6bb5\u843d\u4e4b\u95f4\u7684\u8bed\u4e49\u8054\u7cfb\u3002\u901a\u8fc7\u6355\u83b7\u5230\u66f4\u597d\u7684\u5c40\u90e8\u4fe1\u606f\uff0c\u6a21 \u578bPNGL\u589e\u5f3a\u4e86\u5bf9\u5e95\u5c42\u7ed3\u6784\u7684\u8bc6\u522b\u80fd\u529b\uff0c\u4ece\u800c\u4ece\u6574\u4f53\u4e0a\u63d0\u9ad8\u4e86\u6a21\u578b\u7684\u6027\u80fd\u3002 5.2 \u6a21 \u6a21 \u6a21\u578b \u578b \u578b\u5bf9 \u5bf9 \u5bf9\u957f \u957f \u957f\u77ed \u77ed \u77ed\u6587 \u6587 \u6587\u8bc6 \u8bc6 \u8bc6\u522b \u522b \u522b\u6027 \u6027 \u6027\u80fd \u80fd \u80fd\u6bd4 \u6bd4 \u6bd4\u8f83 \u8f83 \u8f83 \u4e3a\u4e86\u6bd4\u8f83\u6a21\u578b\u5bf9\u4e8e\u957f\u6587\u548c\u77ed\u6587\u7684\u8bc6\u522b\u80fd\u529b\uff0c\u672c\u6587\u5206\u522b\u7edf\u8ba1\u4e86\u957f\u6587\u548c\u77ed\u6587\u5185\u90e8\u8282\u70b9\u6b63\u786e\u7387\uff0c \u5982\u88685\u6240\u793a\u3002\u4ece\u8868\u4e2d\u6570\u636e\u53ef\u77e5\uff0c\u6a21\u578b\u5bf9\u77ed\u6587\u7ed3\u6784\u7684\u8bc6\u522b\u6027\u80fd\u8f83\u597d\uff0c\u800c\u957f\u6587\u7ed3\u6784\u7684\u8bc6\u522b\u6027\u80fd\u8f83\u5dee\u3002 \u4e3b\u8981\u539f\u56e0\u5728\u4e8e\u65e0\u8bba\u91c7\u7528\u4ec0\u4e48\u65b9\u6cd5\u6784\u5efa\u7bc7\u7ae0\u7ed3\u6784\u6811\uff0c\u90fd\u4f1a\u4ea7\u751f\u7ea7\u8054\u9519\u8bef\uff0c\u800c\u5bf9\u957f\u6587\u6765\u8bf4\uff0c\u5219\u66f4\u52a0 \u660e\u663e\u3002\u4f46\u548c\u53ea\u8003\u8651\u5c40\u90e8\u4fe1\u606f\u7684\u6a21\u578b\u4ee5\u53ca\u53ea\u8003\u8651\u5168\u5c40\u4fe1\u606f\u7684\u6a21\u578b\u76f8\u6bd4\uff0c\u672c\u6587\u7684\u6a21\u578bPNGL\u7efc\u5408\u8003 \u8651\u5168\u5c40\u548c\u5c40\u90e8\u4fe1\u606f\uff0c\u5728\u77ed\u6587\u548c\u957f\u6587\u7684\u7ed3\u6784\u8bc6\u522b\u7684\u6027\u80fd\u90fd\u6709\u63d0\u5347\u3002 0 \u7531\u4e8e\u6700\u9876\u5c42\u7684\u6839\u8282\u70b9\u6240\u8868\u793a\u7684\u7ed3\u6784\u603b\u662f\u56fa\u5b9a\uff0c\u56e0\u6b64\u672c\u6587\u8003\u8651\u6700\u9876\u4e0a\u4e09\u5c42\u548c\u6700\u5e95\u4e0b\u4e24\u5c42\u5185\u90e8\u7ed3\u70b9\u6b63\u786e\u7387\u6765\u8868\u793a\u6a21 \u578b\u5bf9\u4e8e\u9876\u5c42\u548c\u5e95\u5c42\u7ed3\u6784\u8bc6\u522b\u7684\u6027\u80fd\u597d\u574f\u3002 \u6a21\u578b \u5185\u90e8\u8282\u70b9\u6b63\u786e\u7387% \u2264 6 > 6 LD-CM", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u91c7 \u7528 \u6808 \u6570 \u636e \u7ed3 \u6784 \uff0c \u901a \u8fc7 \u81ea \u9876 \u5411 \u4e0b \u7684 \u65b9 \u6cd5 \u9012 \u5f52 \u786e \u5b9a \u6587 \u7ae0 \u7684 \u5207 \u5206 \u4f4d \u7f6e \uff0c \u4ece \u800c \u5f62 \u6210 \u7ed3 \u6784 \u6811 \u3002PNGL(-local)\u9996 \u5148 \u4f1a \u5bf9DU", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "(1,1) \u548cDU (2,5) \u3001DU (1,2) \u548cDU (3,5) \u3001DU (1,3) \u548cDU (4,5) \u3001DU (1,4) \u548cDU (5,5) \u8fd9 \u56db \u4e2a \u8bed \u4e49 \u8054 \u7cfb\u7684\u7d27\u5bc6\u7a0b\u5ea6\u8fdb\u884c\u6392\u5e8f\uff0c\u786e\u5b9aDU (1,4) \u548cDU (5,5) \u4e4b\u95f4\u7684\u8bed\u4e49\u8054\u7cfb\u6700\u677e\u6563\uff0c\u7136\u540e\u9012\u5f52\u5730\u5bf9DU (1,4) \u8fdb \u884c\u4ee5\u4e0a\u8fc7\u7a0b\uff0c\u786e\u5b9aDU (1,2) \u548cDU (3 ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [ | |
| { | |
| "text": "A chtb 0756\u6587 \u6587 \u6587\u7ae0 \u7ae0 \u7ae0\u5185 \u5185 \u5185\u5bb9 \u5bb9 \u5bb9\u53ca \u53ca \u53ca\u6807 \u6807 \u6807\u51c6 \u51c6 \u51c6\u7ed3 \u7ed3 \u7ed3\u6784 \u6784 \u6784\u6811 \u6811 \u6811", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "annex", | |
| "sec_num": null | |
| } | |
| ], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "RST Discourse Treebank.Current and new directions in discourse and dialogue", | |
| "authors": [ | |
| { | |
| "first": "Lynn", | |
| "middle": [], | |
| "last": "Carlson", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Daniel", | |
| "middle": [], | |
| "last": "Marcu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mary", | |
| "middle": [ | |
| "Ellen" | |
| ], | |
| "last": "Okurowski", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "85--112", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lynn Carlson, Daniel Marcu, and Mary Ellen Okurowski. 2007. RST Discourse Treebank.Current and new directions in discourse and dialogue. pages 85-112.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Learning phrase representations using RNN encoder-decoder for statistical machine translation", | |
| "authors": [ | |
| { | |
| "first": "Kyunghyun", | |
| "middle": [], | |
| "last": "Cho", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bart", | |
| "middle": [], | |
| "last": "Van Merrienboer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Caglar", | |
| "middle": [], | |
| "last": "Gulcehre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Dzmitry", | |
| "middle": [], | |
| "last": "Bahdanau", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Fethi", | |
| "middle": [], | |
| "last": "Bougares", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Holger", | |
| "middle": [], | |
| "last": "Schwenk", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yoshua", | |
| "middle": [], | |
| "last": "Bengio", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1724--1734", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kyunghyun Cho, Bart Van Merrienboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning phrase representations using RNN encoder-decoder for statistical machine translation. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, pages 1724-1734.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Macro discourse structure representation schema and corpus construction", | |
| "authors": [ | |
| { | |
| "first": "Xiaomin", | |
| "middle": [], | |
| "last": "Chu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xuefeng", | |
| "middle": [], | |
| "last": "Xi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Feng", | |
| "middle": [], | |
| "last": "Jiang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sheng", | |
| "middle": [], | |
| "last": "Xu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Qiaoming", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Guodong", | |
| "middle": [], | |
| "last": "Zhou", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2020, | |
| "venue": "Journal of Software", | |
| "volume": "31", | |
| "issue": "2", | |
| "pages": "321--343", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Xiaomin Chu, Xuefeng Xi, Feng Jiang, Sheng Xu, Qiaoming Zhu, and Guodong Zhou. 2020. Macro discourse structure representation schema and corpus construction. Journal of Software, 31(2):321- 343.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Empirical evaluation of gated recurrent neural networks on sequence modeling", | |
| "authors": [ | |
| { | |
| "first": "Junyoung", | |
| "middle": [], | |
| "last": "Chung", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Caglar", | |
| "middle": [], | |
| "last": "Gulcehre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kyung", | |
| "middle": [ | |
| "Hyun" | |
| ], | |
| "last": "Cho", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yoshua", | |
| "middle": [], | |
| "last": "Bengio", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Junyoung Chung, Caglar Gulcehre, Kyung Hyun Cho, and Yoshua Bengio. 2014. Empirical evaluation of gated recurrent neural networks on sequence modeling. Computer Research Repository.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Scientific article summarization using citation-context and article's discourse structure", | |
| "authors": [ | |
| { | |
| "first": "Arman", | |
| "middle": [], | |
| "last": "Cohan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Nazli", | |
| "middle": [], | |
| "last": "Goharian", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "390--400", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Arman Cohan and Nazli Goharian. 2015. Scientific article summarization using citation-context and article's discourse structure. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 390-400.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Implicit discourse relation recognition using neural tensor network with interactive attention and sparse learning", | |
| "authors": [ | |
| { | |
| "first": "Fengyu", | |
| "middle": [], | |
| "last": "Guo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ruifang", | |
| "middle": [], | |
| "last": "He", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Di", | |
| "middle": [], | |
| "last": "Jin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jianwu", | |
| "middle": [], | |
| "last": "Dang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Longbiao", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xiangang", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the 27th International Conference on Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "547--558", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Fengyu Guo, Ruifang He, Di Jin, Jianwu Dang, Longbiao Wang, and Xiangang Li. 2018. Implicit discourse relation recognition using neural tensor network with interactive attention and sparse learning. In Proceedings of the 27th International Conference on Computational Linguistics, pages 547-558.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "HILDA: A discourse parser using support vector machine classification", | |
| "authors": [ | |
| { | |
| "first": "Hugo", | |
| "middle": [], | |
| "last": "Hernault", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Helmut", | |
| "middle": [], | |
| "last": "Prendinger", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "David", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Duverle", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mitsuru", | |
| "middle": [], | |
| "last": "Ishizuka", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Dialogue and Discourse", | |
| "volume": "1", | |
| "issue": "3", | |
| "pages": "1--33", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hugo Hernault, Helmut Prendinger, David A. Duverle, and Mitsuru Ishizuka. 2010. HILDA: A discourse parser using support vector machine classification. Dialogue and Discourse, 1(3):1-33.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "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": "", | |
| "issue": "", | |
| "pages": "1735--1780", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Sepp Hochreiter and J\u00fcrgen Schmidhuber. 1997. Long short-term memory. Neural Computation, pages 1735-1780.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Representation learning for text-level discourse parsing", | |
| "authors": [ | |
| { | |
| "first": "Yangfeng", | |
| "middle": [], | |
| "last": "Ji", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jacob", | |
| "middle": [], | |
| "last": "Eisenstein", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "13--24", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Yangfeng Ji and Jacob Eisenstein. 2014. Representation learning for text-level discourse parsing. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13-24.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Recognizing macro Chinese discourse structure on label degeneracy combination model", | |
| "authors": [ | |
| { | |
| "first": "Feng", | |
| "middle": [], | |
| "last": "Jiang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Peifeng", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xiaomin", | |
| "middle": [], | |
| "last": "Chu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Qiaoming", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Guodong", | |
| "middle": [], | |
| "last": "Zhou", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the 7th CCF International Conference on Natural Language Processing and Chinese Computing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "92--104", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Feng Jiang, Peifeng Li, Xiaomin Chu, Qiaoming Zhu, and Guodong Zhou. 2018a. Recognizing macro Chinese discourse structure on label degeneracy combination model. In Proceedings of the 7th CCF International Conference on Natural Language Processing and Chinese Computing, pages 92-104.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "MCDTB: A macro-level Chinese discourse treebank", | |
| "authors": [ | |
| { | |
| "first": "Feng", | |
| "middle": [], | |
| "last": "Jiang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sheng", | |
| "middle": [], | |
| "last": "Xu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xiaomin", | |
| "middle": [], | |
| "last": "Chu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Peifeng", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Qiaoming", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Guodong", | |
| "middle": [], | |
| "last": "Zhou", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the 27th International Conference on Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "3493--3504", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Feng Jiang, Sheng Xu, Xiaomin Chu, Peifeng Li, Qiaoming Zhu, and Guodong Zhou. 2018b. MCDTB: A macro-level Chinese discourse treebank. In Proceedings of the 27th International Conference on Computational Linguistics, pages 3493-3504.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Combining intra-and multisentential rhetorical parsing for document-level discourse analysis", | |
| "authors": [ | |
| { | |
| "first": "Shafiq", | |
| "middle": [], | |
| "last": "Joty", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Giuseppe", | |
| "middle": [], | |
| "last": "Carenini", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Raymond", | |
| "middle": [], | |
| "last": "Ng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yashar", | |
| "middle": [], | |
| "last": "Mehdad", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "486--496", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Shafiq Joty, Giuseppe Carenini, Raymond Ng, and Yashar Mehdad. 2013. Combining intra-and multi- sentential rhetorical parsing for document-level discourse analysis. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 486-496.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "A discourse-driven content model for summarising scientific articles evaluated in a complex question answering task", | |
| "authors": [ | |
| { | |
| "first": "Maria", | |
| "middle": [], | |
| "last": "Liakata", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Simon", | |
| "middle": [], | |
| "last": "Dobnik", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Shyamasree", | |
| "middle": [], | |
| "last": "Saha", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Colin", | |
| "middle": [], | |
| "last": "Batchelor", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Dietrich", | |
| "middle": [], | |
| "last": "Rebholz-Schuhmann", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "747--757", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Maria Liakata, Simon Dobnik, Shyamasree Saha, Colin Batchelor, and Dietrich Rebholz-Schuhmann. 2013. A discourse-driven content model for summarising scientific articles evaluated in a complex question answering task. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 747-757.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "A unified linear-time framework for sentence-level discourse parsing", | |
| "authors": [ | |
| { | |
| "first": "Xiang", | |
| "middle": [], | |
| "last": "Lin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Shafiq", | |
| "middle": [], | |
| "last": "Joty", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Prathyusha", | |
| "middle": [], | |
| "last": "Jwalapuram", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Bari", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "4190--4200", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Xiang Lin, Shafiq Joty, Prathyusha Jwalapuram, and M Saiful Bari. 2019. A unified linear-time frame- work for sentence-level discourse parsing. In Proceedings of the 57th Annual Meeting of the Associ- ation for Computational Linguistics, pages 4190-4200.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "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": "Greg", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Corrado", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jeff", | |
| "middle": [], | |
| "last": "Dean", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "Proceedings of the 27th Annual Conference on Neural Information Processing Systems", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "3111--3119", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013. Distributed representa- tions of words and phrases and their compositionality. In Proceedings of the 27th Annual Conference on Neural Information Processing Systems, pages 3111-3119.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "How much progress have we made on RST discourse parsing? A replication study of recent results on the RST-DT", | |
| "authors": [ | |
| { | |
| "first": "Mathieu", | |
| "middle": [], | |
| "last": "Morey", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Philippe", | |
| "middle": [], | |
| "last": "Muller", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Nicholas", | |
| "middle": [], | |
| "last": "Asher", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1319--1324", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Mathieu Morey, Philippe Muller, and Nicholas Asher. 2017. How much progress have we made on RST discourse parsing? A replication study of recent results on the RST-DT. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 1319-1324.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Combining hierarchical clustering and machine learning to predict high-level discourse structure", | |
| "authors": [ | |
| { | |
| "first": "Caroline", | |
| "middle": [], | |
| "last": "Sporleder", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Alex", | |
| "middle": [], | |
| "last": "Lascarides", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Proceedings of the 20th International Conference on Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "43--49", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Caroline Sporleder and Alex Lascarides. 2004. Combining hierarchical clustering and machine learning to predict high-level discourse structure. In Proceedings of the 20th International Conference on Computational Linguistics, pages 43-49.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "Sequence to sequence learning with neural networks", | |
| "authors": [ | |
| { | |
| "first": "Ilya", | |
| "middle": [], | |
| "last": "Sutskever", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Oriol", | |
| "middle": [], | |
| "last": "Vinyals", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Quoc V", | |
| "middle": [], | |
| "last": "Le", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the 28th 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 Proceedings of the 28th Annual Conference on Neural Information Processing Systems, pages 3104-3112.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Macrostructures : An interdisciplinary study of global structures in discourse, interaction", | |
| "authors": [ | |
| { | |
| "first": "T A", | |
| "middle": [], | |
| "last": "Dijk", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Van", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1980, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Dijk T A Van. 1980. Macrostructures : An interdisciplinary study of global structures in discourse, interaction, and cognition. hillsdale, n.j.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "Attention is all you need", | |
| "authors": [ | |
| { | |
| "first": "Ashish", | |
| "middle": [], | |
| "last": "Vaswani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Noam", | |
| "middle": [], | |
| "last": "Shazeer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Niki", | |
| "middle": [], | |
| "last": "Parmar", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jakob", | |
| "middle": [], | |
| "last": "Uszkoreit", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Llion", | |
| "middle": [], | |
| "last": "Jones", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Aidan", | |
| "middle": [ | |
| "N" | |
| ], | |
| "last": "Gomez", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Illia", | |
| "middle": [], | |
| "last": "Kaiser", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Polosukhin", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of the 31th Annual Conference on Neural Information Processing Systems", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "5998--6008", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, L ukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Proceedings of the 31th Annual Conference on Neural Information Processing Systems, pages 5998-6008.", | |
| "links": null | |
| }, | |
| "BIBREF20": { | |
| "ref_id": "b20", | |
| "title": "Pointer networks", | |
| "authors": [ | |
| { | |
| "first": "Oriol", | |
| "middle": [], | |
| "last": "Vinyals", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Meire", | |
| "middle": [], | |
| "last": "Fortunato", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Navdeep", | |
| "middle": [], | |
| "last": "Jaitly", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proceedings of the 29th Annual Conference on Neural Information Processing Systems", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2692--2700", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Oriol Vinyals, Meire Fortunato, and Navdeep Jaitly. 2015. Pointer networks. In Proceedings of the 29th Annual Conference on Neural Information Processing Systems, pages 2692-2700.", | |
| "links": null | |
| }, | |
| "BIBREF21": { | |
| "ref_id": "b21", | |
| "title": "Multi-layer attention network based Chinese implicit discourse relation recognition", | |
| "authors": [ | |
| { | |
| "first": "Sheng", | |
| "middle": [], | |
| "last": "Xu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Tishuang", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Peifeng", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Qiaoming", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Journal of Chinese Information Processing", | |
| "volume": "33", | |
| "issue": "8", | |
| "pages": "12--19", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Sheng Xu, Tishuang Wang, Peifeng Li, and Qiaoming Zhu. 2019. Multi-layer attention network based Chinese implicit discourse relation recognition. Journal of Chinese Information Processing, 33(8):12-19.", | |
| "links": null | |
| }, | |
| "BIBREF22": { | |
| "ref_id": "b22", | |
| "title": "Constructing Chinese macro discourse tree via multiple views and word pair similarity", | |
| "authors": [ | |
| { | |
| "first": "Yi", | |
| "middle": [], | |
| "last": "Zhou", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xiaomin", | |
| "middle": [], | |
| "last": "Chu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Peifeng", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Qiaoming", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the 8th CCF International Conference on Natural Language Processing and Chinese Computing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "773--786", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Yi Zhou, Xiaomin Chu, Peifeng Li, and Qiaoming Zhu. 2019. Constructing Chinese macro discourse tree via multiple views and word pair similarity. In Proceedings of the 8th CCF International Conference on Natural Language Processing and Chinese Computing, pages 773-786.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF0": { | |
| "num": null, | |
| "text": "\u7b49\u3002 \u7bc7\u7ae0\u5206\u6790\u7684\u7814\u7a76\u5206\u6790\u53ef\u5206\u4e3a\u5fae\u89c2\u548c\u5b8f\u89c2\u4e24\u4e2a\u5c42\u9762\u3002\u5fae\u89c2\u5c42\u9762\u7814\u7a76\u7684\u662f\u53e5\u5b50\u548c\u53e5\u5b50\u3001\u53e5\u7fa4\u548c \u53e5\u7fa4\u4e4b\u95f4\u7684\u7ed3\u6784\u548c\u5173\u7cfb\uff1b\u5b8f\u89c2\u5c42\u9762\u7814\u7a76\u7684\u662f\u6bb5\u843d\u548c\u6bb5\u843d\u3001\u7ae0\u8282\u548c\u7ae0\u8282\u4e4b\u95f4\u7684\u7ed3\u6784\u548c\u5173\u7cfb\u3002\u5f53\u524d \u7bc7\u7ae0\u5206\u6790\u4e3b\u8981\u96c6\u4e2d\u5728\u5fae\u89c2\u5c42\u9762\uff0c\u800c\u5b8f\u89c2\u5c42\u9762\u7684\u7814\u7a76\u8f83\u5c11\u3002Chu et al. (2020)\u63d0\u51fa\u4e86\u4e00\u4e2a\u5b8f\u89c2\u7bc7\u7ae0 \u7ed3\u6784\u8868\u793a\u4f53\u7cfb\uff0c\u5176\u4e2d\uff0c\u4ee5\u6bb5\u843d\u4e3a\u57fa\u672c\u7bc7\u7ae0\u5355\u5143(Elementary Discourse Units,EDUs)\uff0c\u76f8\u90bb\u4e24 \u4e2a\u6bb5\u843d\u4ee5\u7bc7\u7ae0\u5173\u7cfb\u8fde\u63a5\u5728\u4e00\u8d77\uff0c\u5e76\u6784\u6210\u66f4\u5927\u7684\u7bc7\u7ae0\u5355\u5143(Discourse Units,DUs),\u8fd9\u4e9b\u7bc7\u7ae0\u5355\u5143 \u5c42\u5c42\u5411\u4e0a\uff0c\u6700\u7ec8\u5c06\u4e00\u7bc7\u6587\u7ae0\u6784\u6210\u4e00\u68f5\u5b8c\u6574\u7684\u7bc7\u7ae0\u7ed3\u6784\u6811\u3002 \u5b8f\u89c2\u6c49\u8bed\u7bc7\u7ae0\u6811\u5e93(Macro Chinese Discourse Treebank,MCDTB)(Jiang et al., 2018b)\u5bf9 \u5b8f\u89c2\u7bc7\u7ae0\u7ed3\u6784\u8fdb\u884c\u4e86\u6807\u6ce8\u3002\u672c\u6587\u4ee5MCDTB\u4e2d\u7684\u4e00\u7bc7\u6587\u7ae0(chtb 0282)\u6765\u8bf4\u660e\u5b8f\u89c2\u7bc7\u7ae0\u7ed3\u6784\uff0c \u5982\u4f8b1\u6240\u793a\u3002\u5176\u4e2d\uff0cp 1 \u4ecb\u7ecd\u4e86\u63a8\u884c\u516c\u52a1\u5458\u5236\u5ea6\u4ea4\u6d41\u4f1a\u7684\u60c5\u51b5, p 2 \u8865\u5145\u4e86\u4f1a\u8bae\u65f6\u95f4\u4ee5\u53ca\u53c2\u4f1a\u4eba \u5458\uff1bp 3 \u8bb2\u8ff0\u4e86\u674e\u9e4f\u603b\u7406\u80af\u5b9a\u4e86\u63a8\u884c\u516c\u52a1\u5458\u5236\u5ea6\u7684\u6210\u6548, p 4 \u8bb2\u8ff0\u4e86\u674e\u9e4f\u603b\u7406\u63d0\u51fa\u63a8\u884c\u516c\u52a1\u5458\u5236\u5ea6 \u8981\u4f9d\u6cd5\u529e\u4e8b\uff1bp 5 \u8865\u5145\u5176\u4ed6\u53c2\u4f1a\u4eba\u5458\u3002p 2 \u8865\u5145\u4e86p 1 \u63cf\u8ff0\u7684\u4ea4\u6d41\u4f1a\u7684\u76f8\u5173\u4fe1\u606f\uff0c\u56e0\u6b64p 1 \u4e0ep 2 \u6784\u6210\u8865 \u5145\u5173\u7cfb\uff0cp 2 \u548cp 4 \u5206\u522b\u9610\u8ff0\u4e86\u4ea4\u6d41\u4f1a\u7684\u5185\u5bb9\uff0c\u56e0\u6b64\u6784\u6210\u4e86\u5e76\u5217\u5173\u7cfb\uff0c\u5176\u5f62\u6210\u7684\u7bc7\u7ae0\u5355\u5143\u5bf9\u4e0a\u6587 (p 1 \u548cp 2 \u6784\u6210\u7684\u7bc7\u7ae0\u5355\u5143)\u8fdb\u884c\u89e3\u8bf4\uff0c\u5f62\u6210\u89e3\u8bf4\u5173\u7cfb\uff0cp 5 \u662f\u5bf9\u5168\u6587\u7684\u8865\u5145\uff0c\u5373\u5bf9p 1 \u5230p 4 \u7684\u4fe1\u606f \u8fdb\u884c\u8865\u5145\u3002", | |
| "content": "<table><tr><td>p 1 :\u56fd\u52a1\u9662\u603b\u7406\u674e\u9e4f\u4eca\u5929\u5728\u4e2d\u5357\u6d77\u7d2b\u5149\u9601\u4f1a\u89c1\u4e2d\u56fd\u63a8\u884c\u516c\u52a1\u5458\u5236\u5ea6\u7ecf\u9a8c\u4ea4\u6d41\u4f1a</td></tr><tr><td>\u5168\u4f53\u4ee3\u8868\u65f6\u6307\u51fa\uff0c\u63a8\u884c\u516c\u52a1\u5458\u5236\u5ea6\u662f\u4e2d\u56fd\u653f\u6cbb\u4f53\u5236\u6539\u9769\u7684\u4e00\u9879\u91cd\u8981\u5185\u5bb9\uff0c\u662f\u5e72</td></tr><tr><td>\u90e8\u4eba\u4e8b\u5236\u5ea6\u7684\u91cd\u5927\u6539\u9769\uff0c\u662f\u5efa\u7acb\u793e\u4f1a\u4e3b\u4e49\u5e02\u573a\u7ecf\u6d4e\u4f53\u5236\u7684\u5ba2\u89c2\u9700\u8981\uff0c\u8981\u6709\u9886\u5bfc\u3001</td></tr><tr><td>\u6709\u6b65\u9aa4\u5730\u52a0\u5feb\u63a8\u884c\u6b65\u4f10\u3002</td></tr><tr><td>p2:\u8fd9\u6b21\u63a8\u884c\u516c\u52a1\u5458\u5236\u5ea6\u7ecf\u9a8c\u4ea4\u6d41\u4f1a\u662f\u6628\u5929\u5f00\u59cb\u53ec\u5f00\u7684\uff0c\u5404\u7701\u3001\u81ea\u6cbb\u533a\u3001\u76f4\u8f96</td></tr><tr><td>\u5e02\u4eba\u4e8b\u5385\u5c40\u957f\u3001\u56fd\u52a1\u9662\u5404\u90e8\u59d4\u3001\u76f4\u5c5e\u673a\u6784\u4eba\u4e8b\u90e8\u95e8\u7684\u8d1f\u8d23\u4eba\u5171\u4e00\u767e\u4e8c\u5341\u591a\u4eba\u51fa</td></tr><tr><td>\u5e2d\u4e86\u4f1a\u8bae\u3002</td></tr><tr><td>p3:\u674e\u9e4f\u80af\u5b9a\u4e86\u4e00\u5e74\u6765\u56fd\u5bb6\u516c\u52a1\u5458\u5236\u5ea6\u63a8\u884c\u5de5\u4f5c\u53d6\u5f97\u7684\u6210\u6548\u3002\u4ed6\u8bf4\uff0c\u6211\u4eec\u8981\u8ba4</td></tr><tr><td>\u771f\u603b\u7ed3\u548c\u63a8\u5e7f\u8fd9\u4e9b\u597d\u7684\u7ecf\u9a8c\uff0c\u5efa\u7acb\u8d77\u6fc0\u52b1\u7ade\u4e89\u548c\u52e4\u653f\u5ec9\u653f\u673a\u5236\uff0c\u5efa\u7acb\u4e00\u652f\u4ee5\u4e3a</td></tr><tr><td>\u4eba\u6c11\u670d\u52a1\u4e3a\u5b97\u65e8\u3001\u5bc6\u5207\u8054\u7cfb\u7fa4\u4f17\u3001\u7cbe\u5e72\u9ad8\u6548\u3001\u5ec9\u6d01\u5949\u516c\u3001\u5fe0\u4e8e\u804c\u5b88\u7684\u56fd\u5bb6\u516c\u52a1</td></tr><tr><td>\u5458\u961f\u4f0d\uff0c\u589e\u5f3a\u653f\u5e9c\u673a\u5173\u7684\u751f\u673a\u548c\u6d3b\u529b\u3002</td></tr><tr><td>p4:\u674e\u9e4f\u63d0\u51fa\uff0c\u63a8\u884c\u516c\u52a1\u5458\u5236\u5ea6\uff0c\u8981\u6309\u7167\u300a\u56fd\u5bb6\u516c\u52a1\u5458\u6682\u884c\u6761\u4f8b\u300b\u4f9d\u6cd5\u529e\u4e8b\uff0c</td></tr><tr><td>\u4e0d\u80fd\u6709\u968f\u610f\u6027\u3002\u8981\u628a\u8fd9\u9879\u5de5\u4f5c\u4f5c\u4e3a\u653f\u6cbb\u4f53\u5236\u6539\u9769\u7684\u4e00\u4ef6\u5927\u4e8b\u6765\u6293\uff0c\u7ed3\u5408\u6539\u9769\u3001</td></tr><tr><td>\u7cbe\u7b80\u673a\u6784\u6765\u63a8\u884c\u516c\u52a1\u5458\u5236\u5ea6\uff1b\u8981\u5f62\u6210\u516c\u52a1\u5458\u7684\u65b0\u9648\u4ee3\u8c22\u673a\u5236\uff0c\u4f7f\u9752\u5e74\u4eba\u624d\u4e0d\u65ad</td></tr><tr><td>\u5730\u8fdb\u5165\u5230\u516c\u52a1\u5458\u961f\u4f0d\u5f53\u4e2d\u3002</td></tr><tr><td>p5:\u56fd\u52a1\u59d4\u5458\u674e\u8d35\u9c9c\u3001\u7f57\u5e72\u53c2\u52a0\u4e86\u4f1a\u89c1\u3002</td></tr></table>", | |
| "html": null, | |
| "type_str": "table" | |
| }, | |
| "TABREF1": { | |
| "num": null, | |
| "text": "\u843d\u4ea4\u4e92\u5c42(Paragraph Interactive Layer,PIL),\u7528\u6765\u6355\u83b7\u76f8\u90bb\u4e24\u4e2a\u6bb5\u843d\u7684\u8bed\u4e49\u8054\u7cfb\uff0c\u5373\u5c40\u90e8\u4fe1 \u606f\uff1b3)\u6307\u9488\u7f51\u7edc(Pointer Network)\uff0c\u6307\u9488\u7f51\u7edc\u7684\u7f16\u7801\u5c42\u7528\u6765\u6355\u83b7\u6574\u4e2a\u7bc7\u7ae0\u7684\u8bed\u4e49\u8868\u793a\uff0c\u5373\u5168 \u5c40\u4fe1\u606f\uff0c\u89e3\u7801\u5c42\u878d\u5408\u5c40\u90e8\u548c\u5168\u5c40\u4fe1\u606f,\u7528\u6765\u8bc6\u522b\u7bc7\u7ae0\u7ed3\u6784\u5e76\u81ea\u9876\u5411\u4e0b\u7684\u6784\u5efa\u7bc7\u7ae0\u7ed3\u6784\u6811\u3002", | |
| "content": "<table><tr><td/><td/><td>\u8865\u5145</td><td/><td/></tr><tr><td/><td>\u89e3\u8bf4</td><td/><td/><td/></tr><tr><td>\u8865\u5145</td><td/><td>\u5e76\u5217</td><td/><td/></tr><tr><td>p 1</td><td>p 2</td><td>p 3</td><td>p 4</td><td>p 5</td></tr><tr><td colspan=\"5\">\u56fe 1. \u5b8f\u89c2\u7bc7\u7ae0\u7ed3\u6784\u6811(chtb 0282)</td></tr><tr><td colspan=\"5\">\u5b8f\u89c2\u8bed\u4e49\u4fe1\u606f(\u5373\u5168\u5c40\u4fe1\u606f)\u5f80\u5f80\u80fd\u4f53\u73b0\u7bc7\u7ae0\u7684\u5c55\u5f00\u7ed3\u6784\uff0c\u53ef\u7528\u4e8e\u68c0\u9a8c\u4e00\u4e2a\u7bc7\u7ae0\u662f\u5426\u8fde\u8d2f\u3002\u56e0</td></tr><tr><td colspan=\"5\">\u6b64\u672c\u6587\u8ba4\u4e3a\u5728\u8003\u8651\u5c40\u90e8\u4fe1\u606f\u7684\u540c\u65f6\uff0c\u5168\u5c40\u4fe1\u606f\u4e5f\u5e94\u8be5\u88ab\u8003\u8651\u7528\u6765\u8f85\u52a9\u7bc7\u7ae0\u7ed3\u6784\u7684\u8bc6\u522b\u3002</td></tr><tr><td colspan=\"5\">\u57fa\u4e8e\u4ee5\u5f80\u7684\u7814\u7a76\u90fd\u53ea\u8003\u8651\u5c40\u90e8\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\uff0c\u4e14\u53d7\u5230\u5b8f\u89c2\u7bc7\u7ae0\u7ed3\u6784\u7406\u8bba\u7684\u542f\u53d1\uff0c\u672c\u6587\u63d0\u51fa</td></tr><tr><td colspan=\"5\">\u4e00\u79cd\u878d\u5408\u5168\u5c40\u548c\u5c40\u90e8\u4fe1\u606f\u7684\u6307\u9488\u7f51\u7edc\u6a21\u578b\uff0c\u7528\u4e8e\u81ea\u9876\u5411\u4e0b\u7684\u8bc6\u522b\u7bc7\u7ae0\u7ed3\u6784\uff0c\u5e76\u6784\u5efa\u7bc7\u7ae0\u7ed3\u6784</td></tr><tr><td colspan=\"5\">\u6811\u3002\u5728\u8be5\u6a21\u578b\u4e2d\uff0c\u672c\u6587\u91c7\u7528\u4ea4\u4e92\u6ce8\u610f\u529b\u673a\u5236\u6355\u83b7\u76f8\u90bb\u4e24\u4e2a\u6bb5\u843d\u4e4b\u95f4\u7684\u8bed\u4e49\u8054\u7cfb\uff0c\u5373\u5c40\u90e8\u4fe1\u606f\uff1b</td></tr><tr><td colspan=\"5\">\u6307\u9488\u7f51\u7edc\u7684\u7f16\u7801\u5c42\u7528\u6765\u6355\u83b7\u6574\u4e2a\u7bc7\u7ae0\u7684\u8bed\u4e49\uff0c\u5373\u5168\u5c40\u4fe1\u606f\uff1b\u800c\u6307\u9488\u7f51\u7edc\u7684\u89e3\u7801\u5c42\u7528\u6765\u878d\u5408\u5168\u5c40</td></tr><tr><td colspan=\"5\">\u548c\u5c40\u90e8\u4fe1\u606f\uff0c\u4e3a\u4e24\u4e2a\u6bb5\u843d\u4e4b\u95f4\u7684\u8bed\u4e49\u5206\u914d\u4e00\u4e2a\u6982\u7387\uff0c\u6982\u7387\u8d8a\u5927\uff0c\u8868\u660e\u8fd9\u4e24\u4e2a\u6bb5\u843d\u4e4b\u95f4\u7684\u8bed\u4e49\u8054</td></tr><tr><td colspan=\"5\">\u7cfb\u8d8a\u5f31\uff0c\u5219\u9700\u8981\u8fdb\u884c\u7bc7\u7ae0\u5355\u5143\u7684\u5207\u5206\u3002\u5bf9\u5207\u5206\u5f62\u6210\u7684\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\uff0c\u6839\u636e\u6df1\u5ea6\u4f18\u5148\u539f\u5219\uff0c\u9012\u5f52</td></tr><tr><td colspan=\"5\">\u5730\u8fdb\u884c\u5207\u5206\uff0c\u4ece\u800c\u81ea\u9876\u5411\u4e0b\u7684\u6784\u5efa\u5b8c\u6574\u7684\u7bc7\u7ae0\u7ed3\u6784\u6811\u3002\u5728MCDTB\u4e0a\u7684\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u672c\u6587\u7684</td></tr><tr><td>\u6a21\u578b\u4f18\u4e8e\u76ee\u524d\u6027\u80fd\u6700\u597d\u7684\u6a21\u578b\u3002</td><td/><td/><td/><td/></tr><tr><td>2 \u76f8 \u76f8 \u76f8\u5173 \u5173 \u5173\u5de5 \u5de5 \u5de5\u4f5c \u4f5c \u4f5c</td><td/><td/><td/><td/></tr><tr><td colspan=\"5\">\u5728\u5df2\u6709\u7684\u7814\u7a76\u5de5\u4f5c\u4e2d\uff0c\u65e0\u8bba\u4e2d\u6587\u8fd8\u662f\u82f1\u6587\u90fd\u66f4\u6ce8\u91cd\u5fae\u89c2\u7bc7\u7ae0\u7ed3\u6784\u7684\u5206\u6790\uff0c\u800c\u5bf9\u4e8e\u5b8f\u89c2</td></tr><tr><td colspan=\"5\">\u7bc7\u7ae0\u7ed3\u6784\u7684\u5206\u6790\u8fd8\u5904\u4e8e\u8d77\u6b65\u9636\u6bb5\u3002\u6d89\u53ca\u5230\u5b8f\u89c2\u7bc7\u7ae0\u7ed3\u6784\u7684\u8bed\u6599\u5e93\u4e3b\u8981\u6709\u82f1\u6587\u4fee\u8f9e\u7ed3\u6784\u7bc7\u7ae0</td></tr><tr><td colspan=\"5\">\u6811\u5e93(RST Discourse Treebank,RST-DT)(Carlson et al., 2007)\u548c\u4e2d\u6587\u7684\u5b8f\u89c2\u6c49\u8bed\u7bc7\u7ae0\u6811\u5e93</td></tr><tr><td colspan=\"5\">(MCDTB)(Jiang et al., 2018b)\u3002\u73b0\u5c06\u4e24\u4e2a\u8bed\u6599\u548c\u76f8\u5173\u6a21\u578b\u4ecb\u7ecd\u5982\u4e0b\uff1a</td></tr><tr><td colspan=\"5\">\u4fee\u8f9e\u7ed3\u6784\u7bc7\u7ae0\u6811\u5e93(RST-DT)\u4ee5\u4fee\u8f9e\u7ed3\u6784\u7406\u8bba(RST)\u4e3a\u7406\u8bba\u4f9d\u636e\uff0c\u6807\u6ce8\u4e86385\u7bc7\u300a\u534e</td></tr><tr><td colspan=\"5\">\u5c14\u8857\u65e5\u62a5\u300b\u6587\u7ae0\u3002\u5728\u8be5\u8bed\u6599\u5e93\u7684\u7814\u7a76\u4e2d\uff0cHernault et al. (2010)\u63d0\u51fa\u4e86\u57fa\u4e8eSVM\u7684\u7bc7\u7ae0\u5206\u6790</td></tr><tr><td colspan=\"5\">\u5668HILDA,\u8be5\u6a21\u578b\u4ee5\u8d2a\u5a6a\u7684\u65b9\u5f0f\u81ea\u5e95\u5411\u4e0a\u6784\u5efa\u7bc7\u7ae0\u7ed3\u6784\u6811\uff1bJoty et al. (2013)\u7b49\u5229\u7528\u52a8\u6001CRF\u6a21</td></tr><tr><td colspan=\"5\">\u578b\u5206\u522b\u6784\u5efa\u4e86\u53e5\u5b50\u7ea7\u522b\u548c\u7bc7\u7ae0\u7ea7\u522b\u7684\u5206\u6790\u5668\uff1bJi and Eisenstein (2014)\u53c2\u8003\u6df1\u5ea6\u5b66\u4e60\u7684\u505a\u6cd5\uff0c</td></tr><tr><td colspan=\"5\">\u91c7\u7528\u7ebf\u6027\u53d8\u6362\u5c06\u8868\u9762\u7279\u5f81\u8f6c\u6362\u6210\u9690\u7a7a\u95f4\u901a\u8fc7\u79fb\u8fdb\u89c4\u7ea6\u8fdb\u884c\u7bc7\u7ae0\u89e3\u6790\uff1bLin et al. (2019)\u91c7\u7528\u6307\u9488</td></tr><tr><td colspan=\"5\">\u7f51\u7edc\uff0c\u6784\u5efa\u4e86\u4e00\u4e2a\u53e5\u5b50\u7ea7\u7684\u7bc7\u7ae0\u89e3\u6790\u5668\uff0c\u4f46\u4e0a\u8ff0\u7814\u7a76\u90fd\u662f\u5728\u5fae\u89c2\u5c42\u9762\u3002\u5728\u5b8f\u89c2\u5c42\u9762\uff0cSporleder</td></tr><tr><td colspan=\"5\">and Lascarides (2004)\u5bf9RST-DT\u4fee\u6b63\u548c\u88c1\u526a\u540e\u91c7\u7528\u6700\u5927\u71b5\u6a21\u578b\u8fdb\u884c\u4e86\u5b8f\u89c2\u7bc7\u7ae0\u7ed3\u6784\u8bc6\u522b\u3002</td></tr><tr><td colspan=\"5\">\u5b8f\u89c2\u6c49\u8bed\u7bc7\u7ae0\u6811\u5e93(MCDTB)\u9075\u5faaRST\u4fee\u8f9e\u7ed3\u6784\u7406\u8bba\uff0c\u5bf9720\u7bc7\u6587\u7ae0\u8fdb\u884c\u4e86\u5b8f\u89c2\u7bc7\u7ae0\u4fe1\u606f</td></tr><tr><td colspan=\"5\">\u7684\u6807\u6ce8\uff0c\u5305\u62ec\u7bc7\u7ae0\u7ed3\u6784\u3001\u4e3b\u6b21\u548c\u8bed\u4e49\u5173\u7cfb\u7b49\u3002\u5728MCDTB\u4e0a\u8fdb\u884c\u7bc7\u7ae0\u7ed3\u6784\u8bc6\u522b\uff0c\u6784\u5efa\u5b8c\u6574\u7bc7\u7ae0</td></tr><tr><td colspan=\"5\">\u7ed3\u6784\u6811\u7684\u7814\u7a76\u4e0d\u591a\u3002Jiang et al. (2018a)\u91c7\u7528\u5e8f\u5217\u6807\u6ce8\u7684\u601d\u60f3\uff0c\u63d0\u51fa\u4e00\u4e2a\u57fa\u4e8e\u6761\u4ef6\u968f\u673a\u573a\u7684\u6a21</td></tr><tr><td colspan=\"5\">\u578b(LD-CM)\u3002\u8be5\u6a21\u578b\u5bf9\u7ed3\u6784\u548c\u4e3b\u6b21\u8fdb\u884c\u8054\u5408\u5b66\u4e60\uff0c\u4ece\u800c\u81ea\u5e95\u5411\u4e0a\u7684\u6784\u5efa\u7bc7\u7ae0\u7ed3\u6784\u6811\uff1bZhou</td></tr><tr><td colspan=\"5\">et al. (2019)\u63d0\u51fa\u4e86\u4e00\u4e2a\u57fa\u4e8e\u795e\u7ecf\u7f51\u7edc\u7684\u6a21\u578b(MVM)\u3002\u8be5\u6a21\u578b\u4ece\u591a\u4e2a\u89d2\u5ea6\u5339\u914d\u4e24\u4e2a\u7bc7\u7ae0\u5355</td></tr><tr><td colspan=\"5\">\u5143\u4e4b\u95f4\u7684\u8bed\u4e49\uff0c\u4ece\u800c\u8bc6\u522b\u7bc7\u7ae0\u7ed3\u6784\uff0c\u5e76\u91c7\u7528\u79fb\u8fdb\u89c4\u7ea6\u7684\u65b9\u6cd5\u6784\u5efa\u7bc7\u7ae0\u7ed3\u6784\u6811\u3002\u7136\u800cLD-CM\u662f</td></tr><tr><td colspan=\"5\">\u57fa\u4e8e\u4f20\u7edf\u673a\u5668\u5b66\u4e60\u7684\u65b9\u6cd5\uff0c\u7528\u5230\u4e86\u8f83\u591a\u7684\u624b\u5de5\u7279\u5f81\uff0c\u8003\u8651\u76f8\u90bb\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\u7684\u8bed\u4e49\u8054\u7cfb\uff1b\u540c</td></tr><tr><td colspan=\"5\">\u6837MVM\u4e5f\u53ea\u8003\u8651\u76f8\u90bb\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\u7684\u8bed\u4e49\u8054\u7cfb\u3002\u8fd9\u4e24\u79cd\u65b9\u6cd5\u90fd\u53ea\u8003\u8651\u4e86\u5c40\u90e8\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\uff0c</td></tr><tr><td colspan=\"2\">\u6ca1\u6709\u6709\u6548\u8fd0\u7528\u5168\u5c40\u4fe1\u606f\u8f85\u52a9\u7bc7\u7ae0\u7ed3\u6784\u7684\u8bc6\u522b\u3002</td><td/><td/><td/></tr><tr><td>3 PNGL\u6a21 \u6a21 \u6a21\u578b \u578b \u578b</td><td colspan=\"4\">\u7684\u7bc7\u7ae0\u7ed3\u6784\u8bc6\u522b\u4efb\u52a1\u4e2d,Lin et al. (2019)\u63d0\u5230\u6bcf\u6b21\u8003\u8651\u76f8\u90bb</td></tr><tr><td colspan=\"5\">\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\u5bb9\u6613\u53d7\u5230\u5c40\u90e8\u4fe1\u606f\u7684\u5f71\u54cd\uff0c\u800c\u9519\u8bef\u7684\u76f8\u90bb\u7bc7\u7ae0\u7ed3\u6784\u5224\u65ad\u4f1a\u5c06\u9519\u8bef\u7684\u4fe1\u606f\u4f20\u64ad\u5230\u4e0a</td></tr><tr><td colspan=\"5\">\u5c42\uff0c\u4ece\u800c\u5f71\u54cd\u4e0a\u5c42\u7ed3\u6784\u7684\u8bc6\u522b\u3002\u800cVan (1980)\u7684\u5b8f\u89c2\u7bc7\u7ae0\u7ed3\u6784\u7406\u8bba\u4e5f\u6307\u51fa\u5b8f\u89c2\u7ed3\u6784\u662f\u66f4\u9ad8\u5c42\u6b21</td></tr><tr><td colspan=\"5\">\u7684\u7ed3\u6784\uff0c\u8868\u73b0\u4e3a\u7bc7\u7ae0\u6574\u4f53\u7684\u8bed\u4e49\u8fde\u8d2f\uff0c\u6bcf\u4e00\u5c42\u7684\u5b8f\u89c2\u7ed3\u6784\u90fd\u662f\u7531\u4e0b\u5c42\u7ed3\u6784\u652f\u6491\u8d77\u6765\u7684\u3002\u7bc7\u7ae0\u7684</td></tr></table>", | |
| "html": null, | |
| "type_str": "table" | |
| }, | |
| "TABREF3": { | |
| "num": null, | |
| "text": ", x 2 , \u2022 \u2022 \u2022 , x n }\uff0c\u9996\u5148\u7ecf\u8fc7\u7f16\u7801\u5c42\u5f97\u5230\u8f93\u51faY = {y 1 , y 2 , \u2022 \u2022 \u2022 , y n }\u3002\u5728\u89e3\u7801\u5c42\u7684\u6bcf\u4e00\u4e2a\u65f6 \u95f4\u6b65t\uff0c\u8f93\u51fa\u7684\u72b6\u6001d t \u4f1a\u548c\u5e8f\u5217Y\u8fdb\u884c\u4ea4\u4e92\uff0c\u8ba1\u7b97\u6ce8\u610f\u529b\uff0c\u7136\u540e\u901a\u8fc7softmax\u5c42\u83b7\u5f97\u5173\u4e8e\u8f93\u5165\u5e8f\u5217 \u7684\u6982\u7387\u5206\u5e03\u3002\u56e0\u6b64\uff0c\u5728PNGL\u6a21\u578b\u4e2d\uff0c\u672c\u6587\u8fd0\u7528\u6307\u9488\u7f51\u7edc\u83b7\u5f97\u5173\u4e8e\u6587\u7ae0\u76f8\u90bb\u4e24\u4e2a\u6bb5\u843d\u4e4b\u95f4\u7684\u8bed \u4e49\u8054\u7cfb(H)\u7684\u6982\u7387\u5206\u5e03\uff0c\u8fdb\u800c\u786e\u5b9a\u6587\u7ae0\u7684\u5207\u5206\u4f4d\u7f6e\u3002 3.3.1 \u7f16 \u7f16 \u7f16\u7801 \u7801 \u7801\u5c42 \u5c42 \u5c42Chung et al. (2014)\u7684\u7814\u7a76\u8868\u660e\uff0cGRU\u548cLSTM\u5728\u5f88\u591a\u4efb\u52a1\u4e0a\u7684\u6027\u80fd\u4e0d \u5206\u4f2f\u4ef2\uff0c\u4f46\u662fGRU\u62e5\u6709\u66f4\u5c11\u7684\u53c2\u6570\uff0c\u5bb9\u6613\u6536\u655b\uff0c\u56e0\u6b64\u5728\u7f16\u7801\u5c42\u672c\u6587\u4f7f\u7528\u4e24\u5c42\u7684\u53cc\u5411GRU\u8fdb \u884c\u7f16\u7801\u3002\u4ee5chtb 0282\u4e3a\u4f8b\uff0c\u672c\u6587\u5c06\u6587\u7ae0P = {p 1 , p 2 , p 3 , p 4 , p 5 }\u901a\u8fc7\u6bb5\u843d\u7f16\u7801\u5c42\uff0c\u5f97\u5230\u6bb5\u843d\u7f16 \u7801R = {r 1 , r 2 , r 3 , r 4 , r 5 },\u7136\u540e\u91c7\u7528\u5e73\u5747\u6c60\u5316\u64cd\u4f5c\u8f93\u5165\u5230\u53cc\u5411GRU\u4e2d\u3002\u53cc\u5411GRU\u7684\u8f93\u51fa\u4e3aE = {e 1 , e 2 , e 3 , e 4 , e 5 },\u5176\u4e2de", | |
| "content": "<table><tr><td>3.3 \u6307 \u6307 \u6307\u9488 \u9488 \u9488\u7f51 \u7f51 \u7f51\u7edc \u7edc \u7edc \u5e8f\u5217\u5230\u5e8f\u5217\u7684\u6a21\u578b(Sutskever et al., 2014)\u63d0\u4f9b\u4e86\u8f93\u5165\u5e8f\u5217\u548c\u8f93\u51fa\u5e8f\u5217\u957f\u5ea6\u53ef\u4ee5\u4e0d\u540c\u7684 \u7075\u6d3b\u6027,\u4f46\u662f\u7531\u4e8e\u8be5\u6a21\u578b\u4ecd\u7136\u9700\u8981\u56fa\u5b9a\u8f93\u51fa\u8bcd\u6c47\u8868\u7684\u5927\u5c0f\uff0c\u800c\u8f93\u51fa\u8bcd\u8868\u7684\u5927\u5c0f\u53d6\u51b3\u4e8e\u8f93\u5165\u5e8f \u5217\u7684\u957f\u5ea6\uff0c\u4ece\u800c\u9650\u5236\u4e86\u9700\u8981\u6307\u5411\u8f93\u5165\u5e8f\u5217\u67d0\u4e2a\u4f4d\u7f6e\u7684\u95ee\u9898\u7684\u9002\u7528\u6027\u3002\u800c\u6307\u9488\u7f51\u7edc(Vinyals et al., 2015)\u901a\u8fc7\u4f7f\u7528\u6ce8\u610f\u529b\u4f5c\u4e3a\u4e00\u4e2a\u6307\u5411\u673a\u5236\u89e3\u51b3\u4e86\u8fd9\u4e2a\u95ee\u9898\u3002\u5177\u4f53\u8bf4\u6765\uff0c\u5bf9\u4e8e\u8f93\u5165\u5e8f \u5217X = {x 1 i = [e f i ; e b i ]\u3002e f i \u548ce b i</td></tr></table>", | |
| "html": null, | |
| "type_str": "table" | |
| }, | |
| "TABREF4": { | |
| "num": null, | |
| "text": "\u5168 \u5168 \u5168\u5c40 \u5c40 \u5c40\u548c \u548c \u548c\u5c40 \u5c40 \u5c40\u90e8 \u90e8 \u90e8\u4fe1 \u4fe1 \u4fe1\u606f \u606f \u606f\u7684 \u7684 \u7684\u5f71 \u5f71 \u5f71\u54cd \u54cd \u54cd", | |
| "content": "<table><tr><td colspan=\"5\">\u672c\u6587\u7684\u6a21\u578bPNGL\u901a\u8fc7\u6539\u8fdb\u6bb5\u843d\u7684\u8bed\u4e49\u7f16\u7801\uff0c\u5728\u6307\u9488\u7f51\u7edc\u7f16\u7801\u5c42\u5b66\u4e60\u5230\u66f4\u597d\u7684\u5168\u5c40\u4fe1\u606f\u7684</td></tr><tr><td colspan=\"5\">\u540c\u65f6\uff0c\u53c8\u8003\u8651\u76f8\u90bb\u4e24\u4e2a\u6bb5\u843d\u4e4b\u95f4\u8bed\u4e49\u8054\u7cfb\u7684\u7d27\u5bc6\u7a0b\u5ea6\uff0c\u4ece\u800c\u5728\u6027\u80fd\u4e0a\u6709\u6240\u63d0\u5347\uff0c\u8fd9\u8bf4\u660e\u7efc\u5408\u8003</td></tr><tr><td colspan=\"5\">\u8651\u5168\u5c40\u548c\u5c40\u90e8\u4fe1\u606f\u5bf9\u4e8e\u8bc6\u522b\u7bc7\u7ae0\u7ed3\u6784\u5e76\u6784\u5efa\u7bc7\u7ae0\u7ed3\u6784\u6811\u975e\u5e38\u6709\u6548\u3002</td></tr><tr><td colspan=\"2\">5 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u5206 \u5206 \u5206\u6790 \u6790 \u6790</td><td/><td/></tr><tr><td colspan=\"5\">\u88681\u6240\u793a\u3002 5.1 \u4ee5\u5f80\u7684\u7814\u7a76\u8868\u660e(Lin et al., 2019)\uff0c\u91c7\u7528\u57fa\u4e8e\u8f6c\u79fb\u7684\u65b9\u6cd5\u8fdb\u884c\u7ed3\u6784\u8bc6\u522b\uff0c\u5f80\u5f80\u5bf9\u4e8e\u5e95\u5c42\u7684\u8bc6 \u6bb5\u843d 2 3 4 5 6 7 8 9 10 11 12 > 12 \u6570\u91cf 29 122 159 144 91 58 37 33 15 13 14 \u522b\u80fd\u529b\u6bd4\u8f83\u597d\uff0c\u800c\u4e0a\u5c42\u7684\u8bc6\u522b\u80fd\u529b\u6bd4\u8f83\u5dee\u3002\u4e3b\u8981\u539f\u56e0\u662f\u6bcf\u4e00\u6b65\u7684\u8bc6\u522b\u90fd\u53ea\u8003\u8651\u5c40\u90e8\u4fe1\u606f\uff0c\u8fd9\u4f1a 15 \u5c06\u9519\u8bef\u4f20\u64ad\u5230\u540e\u7eed\u6b65\u9aa4\uff0c\u5bfc\u81f4\u4e0a\u5c42\u7684\u7ed3\u6784\u7684\u8bc6\u522b\u80fd\u529b\u8f83\u5dee\u3002</td></tr><tr><td colspan=\"5\">\u88681.\u6bb5\u843d\u5206\u5e03 \u4e3a\u4e86\u7814\u7a76\u5c40\u90e8\u4fe1\u606f\u548c\u5168\u5c40\u4fe1\u606f\u5206\u522b\u5bf9\u5e95\u5c42\u548c\u9876\u5c42\u7ed3\u6784\u8bc6\u522b\u7684\u5f71\u54cd\uff0c\u672c\u6587\u5728PNGL\u6a21\u578b\u7684\u57fa</td></tr><tr><td colspan=\"5\">\u672c\u6587\u4f7f\u7528Jiang et al. (2018a)\u9075\u5faa\u6bb5\u843d\u5206\u5e03\u5212\u5206\u597d\u7684\u6570\u636e\u96c6\u8fdb\u884c\u8bd5\u9a8c\uff0c\u5176\u4e2d\u8bad\u7ec3\u96c6576\u7bc7\uff0c \u7840\u4e4b\u4e0a\u53bb\u6389\u6bb5\u843d\u4ea4\u4e92\u5c42\uff0c\u5373\u53ea\u8003\u8651\u5168\u5c40\u4fe1\u606f\uff0c\u5f97\u5230\u6a21\u578bPNGL(-local)\u3002\u672c\u6587\u5bf9\u53ea\u8003\u8651\u5c40\u90e8\u4fe1\u606f</td></tr><tr><td colspan=\"5\">\u6d4b\u8bd5\u96c6144\u7bc7\u3002\u4e3a\u4e86\u4e0eZhou et al. (2019)\u7684\u5b9e\u9a8c\u8bbe\u7f6e\u4e00\u81f4\uff0c\u672c\u6587\u5c06\u6240\u6709\u7684\u975e\u4e8c\u53c9\u6811\u90fd\u8f6c\u6362\u4e3a\u53f3 \u6700\u597d\u7684\u6a21\u578bMVM\u4ee5\u53ca\u53ea\u8003\u8651\u5168\u5c40\u4fe1\u606f\u6700\u597d\u7684\u6a21\u578bPNGL(-local)\u5728\u6700\u5e95\u4e0b\u4e24\u5c42\u5185\u90e8\u8282\u70b9\u6b63\u786e\u7387\u548c</td></tr><tr><td colspan=\"5\">\u4e8c\u53c9\u6811\u3002\u53e6\u5916\uff0c\u672c\u6587\u9075\u5faaMorey et al. (2017)\u5bf9RST-DT\u4e0a\u7bc7\u7ae0\u7ed3\u6784\u5206\u6790\u6a21\u578b\u7684\u8bc4\u4ef7\u6807\u51c6\uff0c\u540c\u6837 \u6700\u9876\u4e0a\u4e09\u5c42\u5185\u90e8\u8282\u70b9\u6b63\u786e\u7387 0 \u8fdb\u884c\u4e86\u7edf\u8ba1\u5206\u6790\uff0c\u5982\u88683\u6240\u793a\u3002</td></tr><tr><td colspan=\"5\">\u91c7\u7528\u5185\u90e8\u8282\u70b9\u6b63\u786e\u7387(\u7b49\u4ef7\u4e8emicro-F1)\u6765\u8861\u91cf\u6a21\u578b\u6027\u80fd\u3002\u672c\u6587\u5c06\u8bcd\u5411\u91cf\u7ef4\u5ea6\u8bbe\u7f6e\u4e3a300\uff0c\u91c7 \u6a21\u578b \u6700\u5e95\u4e0b\u4e24\u5c42\u8282\u90e8\u7ed3\u70b9\u6b63\u786e\u7387% \u6700\u9876\u4e0a\u4e09\u5c42\u5185\u90e8\u8282\u70b9\u6b63\u786e\u7387% \u7528Word2Vec(Mikolov et al., 2013)\u8fdb\u884c\u9884\u8bad\u7ec3\u3002\u5728\u6bb5\u843d\u7f16\u7801\u5c42\u548c\u6bb5\u843d\u4ea4\u4e92\u5c42\u8f6c\u6362\u77e9\u9635\u6620\u5c04\u7684\u7ef4 MVM 46.95 60.28 \u5ea6d m \u548cd i \u90fd\u88ab\u8bbe\u7f6e\u4e3a512;\u6bb5\u843d\u7f16\u7801\u5c42\u591a\u5934\u6ce8\u610f\u529b\u673a\u5236\u4e2d\u5934\u6570h\u8bbe\u7f6e\u4e3a8\uff0c\u5176\u4e2dd k = d v = d m /h = 64;\u8bad\u7ec3\u8fc7\u7a0b\u4e2dbatch\u5927\u5c0f\u8bbe\u7f6e\u4e3a32\uff0cdropout\u7387\u8bbe\u7f6e\u4e3a0.5\u3002 PNGL(-local) 42.68 65.35</td></tr><tr><td/><td colspan=\"4\">\u88683.\u5c40\u90e8\u548c\u5168\u5c40\u4fe1\u606f\u5206\u522b\u5bf9\u5e95\u5c42\u548c\u9876\u5c42\u7ed3\u6784\u8bc6\u522b\u7684\u5f71\u54cd</td></tr><tr><td colspan=\"5\">4.2 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u7ed3 \u7ed3 \u7ed3\u679c \u679c \u679c \u7531\u88683\u7684\u5b9e\u9a8c\u7ed3\u679c\u53ef\u77e5\uff0c\u76f8\u6bd4\u4e8e\u53ea\u8003\u8651\u5168\u5c40\u4fe1\u606f\u7684\u6a21\u578b\uff0cMVM\u5728\u6700\u5e95\u4e0b\u4e24\u5c42\u8282\u70b9\u6b63\u786e\u7387\u66f4</td></tr><tr><td colspan=\"5\">\u9ad8\uff0c\u8fd9\u8bf4\u660e\u8003\u8651\u5c40\u90e8\u4fe1\u606f\u7684\u5bf9\u4e8e\u5e95\u5c42\u7ed3\u6784\u8bc6\u522b\u6709\u5e2e\u52a9\u3002PNGL(-local)\u5728\u6700\u4e0a\u4e09\u5c42\u7684\u8282\u70b9\u6b63\u786e \u672c\u6587\u5c06\u6587\u4e2d\u63d0\u51fa\u7684\u6a21\u578bPNGL\u548c\u57fa\u51c6\u7cfb\u7edf\u8fdb\u884c\u4e86\u5bf9\u6bd4\uff0c\u57fa\u51c6\u7cfb\u7edf\u5206\u4e3a\u4e24\u79cd\uff1a1)\u53ea\u8003\u8651\u5c40\u90e8 \u7387\u8981\u9ad8\u4e8eMVM\uff0c\u8bf4\u660e\u76f8\u6bd4\u4e8e\u8003\u8651\u5c40\u90e8\u4fe1\u606f\u7684\u6a21\u578b\uff0c\u53ea\u8003\u8651\u5168\u5c40\u4fe1\u606f\u5bf9\u4e0a\u5c42\u7ed3\u6784\u8bc6\u522b\u6709\u5e2e\u52a9\u3002 \u4fe1\u606f2)\u53ea\u8003\u8651\u5168\u5c40\u4fe1\u606f\uff0c\u57fa\u51c6\u7cfb\u7edf\u4ecb\u7ecd\u5982\u4e0b\uff1a \u56e0\u6b64\u672c\u6587\u8ba4\u4e3a\u5728\u5168\u5c40\u4fe1\u606f\u7684\u57fa\u7840\u4e0a\u52a0\u5165\u5c40\u90e8\u4fe1\u606f\u53ef\u4ee5\u589e\u5f3a\u6a21\u578b\u5bf9\u4e8e\u5e95\u5c42\u8282\u70b9\u7684\u8bc6\u522b\u80fd\u529b\u3002 LD-CM\uff1a\u6027\u80fd\u6700\u597d\u7684\u4f20\u7edf\u6a21\u578b(Jiang et al., 2018a)\uff0c\u53ea\u8003\u8651\u5c40\u90e8\u4fe1\u606f\u3002\u8be5\u6a21\u578b\u91c7\u7528\u6761\u4ef6 \u4e3a\u4e86\u7814\u7a76\u5728\u5168\u5c40\u4fe1\u606f\u7684\u57fa\u7840\u4e4b\u4e0a\u878d\u5408\u5c40\u90e8\u4fe1\u606f\u5bf9\u4e8e\u7ed3\u6784\u8bc6\u522b\u7684\u5f71\u54cd\uff0c\u672c\u6587\u5728\u6a21\u578bPN\u7684\u57fa\u7840 \u968f\u673a\u573a\uff0c\u8fd0\u7528\u8f83\u591a\u7684\u624b\u5de5\u7279\u5f81\uff0c\u8003\u8651\u76f8\u90bb\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\u80fd\u591f\u5408\u5e76\uff0c\u8d2a\u5a6a\u7684\u81ea\u5e95\u5411\u4e0a\u8bc6\u522b\u7bc7\u7ae0\u7ed3 \u4e4b\u4e0a\uff0c\u52a0\u5165\u6bb5\u843d\u4ea4\u4e92\u5c42\uff0c\u7efc\u5408\u8003\u8651\u5168\u5c40\u548c\u5c40\u90e8\u4fe1\u606f\uff0c\u5f97\u5230\u6a21\u578bPN(+local)\uff1b\u800cPN\u548cPNGL(-\u6784\uff0c\u4ece\u800c\u6784\u5efa\u7bc7\u7ae0\u7ed3\u6784\u6811\u3002 local)\u90fd\u662f\u53ea\u8003\u8651\u5168\u5c40\u4fe1\u606f\u7684\u6307\u9488\u7f51\u7edc\u6a21\u578b\uff0c\u5b83\u4eec\u7684\u533a\u522b\u5728\u4e8ePN\u91c7\u7528\u53cc\u5411GRU\u5bf9\u6bb5\u843d\u8fdb\u884c\u7f16 MVM:\u6027\u80fd\u6700\u597d\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b(Zhou et al., 2019)\uff0c\u53ea\u8003\u8651\u5c40\u90e8\u4fe1\u606f\u3002\u8be5\u6a21\u578b\u4ece\u8bcd\u3001\u5c40\u90e8 \u7801\uff0c\u800cPNGL(-local)\u91c7\u7528\u591a\u5934\u6ce8\u610f\u529b\u673a\u5236\u5bf9\u6bb5\u843d\u8fdb\u884c\u7f16\u7801\u3002\u672c\u6587\u7edf\u8ba1\u4e86\u5185\u90e8\u7ed3\u70b9\u6b63\u786e\u7387\u4ee5\u53ca \u4e0a\u4e0b\u6587\u4ee5\u53ca\u8bdd\u9898\u8fd9\u4e09\u4e2a\u89d2\u5ea6\u51fa\u53d1\uff0c\u63d0\u51fa\u4e86\u8bcd\u5bf9\u76f8\u4f3c\u5ea6\u673a\u5236\u6765\u8861\u91cf\u76f8\u90bb\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\u7684\u8bed\u4e49\u3002\u5e76 \u91c7\u7528\u79fb\u8fdb\u89c4\u7ea6\u7684\u65b9\u6cd5\u6bcf\u6b21\u8003\u8651\u76f8\u90bb\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\u80fd\u5426\u5408\u5e76\uff0c\u4ece\u5de6\u5230\u53f3\u8bc6\u522b\u7bc7\u7ae0\u7ed3\u6784\uff0c\u4ece\u800c\u6784\u5efa \u6700\u5e95\u4e0b\u4e24\u5c42\u5185\u90e8\u7ed3\u70b9\u6b63\u786e\u7387\uff0c\u5982\u88684\u6240\u793a\u3002</td></tr><tr><td>\u7bc7\u7ae0\u7ed3\u6784\u6811\u3002</td><td>\u6a21\u578b</td><td colspan=\"3\">\u5185\u90e8\u8282\u70b9\u6b63\u786e\u7387(%) \u6700\u5e95\u4e0b\u4e24\u5c42\u5185\u90e8\u8282\u70b9\u6b63\u786e\u7387(%)</td></tr><tr><td colspan=\"5\">PN:\u672c\u6587\u590d\u73b0\u4e86\u5728RST-DT\u4e0a\u8868\u73b0\u4f18\u5f02\u7684\u7ed3\u6784\u8bc6\u522b\u6a21\u578bPN(Lin et al., 2019)\uff0c\u53ea\u8003\u8651\u5168\u5c40 PN 56.25 46.65</td></tr><tr><td colspan=\"5\">\u4fe1\u606f\u3002\u8be5\u6a21\u578b\u662f\u4e00\u4e2a\u6307\u9488\u7f51\u7edc\uff0c\u5728\u7f16\u7801\u5c42\u4f7f\u7528\u53cc\u5411GRU\u5bf9\u6574\u4e2a\u6587\u7ae0\u8fdb\u884c\u7f16\u7801\uff0c\u89e3\u7801\u5c42\u4f7f\u7528\u5355 PN(+local) 56.87 47.26</td></tr><tr><td colspan=\"4\">\u5411GRU\u8fdb\u884c\u89e3\u7801\uff0c\u81ea\u9876\u5411\u4e0b\u7684\u8bc6\u522b\u7bc7\u7ae0\u7ed3\u6784\uff0c\u6784\u5efa\u7bc7\u7ae0\u7ed3\u6784\u6811\u3002 PNGL(-local) 56.57</td><td>42.68</td></tr><tr><td colspan=\"2\">PNGL</td><td>\u6a21\u578b 58.42</td><td>\u5185\u90e8\u8282\u70b9\u6b63\u786e\u7387(%)</td><td>48.48</td></tr><tr><td/><td/><td>LD-CM</td><td>54.71</td></tr><tr><td/><td/><td>MVM</td><td>56.11</td></tr><tr><td/><td/><td>PN</td><td>56.25</td></tr><tr><td/><td/><td>PNGL</td><td>58.42</td></tr><tr><td/><td/><td colspan=\"2\">\u88682.\u6a21\u578b\u5728MCDTB\u4e0a\u7684\u6027\u80fd\u6bd4\u8f83</td></tr><tr><td colspan=\"5\">\u5b9e\u9a8c\u7ed3\u679c\u5982\u88682\u6240\u793a\u3002PNGL\u6a21\u578b\u6bd4\u4ec5\u8003\u8651\u5c40\u90e8\u4fe1\u606f\u7684LD-CM\u6a21\u578b\u6027\u80fd\u63d0\u5347\u4e863.71\uff0c\u6bd4\u4ec5\u8003</td></tr><tr><td colspan=\"5\">\u8651\u5c40\u90e8\u4fe1\u606f\u7684MVM\u6a21\u578b(\u76ee\u524d\u5728MCDTB\u4e0a\u6700\u597d\u7684\u7ed3\u6784\u8bc6\u522b\u7684\u6a21\u578b)\u6027\u80fd\u63d0\u5347\u4e862.31\uff0c\u6bd4\u4ec5\u8003</td></tr><tr><td colspan=\"5\">\u8651\u5168\u5c40\u4fe1\u606f\u7684PN\u6a21\u578b\u6027\u80fd\u63d0\u5347\u4e862.17\u3002\u5b8f\u89c2\u7bc7\u7ae0\u7ed3\u6784\u7406\u8bba(Van, 1980)\u6307\u51fa\uff0c\u6587\u7ae0\u4f1a\u6709\u4e00\u4e2a\u603b\u6444</td></tr><tr><td colspan=\"5\">\u5168\u7bc7\u7684\u4e3b\u9898\uff0c\u5e76\u5c42\u5c42\u5206\u89e3\uff0c\u7531\u4e0b\u5c42\u547d\u9898\u5c55\u5f00\u3002\u8fd9\u8bf4\u660e\u6bb5\u843d\u6216\u7bc7\u7ae0\u5355\u5143\u4e4b\u95f4\u7684\u5173\u7cfb\u5e76\u975e\u5f88\u677e\u6563\uff0c</td></tr><tr><td colspan=\"3\">\u90fd\u662f\u5728\u5bf9\u4e3b\u9898\u8fdb\u884c\u5206\u5c42\u9762\u7684\u5c55\u5f00\u53d9\u8ff0\u3002</td><td/></tr><tr><td colspan=\"5\">\u800cLD-CM\u548cMVM\u90fd\u662f\u8003\u8651\u76f8\u90bb\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\u8054\u7cfb\u7684\u7d27\u5bc6\u7a0b\u5ea6\uff0c\u4f46\u662f\u8fd9\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\u662f\u56f4</td></tr><tr><td colspan=\"5\">\u7ed5\u5171\u540c\u7684\u4e3b\u9898\u5c55\u5f00\u7684\uff0c\u5982\u679c\u4ec5\u4ec5\u8003\u8651\u4e24\u4e2a\u7bc7\u7ae0\u5355\u5143\u4e4b\u95f4\u7684\u8054\u7cfb\uff0c\u6a21\u578b\u5f80\u5f80\u4f1a\u504f\u5411\u4e8e\u5c06\u8fd9\u4e24\u4e2a\u7bc7</td></tr><tr><td colspan=\"5\">\u7ae0\u5355\u5143\u5408\u5e76\u6210\u66f4\u5927\u7684\u7bc7\u7ae0\u5355\u5143\u3002\u800cPN\u6a21\u578b\u901a\u8fc7\u8003\u8651\u6574\u4e2a\u7bc7\u7ae0\u5355\u5143\u7684\u8bed\u4e49\u4fe1\u606f\uff0c\u5c06\u7bc7\u7ae0\u5355\u5143\u5207\u5206</td></tr><tr><td colspan=\"5\">\u6210\u4e24\u4e2a\u8f83\u5c0f\u7684\u7bc7\u7ae0\u5355\u5143\u3002PN\u6a21\u578b\u4f1a\u5bf9\u6240\u6709\u53ef\u80fd\u5f62\u6210\u7684\u4e24\u4e2a\u8f83\u5c0f\u7bc7\u7ae0\u5355\u5143\u8bed\u4e49\u8054\u7cfb\u7684\u7d27\u5bc6\u7a0b\u5ea6\u8fdb</td></tr><tr><td colspan=\"5\">\u884c\u6392\u5e8f\uff0c\u53d6\u8bed\u4e49\u8054\u7cfb\u6700\u677e\u6563\u7684\u4e24\u4e2a\u8f83\u5c0f\u7bc7\u7ae0\u5355\u5143\u4f5c\u4e3a\u5207\u5206\u7ed3\u679c\u3002\u4f46\u662f\u6bcf\u4e2a\u7bc7\u7ae0\u5355\u5143\u5f80\u5f80\u5305\u542b\u8f83</td></tr><tr><td colspan=\"5\">\u590d\u6742\u7684\u6bb5\u843d\u8bed\u4e49\u4fe1\u606f\uff0c\u4ec5\u4ec5\u8003\u8651\u5168\u5c40\u4fe1\u606f\uff0c\u6a21\u578b\u5f88\u96be\u5bf9\u4e24\u4e2a\u8f83\u5c0f\u7bc7\u7ae0\u5355\u5143\u4e4b\u95f4\u7684\u8bed\u4e49\u8054\u7cfb\u7684\u7d27</td></tr><tr><td colspan=\"2\">\u5bc6\u7a0b\u5ea6\u8fdb\u884c\u6b63\u786e\u7684\u6392\u5e8f\u3002</td><td/><td/></tr></table>", | |
| "html": null, | |
| "type_str": "table" | |
| } | |
| } | |
| } | |
| } |