| { |
| "paper_id": "2020", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T12:54:43.556421Z" |
| }, |
| "title": "\u591a \u591a \u591a\u6a21 \u6a21 \u6a21\u5757 \u5757 \u5757\u8054 \u8054 \u8054\u5408 \u5408 \u5408\u7684 \u7684 \u7684\u9605 \u9605 \u9605\u8bfb \u8bfb \u8bfb\u7406 \u7406 \u7406\u89e3 \u89e3 \u89e3\u5019 \u5019 \u5019\u9009 \u9009 \u9009\u53e5 \u53e5 \u53e5\u62bd \u62bd \u62bd\u53d6 \u53d6 \u53d6 *", |
| "authors": [ |
| { |
| "first": "\u5409", |
| "middle": [ |
| "\u5409" |
| ], |
| "last": "\u5409\u5b87", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Technology\uff0cShanxi University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "\u738b", |
| "middle": [], |
| "last": "\u738b\u7b11 \u7b11 \u7b11\u6708", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Technology\uff0cShanxi University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "\u90ed", |
| "middle": [], |
| "last": "\u90ed\u5c11 \u5c11 \u5c11\u8339", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Technology\uff0cShanxi University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "\u5c71\u897f\u5927\u5b66", |
| "middle": [], |
| "last": "\u8ba1\u7b97\u673a\u4e0e\u4fe1\u606f\u6280\u672f\u5b66\u9662", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Technology\uff0cShanxi University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "\u5c71\u897f\u5927\u5b66", |
| "middle": [], |
| "last": "\u8ba1\u7b97\u667a\u80fd\u4e0e\u4e2d\u6587\u4fe1\u606f\u5904\u7406\u6559\u80b2\u90e8\u91cd\u70b9\u5b9e\u9a8c\u5ba4", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Technology\uff0cShanxi University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Yu", |
| "middle": [], |
| "last": "Ji", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Technology\uff0cShanxi University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "\u2021", |
| "middle": [ |
| "Xiaoyue" |
| ], |
| "last": "Wang", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Technology\uff0cShanxi University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Ru", |
| "middle": [], |
| "last": "Li", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Technology\uff0cShanxi University", |
| "location": {} |
| }, |
| "email": "liru@sxu.edu.cn" |
| }, |
| { |
| "first": "Shaoru", |
| "middle": [], |
| "last": "Guo", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Technology\uff0cShanxi University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Yong", |
| "middle": [], |
| "last": "Guan", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Technology\uff0cShanxi University", |
| "location": {} |
| }, |
| "email": "guanyong0130@163.com" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "downstream answer tasks is 3.68% and 3.6% respectively higher than that of full text input. The above results confirm the effectiveness of the proposed method.", |
| "pdf_parse": { |
| "paper_id": "2020", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "downstream answer tasks is 3.68% and 3.6% respectively higher than that of full text input. The above results confirm the effectiveness of the proposed method.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "As a key task of natural language understanding, machine reading comprehension has been widely concerned by scholars at domestic and foreign. In order to solve the problem of multiple choice reading comprehension, which is difficult to extract evidence sentences due to the absence of clue annotation and questions involve multi-hop reasoning, we proposes a model of evidence sentence extraction based on multi-module combination. Firstly,we use some labeled data to fine-tune the pre-training model; secondly, the evidence sentences in the multi-hop reasoning problem are extracted recursively through TF-IDF; finally, the unsupervised method is combined to further filter the model prediction results to reduce redundancy. This paper is verified on the Chinese Gaokao and the RACE data set. In the extraction of evidence sentences, compared with the optimal baseline model, the F1 value of the method in this paper is increased by 3.44%. The accuracy of using evidence sentences as model input in 1 \u5f15 \u5f15 \u5f15\u8a00 \u8a00 \u8a00 \u968f\u7740\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6280\u672f\u7684\u53d1\u5c55\uff0c\u56fd\u5185\u5916\u5bf9\u4e8e\u673a\u5668\u9605\u8bfb\u7406\u89e3\u7684\u7814\u7a76\u4e0d\u65ad\u6df1\u5165\u3002\u672c\u6587\u91cd\u70b9\u5173 \u6ce8\u673a\u5668\u9605\u8bfb\u7406\u89e3\u4e2d\u7684\u591a\u9879\u9009\u62e9\u9898\u4efb\u52a1 (Mostafazadeh et al., 2016; Lai et al., 2017) \uff0c\u5373\uff1a\u7ed9\u5b9a\u6587 \u7ae0\u3001\u95ee\u9898\u548c\u9009\u9879\uff0c\u8981\u6c42\u6839\u636e\u6587\u7ae0\u56de\u7b54\u95ee\u9898\uff0c\u4ece\u591a\u4e2a\u9009\u9879\u4e2d\u9009\u62e9\u6700\u4f73\u9009\u9879\u3002 \u5bf9\u4e8e\u8be5\u4efb\u52a1\uff0c\u7814\u7a76\u8005\u901a\u5e38\u5c06\u6574\u7bc7\u6587\u7ae0\u3001\u95ee\u9898\u53ca\u9009\u9879\u4f5c\u4e3a\u8f93\u5165 (Wang et al., 2018; Ran et al., 2019) \u5e76\u5728\u4e09\u8005\u4e4b\u95f4\u4e24\u4e24\u4ea4\u4e92\uff0c\u8fdb\u884c\u4fe1\u606f\u6574\u5408\u7ee7\u800c\u9009\u51fa\u6700\u4f73\u9009\u9879\u3002\u7136\u4e0e\u7247\u6bb5\u62bd\u53d6\u5f0f\u9605\u8bfb\u7406\u89e3\u4e0d \u540c\uff0c\u591a\u9879\u9009\u62e9\u7684\u7b54\u6848\u96be\u4ee5\u76f4\u63a5\u4ece\u7ed9\u5b9a\u7684\u6587\u7ae0\u4e2d\u63d0\u53d6 (Wang et al., 2019 )\uff0c(\u5982\u5728RACE (Lai et al., 2017) (Yadav et al., 2019) et al., 2015; Williams et al., 2018 )\u4e3a\u8bad\u7ec3\u5019\u9009\u53e5\u62bd\u53d6\u6a21\u578b\u3002\u5bf9\u4e8e\u4e0d\u63d0\u4f9b\u5019\u9009\u53e5\u6807 \u6ce8\u7684\u6570\u636e\u96c6\uff0c\u7814\u7a76\u8005\u4ece\u7ed3\u6784\u5316\u77e5\u8bc6\u5e93 (Speer et al., 2016) \u4e2d\u9009\u53d6\u76f8\u5173\u7ebf\u7d22\u77e5\u8bc6\uff0c\u8bad\u7ec3\u6a21\u578b (Hao et al., 2017; Lukovnikov et al., 2017 )\uff1b(3)\u4f7f\u7528\u4fe1\u606f\u68c0\u7d22\u8fdb\u884c\u5019\u9009\u53e5\u62bd\u53d6\u5de5\u4f5c\uff0c\u901a\u8fc7\u5f3a\u5316\u5b66 \u4e60(Geva and Berant, 2018)\u6216pagerank (Surdeanu et al., 2008) \u5b66\u4e60\u5982\u4f55\u5728\u7f3a\u5c11\u660e\u786e\u8bad\u7ec3\u6570\u636e\u7684 \u524d\u63d0\u4e0b\u8fdb\u884c\u5019\u9009\u53e5\u62bd\u53d6\u3002\u6216\u662f\u4f7f\u7528\u6ce8\u610f\u529b\u673a\u5236\u5728\u6587\u672c\u4e0e\u9009\u9879\u53ca\u95ee\u9898\u4e4b\u95f4\u4ea4\u4e92\uff0c\u4f7f\u6587\u7ae0\u4e2d\u4e0e\u9009\u9879 \u548c\u95ee\u9898\u76f8\u5173\u90e8\u5206\u7684\u6ce8\u610f\u529b\u6743\u91cd\u66f4\u5927 (Ran et al., 2019; Tang et al., 2019) ", |
| "cite_spans": [ |
| { |
| "start": 1068, |
| "end": 1095, |
| "text": "(Mostafazadeh et al., 2016;", |
| "ref_id": "BIBREF9" |
| }, |
| { |
| "start": 1096, |
| "end": 1113, |
| "text": "Lai et al., 2017)", |
| "ref_id": "BIBREF5" |
| }, |
| { |
| "start": 1181, |
| "end": 1200, |
| "text": "(Wang et al., 2018;", |
| "ref_id": "BIBREF16" |
| }, |
| { |
| "start": 1201, |
| "end": 1218, |
| "text": "Ran et al., 2019)", |
| "ref_id": "BIBREF10" |
| }, |
| { |
| "start": 1281, |
| "end": 1299, |
| "text": "(Wang et al., 2019", |
| "ref_id": "BIBREF17" |
| }, |
| { |
| "start": 1310, |
| "end": 1328, |
| "text": "(Lai et al., 2017)", |
| "ref_id": "BIBREF5" |
| }, |
| { |
| "start": 1329, |
| "end": 1349, |
| "text": "(Yadav et al., 2019)", |
| "ref_id": "BIBREF19" |
| }, |
| { |
| "start": 1350, |
| "end": 1363, |
| "text": "et al., 2015;", |
| "ref_id": "BIBREF0" |
| }, |
| { |
| "start": 1364, |
| "end": 1385, |
| "text": "Williams et al., 2018", |
| "ref_id": "BIBREF18" |
| }, |
| { |
| "start": 1425, |
| "end": 1445, |
| "text": "(Speer et al., 2016)", |
| "ref_id": "BIBREF12" |
| }, |
| { |
| "start": 1461, |
| "end": 1479, |
| "text": "(Hao et al., 2017;", |
| "ref_id": "BIBREF3" |
| }, |
| { |
| "start": 1480, |
| "end": 1503, |
| "text": "Lukovnikov et al., 2017", |
| "ref_id": "BIBREF7" |
| }, |
| { |
| "start": 1565, |
| "end": 1588, |
| "text": "(Surdeanu et al., 2008)", |
| "ref_id": "BIBREF13" |
| }, |
| { |
| "start": 1662, |
| "end": 1680, |
| "text": "(Ran et al., 2019;", |
| "ref_id": "BIBREF10" |
| }, |
| { |
| "start": 1681, |
| "end": 1699, |
| "text": "Tang et al., 2019)", |
| "ref_id": "BIBREF14" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u8fc7\u6ee4\u91cd\u590d\u4fe1\u606f\uff0c\u63d0\u5347\u7cbe\u786e\u7387\u3002 \u4e3a \u8fdb \u4e00 \u6b65 \u8bc4 \u4f30 \u5019 \u9009 \u53e5 \u62bd \u53d6 \u8d28 \u91cf \u53ca \u6240 \u63d0 \u65b9 \u6cd5 \u5bf9 \u540e \u7eed \u7b54 \u9898 \u7684 \u5e2e \u52a9 \uff0c \u672c \u6587 \u5c06 \u62bd \u53d6 \u51fa \u7684 \u5019 \u9009 \u53e5 \u96c6 \u62fc \u63a5 \uff0c \u91c7 \u7528BERT\u4e0eco-matching\u6a21 \u578b \u5206 \u522b \u5728RACE\u3001 \u9ad8 \u8003 \u8bed \u6587 \u9605 \u8bfb \u7406 \u89e3 \u9009 \u62e9 \u9898 \u6570 \u636e \u96c6 \u4e0a \u8fdb \u884c \u5b9e \u9a8c \uff0c \u5b9e \u9a8c \u7ed3 \u679c \u8868 \u660e \u91c7 \u7528 \u5019 \u9009 \u53e5 \u96c6 \u4f5c \u4e3a \u8f93 \u5165 \u76f8 \u6bd4 \u5168 \u6587 \u5728 \u9ad8 \u8003 \u53caRACE\u6570 \u636e \u96c6 \u4e0a \u5206 \u522b \u63d0 \u5347 \u4e863.68%\u53ca3.6%\u3002 \u5728 \u5019 \u9009 \u53e5 \u62bd \u53d6 \u4e0a \uff0c \u672c \u6587 \u6240 \u63d0 \u65b9 \u6848 \u76f8 \u6bd4 \u4e8e \u57fa \u7ebfF1\u503c \u8fdb \u4e00 \u6b65 \u63d0 \u5347 \u4e863.44%\u53ca3.95%\u3002 2 \u76f8 \u76f8 \u76f8\u5173 \u5173 \u5173\u5de5 \u5de5 \u5de5\u4f5c \u4f5c \u4f5c \u5019\u9009\u53e5\u62bd\u53d6\u5de5\u4f5c\uff0c\u4f9d\u636e\u8bad\u7ec3\u65b9\u5f0f\u53ef\u5212\u5206\u4e3a\u56db\u79cd\u7c7b\u578b\u3002(1)\u4f7f\u7528\u65e0\u76d1\u7763\u65b9\u6cd5\u4e3a\u5019\u9009\u53e5\u62bd\u53d6 \u63d0\u4f9b\u4e86\u6307\u5bfc\uff0c\u540c\u65f6\u51cf\u7701\u4eba\u5de5\u6807\u6ce8\u7684\u6d88\u8017(Yadav et al., 2019)\uff1b(2)\u6709\u76d1\u7763\u65b9\u6cd5\u901a\u8fc7\u6807\u6ce8\u6570\u636e \u8bad\u7ec3\u6a21\u578b\uff0c\u4ece\u800c\u5b9e\u73b0\u4e0b\u6e38\u4efb\u52a1\u4e2d\u81ea\u52a8\u62bd\u53d6\u5019\u9009\u53e5\u7684\u76ee\u7684\u3002Trivedi et al. (2019)\u4f7f\u7528\u6587\u672c\u8574\u542b\u8bed \u6599(Bowman", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\uff1b(4)\u901a\u8fc7\u4eba\u5de5\u5b9a\u4e49\u89c4 \u5219\uff0c\u62bd\u53d6\u51fa\u542b\u6709\u566a\u58f0\u4fe1\u606f\u7684\u5019\u9009\u53e5\uff0c\u4f7f\u7528\u5f31\u76d1\u7763\u65b9\u5f0f\u8bad\u7ec3\u6a21\u578b(Min et al., 2018)\u3002\u4e0a\u8ff0\u5de5\u4f5c\uff0c \u5404\u6709\u5176\u8d21\u732e\u4e4b\u5904\u4e0e\u610f\u4e49\uff0c\u63a8\u52a8\u4e86\u6a21\u578b\u5728\u76f8\u5e94\u4e0b\u6e38\u4efb\u52a1\u4e0a\u7684\u6027\u80fd\u8868\u73b0\u3002\u672c\u6587\u6240\u63d0\u5de5\u4f5c\u7740\u91cd\u5728\u5bf9\u4e0a \u8ff0\u5de5\u4f5c\u758f\u6f0f\u4e4b\u5904\u8fdb\u884c\u5f3a\u5316\uff0c\u7efc\u5408\u4f7f\u7528\u6709\u76d1\u7763\u4e0e\u65e0\u76d1\u7763\u65b9\u5f0f\uff0c\u4f7f\u62bd\u53d6\u7ed3\u679c\u53ef\u8bc4\u4ef7\u5e76\u4e14\u63d0\u9ad8\u62bd\u53d6\u7ed3 \u679c\u7684\u7cbe\u786e\u6027\u4e5f\u51cf\u7701\u4e86\u6570\u636e\u6807\u6ce8\u5de5\u4f5c\u91cf\u3002\u540c\u65f6\uff0c\u5bf9\u4e0a\u8ff0\u6a21\u578b\u4e2d\u672a\u80fd\u8003\u8651\u5230\u7684\u9009\u9879\u4fe1\u606f\u7f3a\u5931\u95ee\u9898\u4ee5 \u53ca\u6b63\u8d1f\u6837\u672c\u4e0d\u5747\u8861\u4e5f\u8fdb\u884c\u4e86\u76f8\u5e94\u5904\u7406\u3002\u6b64\u5916\uff0c\u672c\u6587\u9488\u5bf9\u591a\u6b65\u63a8\u7406\u95ee\u9898\u63d0\u51fa\u4e86\u4e00\u79cd\u591a\u6b65\u4fe1\u606f\u62bd\u53d6 \u65b9\u5f0f\uff0c\u8fdb\u4e00\u6b65\u63d0\u5347\u4e86\u6a21\u578b\u62bd\u53d6\u6548\u679c\u3002\u5e76\u5728\u4e0b\u6e38\u4efb\u52a1\u4e2d\u9a8c\u8bc1\u4e86\u6a21\u578b\u7684\u6709\u6548\u6027\u3002 3 \u5019 \u5019 \u5019\u9009 \u9009 \u9009\u53e5 \u53e5 \u53e5\u62bd \u62bd \u62bd\u53d6 \u53d6 \u53d6\u6a21 \u6a21 \u6a21\u578b \u578b \u578b \u672c\u6587\u63d0\u51fa\u4e00\u79cd\u65b0\u7684\u5019\u9009\u53e5\u62bd\u53d6\u6a21\u578b\uff0c\u6a21\u578b\u6574\u4f53\u67b6\u6784\u5982\u56fe2\u6240\u793a\u3002\u5176\u4e3b\u8981\u5305\u542b\u56db\u90e8\u5206\uff1a(1) \u9009\u9879\u6539\u5199\u6a21\u5757\uff1a\u878d\u5408\u9009\u9879\u4e0e\u95ee\u9898\u6240\u6db5\u76d6\u7684\u4fe1\u606f\uff0c\u786e\u4fdd\u5176\u7ed3\u679c\u65e0\u8bed\u6cd5\u9519\u8bef\uff1b(2)\u5019\u9009\u53e5\u62bd\u53d6\u6a21 \u5757\uff1a\u4ece\u6587\u7ae0\u4e2d\u521d\u6b65\u7b5b\u9009\u51fa\u4e0e\u5224\u65ad\u9009\u9879\u6b63\u8bef\u6709\u5173\u7684\u53e5\u5b50\u96c6\u5408\uff1b(3)TF-IDF\u9012\u5f52\u62bd\u53d6\u6a21\u5757\uff1a\u5728\u524d \u4e00\u6b65\u7684\u57fa\u7840\u4e0a\uff0c\u4f7f\u7528TF-IDF\u4f5c\u4e3a\u5f15\u5bfc\uff0c\u62bd\u53d6\u591a\u6b65\u63a8\u7406\u95ee\u9898\u5019\u9009\u53e5\uff0c\u907f\u514d\u5173\u952e\u4fe1\u606f\u9057\u6f0f\uff1b(4) \u7b5b\u9009\u6a21\u5757\uff1a\u5728\u6240\u5f97\u53e5\u5b50\u96c6\u5408\u4e0a\u8fdb\u4e00\u6b65\u7b5b\u9009\uff0c\u63d0\u9ad8\u5019\u9009\u53e5\u62bd\u53d6\u7cbe\u786e\u7387\uff0c\u964d\u4f4e\u4fe1\u606f\u5197\u4f59\u3002 \u56fe 2. \u5019\u9009\u53e5\u62bd\u53d6\u6a21\u578b \u8ba1\u7b97\u8bed\u8a00\u5b66 3.1 \u9009 \u9009 \u9009\u9879 \u9879 \u9879\u6539 \u6539 \u6539\u5199 \u5199 \u5199\u6a21 \u6a21 \u6a21\u5757 \u5757 \u5757", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u901a\u8fc7\u5bf9\u9ad8\u8003\u9605\u8bfb\u7406\u89e3\u53caRACE\u6570\u636e\u96c6\u5206\u6790\u540e\u53d1\u73b0\uff0c\u5982\u56fe3\u6240\u793a\uff0c\u5f53\u95ee\u9898\u4e3a\"\u4e0b\u5217\u8bf4\u6cd5\u7b26\u5408 (\u4e0d\u7b26\u5408)\u6587\u610f\u7684\u4e00\u9879\u662f\uff1f(\u6216\u5176\u540c\u4e49\u8868\u8ff0),\u8be5\u7c7b\u95ee\u9898\u6240\u8574\u542b\u7684\u4fe1\u606f\u91cf\u8f83\u5c11\uff0c\u9009\u9879\u4fe1\u606f\u5b8c\u6574\uff0c \u65e0\u9700\u5bf9\u9009\u9879\u6539\u5199\uff1b\u800c\u5f53\u95ee\u9898\u4e3a\"\u4e0b\u5217\u5bf9'\u56fd\u5916\u5a92\u4f53\u5173\u6ce8\u70b9'\u7684\u7406\u89e3\uff0c\u4e0d\u6b63\u786e(\u6b63\u786e)\u7684\u4e00\u9879\u662f\uff1f\"\uff0c \u9009\u9879\u5185\u5bb9\u4e3a\"\u79d1\u6280\u7ade\u4e89\u529b\"\uff0c\u82e5\u4ec5\u4f7f\u7528\u9009\u9879\u5185\u5bb9\uff0c\u5176\u6db5\u76d6\u4fe1\u606f\u91cf\u8fc7\u5c11\uff0c\u62bd\u53d6\u5bf9\u5e94\u5019\u9009\u53e5\u4f1a\u8f83\u4e3a\u56f0 \u96be\uff1b\u800c\u82e5\u5c06\u95ee\u9898\u4e0e\u9009\u9879\u76f4\u63a5\u62fc\u63a5\uff0c\u6240\u5f97\u7ed3\u679c\u4e0d\u7b26\u5408\u8bed\u6cd5\u89c4\u5219\u3002\u6545\u9700\u63d0\u53d6\u95ee\u9898\u7684\u5173\u952e\u4fe1\u606f\uff0c\u5e76\u5c06 \u8868 4. \u5019\u9009\u53e5\u62bd\u53d6\u9519\u8bef\u793a\u4f8b \u7531\u8868\u53ef\u77e5\uff0c\u9519\u8bef\u539f\u56e0\u4e3b\u8981\u6709\u4ee5\u4e0b\u4e09\u70b9\uff1a(1)\u6307\u4ee3\u95ee\u9898\uff0c\u9700\u8fa8\u522b\u8868\u4e2d\u7684\"\u5b83\u4eec\"\u6307\u4ee3\"\u8709\u8763\"\uff0c \u624d\u53ef\u77e5\"\u7e41\u6b96\"\u4e0e\"\u751f\u547d\u5ef6\u7eed\"\u8574\u542b\u3002(2)\u5f52\u7eb3\u6982\u62ec\u95ee\u9898\uff1a\u5982\"\u6211\u4eec\u90fd\u4f9d\u7136\u5f97\u548c\u4ed6\u4e00\u8d77\uff0c\u627f\u53d7\u4e00\u4e2a \u5404\u4eba\u5fc3\u5e95\u7684\u8bda\u4e0e\u7231\u90fd\u5c1a\u6709\u4e0d\u8db3\u7684\u65f6\u4ee3\u3002\"\u5c1a\u6709\u4e0d\u8db3\u7684\u8a00\u5916\u4e4b\u610f\u662f\uff1a\"\u4f46\u90a3\u4e2a\u65f6\u4ee3\u7684\u56fd\u6c11\u52a3\u6839\u6027\u4eca \u5929\u4f9d\u7136\u5b58\u5728\"\uff0c\u7136\u5176\u8868\u8ff0\u5dee\u5f02\u6027\u8f83\u5927\uff0c\u5bfc\u81f4\u8ba1\u7b97\u673a\u65e0\u6cd5\"\u7406\u89e3\"\u3002(3)\u6d89\u53ca\u5f52\u7eb3\u4e0e\u77e5\u8bc6\u878d\u5408\uff1a\u5982 \u9700\u4f7f\u6a21\u578b\u77e5\u9053\"\u4e66\u7248\u5b8b\u3001\u62a5\u7248\u5b8b\u3001\u6807\u9898\u5b8b\u3001\u4eff\u5b8b\u7b49\"\u5373\u4e3a\"\u5b57\u5f62\u5b57\u4f53\"\u3002 7 \u7ed3 \u7ed3 \u7ed3\u8bba \u8bba \u8bba\u4e0e \u4e0e \u4e0e\u5c55 \u5c55 \u5c55\u671b \u671b \u671b \u672c\u6587\u9488\u5bf9\u591a\u9879\u9009\u62e9\u9605\u8bfb\u7406\u89e3\u5019\u9009\u53e5\u62bd\u53d6\u4efb\u52a1\uff0c\u4ee5\u6709\u76d1\u7763\u65b9\u5f0f\u62bd\u53d6\u4e3a\u57fa\u7840\uff0c\u9488\u5bf9\u9009\u9879\u8bed\u4e49\u4e0d \u5b8c\u6574\u3001\u6570\u636e\u96c6\u6b63\u8d1f\u6837\u672c\u4e0d\u5747\u8861\u3001\u53ca\u62bd\u53d6\u7ed3\u679c\u4fe1\u606f\u5197\u4f59\u7b49\u65b9\u9762\u8fdb\u884c\u6539\u8fdb\u3002\u5728\u9ad8\u8003\u53caRACE\u6570\u636e\u96c6 \u4e0a\u8fdb\u884c\u5b9e\u9a8c\uff0c\u8bc1\u5b9e\u4e86\u8be5\u65b9\u6cd5\u7684\u6709\u6548\u6027\u3002\u540c\u65f6\uff0c\u8fd8\u9a8c\u8bc1\u4e86\u5019\u9009\u53e5\u62bd\u53d6\u5bf9\u591a\u9879\u9009\u62e9\u7b54\u6848\u9884\u6d4b\u7684\u5e2e \u52a9\u3002\u6b64\u5916\uff0c\u4ece\u88682\u4e2d\u53ef\u770b\u51fa\uff0c\u5019\u9009\u53e5\u62bd\u53d6\u4ecd\u5b58\u5728\u8f83\u5927\u63d0\u5347\u7a7a\u95f4\u3002\u7ed3\u5408\u9519\u8bef\u5206\u6790\uff0c\u4e0b\u4e00\u6b65\u8ba1\u5212\u6316\u6398 \u9605\u8bfb\u7406\u89e3\u4e2d\u66f4\u6df1\u5c42\u6b21\u7684\u7ebf\u7d22(\u5982\u53e5\u95f4\u6307\u4ee3\u5173\u8054)\uff0c\u63d0\u5347\u5019\u9009\u53e5\u62bd\u53d6\u6548\u679c\uff0c\u8fdb\u4e00\u6b65\u63d0\u9ad8\u7b54\u6848\u9884\u6d4b \u7684\u51c6\u786e\u7387\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": "\u5176\u4e0e\u9009\u9879\u4fe1\u606f\u878d\u5408\uff0c\u5f62\u6210\u4e00\u6761\u5b8c\u6574\u7684\u53e5\u5b50\u3002 \u9488\u5bf9\u4e0a\u8ff0\u4e24\u79cd\u60c5\u51b5\uff0c\u9996\u5148\u91c7\u7528\u6b63\u5219\u8868\u8fbe\u5f0f\u8fdb\u884c\u9009\u9879\u5185\u5bb9\u6539\u5199\uff0c\u4f7f\u5176\u5f62\u6210\u5b8c\u6574\u9648\u8ff0\u53e5H = {H 1 , H 2 , . . . , H m }\u5176\u4e2dm\u4e3a\u9009\u9879\u6539\u5199\u53e5\u7684\u957f\u5ea6\uff1b\u4e4b\u540e\u5c06\u6587\u7ae0\u5207\u5206\u4e3a\u53e5\u5b50P = {P 1 , P 2 , . . . , P n }\u5176 \u4e2dP \u4e3a\u6587\u7ae0\uff0cn\u4e3a\u6587\u7ae0\u4e2d\u53e5\u5b50\u6570\u91cf\u3002 \u56fe 3. \u9009\u9879\u6539\u5199\u793a\u4f8b 3.2 \u5019 \u5019 \u5019\u9009 \u9009 \u9009\u53e5 \u53e5 \u53e5\u62bd \u62bd \u62bd\u53d6 \u53d6 \u53d6\u6a21 \u6a21 \u6a21\u5757 \u5757 \u5757 \u8be5 \u6a21 \u5757 \u901a \u8fc7 \u8ba1 \u7b97P i \u4e0e \u9009 \u9879 \u6539 \u5199 \u53e5H\u7684 \u5173 \u8054 \u5ea6 \uff0c \u521d \u6b65 \u62bd \u53d6 \u51fa \u5019 \u9009 \u53e5 \u3002 \u672c \u6587 \u5728BERT\u57fa \u7840 \u4e0a \u8fdb \u884c \u5b9e \u9a8c \uff0c \u9996 \u5148 \u5c06[CLS],\u53e5 \u5b50P i ,[SEP ],\u9009 \u9879 \u6539 \u5199H,\u53ca[SEP ]\u62fc \u63a5 \u540e \u8f93 \u5165 \u6a21 \u578b \u4e2d \uff0c \u5176 \u4e2d[SEP ]\u4e3aBERT\u4e2d\u7684\u7247\u6bb5\u5206\u9694\u7b26\uff0c[CLS]\u4e3a\u7279\u6b8a\u5b57\u7b26(\u8f93\u5165\u6574\u4f53\u8868\u793a)\u3002\u7f16\u7801\u540e\uff0c\u53d6[CLS]\u7684 \u7f16\u7801\u7ed3\u679cO i \u2208 R d \u8fdb\u884c\u5206\u7c7b\uff0cd\u4e3aBERT\u9690\u85cf\u5c42\u7ef4\u5ea6\u3002 3.2.1 Focal Loss \u7531\u4e8e\u5019\u9009\u53e5\u6570\u636e\u96c6\u4e2d\u5b58\u5728\u6b63\u8d1f\u6837\u672c\u4e0d\u5747\u8861\u73b0\u8c61(RACE\u5019\u9009\u53e5\u6570\u636e\u96c6\u4e2d\uff0c\u6b63\u8d1f\u6837\u672c\u6bd4 \u4e3a1\uff1a10)\u3002\u672c\u6587\u91c7\u7528FocalLoss\u4f5c\u4e3a\u635f\u5931\u51fd\u6570\uff0c\u4f7f\u6a21\u578b\u805a\u7126\u4e8e\u6b63\u6837\u672c\u7684\u5b66\u4e60,\u7f13\u89e3\u6837\u672c\u7c7b\u522b\u4e0d\u5747 \u8861\u5e26\u6765\u7684\u98ce\u9669\u3002 \u8f93\u5165\u7684\u5019\u9009\u53e5\u5bf9\u4e3a(P i , H),\u6a21\u578b\u9884\u6d4b\u7ed3\u679c\u4e3aP=[p 0 ,p 1 ],\u771f\u503c\u4e3aY=[y 0 ,y 1 ]\u3002\u5bf9\u4e8e\u4f20\u7edf\u7684\u4ea4\u53c9\u71b5 \u635f\u5931\u800c\u8a00\uff0c\u5176\u8868\u793a\u4e3a\uff1aCE = \u2212(y 0 log(p 0 ) + y 1 log(p 1 )),\u663e\u7136\uff0c\u5f53\u8d1f\u6837\u672c\u5360\u6bd4\u8f83\u5927\u65f6\uff0c\u6a21\u578b\u7684\u8bad \u7ec3\u4f1a\u88ab\u8d1f\u6837\u672c\u5360\u636e\uff0c\u4f7f\u5f97\u6a21\u578b\u96be\u4ee5\u4ece\u6b63\u6837\u672c\u4e2d\u5b66\u4e60\u3002 L f l = \u2212\u03b1(1 \u2212 y ) log y y = 1 \u2212(1 \u2212 \u03b1)y \u03b3 log 1\u2212y y = 0 (1) FocalLoss\u5728\u539f\u6709\u7684\u57fa\u7840\u52a0\u5165\u6743\u91cd\u7cfb\u6570\u03b3\u53ca\u03b1,\u03b3\u51cf\u5c11\u6613\u5206\u7c7b\u6837\u672c\u7684\u635f\u5931\u4f7f\u6a21\u578b\u66f4\u5173\u6ce8\u4e8e\u56f0\u96be\u7684\u3001\u9519 \u5206\u7684\u6837\u672c\uff1b\u03b1\u7528\u4e8e\u5e73\u8861\u6b63\u8d1f\u6837\u672c\u672c\u8eab\u6570\u91cf\u6bd4\u4f8b\u4e0d\u5747,\u7531\u6b64\u7f13\u89e3\u4e86\u6b63\u8d1f\u6837\u672c\u4e0d\u5747\u8861\u7684\u73b0\u8c61\u3002 3.3 TF-IDF\u9012 \u9012 \u9012\u5f52 \u5f52 \u5f52\u62bd \u62bd \u62bd\u53d6 \u53d6 \u53d6\u6a21 \u6a21 \u6a21\u5757 \u5757 \u5757 \u7531\u4e8e\u9605\u8bfb\u7406\u89e3\u591a\u9879\u9009\u62e9\u9898\u4e2d\u5b58\u5728\u591a\u6b65\u63a8\u7406\u95ee\u9898\uff0c\u5982\u56fe1\u6240\u793a\uff0c\u8be5\u60c5\u51b5\u96be\u4ee5\u76f4\u63a5\u4f7f\u7528\u6587\u672c\u8574\u542b \u65b9\u5f0f\u5c06\u9009\u9879\u5bf9\u5e94\u5019\u9009\u53e5\u5168\u90e8\u62bd\u51fa\u3002\u8003\u8651\u5230\u591a\u6b65\u63a8\u7406\u95ee\u9898\u4e2d\u5b58\u5728\u94fe\u5f0f\u5173\u7cfb\uff0c\u6545\u57fa\u4e8e\u4e0a\u4e00\u6b65\u6240\u5f97\u7ed3 \u679cE = {E}\uff0c\u9996\u5148\u9009\u51fa\u4e0e\u9009\u9879\u6539\u5199\u53e5\u5173\u8054\u5ea6\u6700\u9ad8\u7684\u53e5\u5b50\u4f5c\u4e3a\u7b2c\u4e00\u8df3\u5019\u9009\u53e5hop1\u3002\u7ee7\u800c\uff0c\u8ba1\u7b97\u5176 \u4e0e\u6587\u7ae0\u53e5\u5b50(\u9664\u672c\u8eab\u4e4b\u5916)\u7684\u76f8\u4f3c\u5ea6\uff0c\u53d6\u76f8\u4f3c\u5ea6\u6700\u9ad8\u7684\u4f5c\u4e3a\u7b2c\u4e8c\u8df3\u5019\u9009\u53e5hop2\u3002\u4e4b\u540e\uff0c\u8ba1\u7b97\u6587 \u7ae0\u53e5\u5b50\u4e2d\u4e0ehop2\u4e4b\u95f4\u7684\u5173\u8054\u5ea6(hop1\u4e0ehop2\u9664\u5916)\uff0c\u5e76\u53d6\u5173\u8054\u5ea6\u6700\u9ad8\u53e5\u5b50\u4f5c\u4e3a\u7b2c\u4e09\u8df3\u5019\u9009\u53e5\uff0c \u4ee5\u6b64\u7c7b\u63a8\uff0c\u91cd\u590dK\u6b21(K\u503c\u89c6\u5177\u4f53\u60c5\u51b5\u8bbe\u5b9a)\u3002\u5c06\u6240\u5f97\u7684\u53e5\u5b50\u4e0e\u5019\u9009\u53e5\u96c6\u5408E\u5408\u5e76\u3002 3.4 \u5019 \u5019 \u5019\u9009 \u9009 \u9009\u53e5 \u53e5 \u53e5\u7b5b \u7b5b \u7b5b\u9009 \u9009 \u9009\u6a21 \u6a21 \u6a21\u5757 \u5757 \u5757 \u4e3a\u63d0\u5347\u62bd\u53d6\u7ed3\u679c\u7cbe\u786e\u6027\uff0c\u964d\u4f4e\u65e0\u5173\u53ca\u5197\u4f59\u4fe1\u606f\u6bd4\u91cd\u3002\u672c\u6587\u4f7f\u7528ROCC\u5bf9\u7ed3\u679c\u8fdb\u4e00\u6b65\u7b5b\u9009\u3002 \u9996\u5148\uff0c\u4f7f\u7528\u4e0a\u4e00\u6b65\u7684\u62bd\u53d6\u7ed3\u679cE\u5e76\u5bf9\u5176\u8fdb\u884c\u5168\u7ec4\u5408 n m \uff0c\u751f\u6210\u5019\u9009\u53e5\u96c6\u5408G\uff0c\u5176\u4e2dn\u4e3a\u62bd\u53d6\u7ed3 \u679c\u7684\u603b\u5171\u53e5\u5b50\u6570\uff0cm\u4e3a\u7ec4\u5408\u5355\u4f4d(\u53ef\u4f9d\u636e\u5177\u4f53\u60c5\u51b5\u8c03\u8282\u5927\u5c0f)\u3002\u4e4b\u540e\uff0c\u5bf9\u6bcf\u7ec4\u96c6\u5408\u5206\u522b\u4ece(1)\u96c6 \u5408\u5185\u90e8\u7684\u4fe1\u606f\u5197\u4f59\u5ea6(2)\u96c6\u5408\u5bf9\u9009\u9879\u7684\u4fe1\u606f\u8986\u76d6\u7387(3)\u96c6\u5408\u4e0e\u9009\u9879\u4e4b\u95f4\u4fe1\u606f\u76f8\u5173\u6027\u4e09\u4e2a\u89d2\u5ea6\u8ba1\u7b97\u5f97 \u5206\u3002 3.4.1 \u5197 \u5197 \u5197\u4f59 \u4f59 \u4f59\u5ea6 \u5ea6 \u5ea6 \u901a\u8fc7\u8ba1\u7b97\u7ed9\u5b9a\u96c6\u5408\u4e2d\u53e5\u5bf9\u95f4\u4fe1\u606f\u91cd\u5408\u5ea6\uff0c\u6765\u786e\u4fdd\u5019\u9009\u53e5\u7684\u591a\u6837\u6027\u548c\u4fe1\u606f\u4e92\u8865\u6027\u3002\u5f97\u5206\u8d8a\u4f4e \u7684\u53e5\u5b50\u96c6\u5408\uff0c\u4fe1\u606f\u5197\u4f59\u5ea6\u8d8a\u4f4e\u3002 O(G) = g i \u2208G g j \u2208G |t|g i |\u2229t|g j || max(t|g i |,t|g j |) |G| 2 (2) \u5176\u4e2dG\u8868\u793a\u7ed9\u5b9a\u53e5\u5b50\u96c6\u5408\uff0cg i \u4e0eg j \u5206\u522b\u8868\u793a\u96c6\u5408\u4e2d\u7684\u67d0\u4e00\u6761\u53e5\u5b50\uff0ct(g i )\u8868\u793ag i \u6240\u5305\u542b\u7684\u8bcd\u96c6 \u5408(\u53bb\u91cd\u540e)\uff0c|t |g i | \u2229 t |g j ||\u8868\u793ag i \u4e0eg j \u7684\u5171\u6709\u8bcd\u6570\u91cf\u3002 3.4.2 \u8986 \u8986 \u8986\u76d6 \u76d6 \u76d6\u7387 \u7387 \u7387 \u8be5 \u6a21 \u5757 \u7528 \u4e8e \u8861 \u91cf \u7ed9 \u5b9a \u96c6 \u5408G\u5bf9 \u9009 \u9879 \u6539 \u5199 \u53e5H\u7684 \u8bcd \u6c47 \u8986 \u76d6 \u7387,\u7531H\u4e0e \u96c6 \u5408G\u4e4b \u95f4 \u7684 \u5171 \u6709 \u8bcd \u7684IDF\u503c\u52a0\u6743\u5e73\u5747\u5f97\u5230\u3002Coverage\u503c\u8d8a\u5927\uff0c\u610f\u5473\u8be5\u96c6\u5408\u5305\u542b\u9009\u9879\u6539\u5199\u53e5\u7684\u4fe1\u606f\u8d8a\u591a\u3002 C t (H) = g i \u2208G t(H) \u2229 t(g 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(H, G) = n i w i \u2022 R(h i , G) (5) \u4ece\u4e0a\u8ff0\u4e09\u4e2a\u89d2\u5ea6\u5206\u522b\u8ba1\u7b97\u51fa\u7ed9\u5b9a\u96c6\u5408\u5f97\u5206\u540e\uff0c\u7efc\u5408\u5f97\u5206\u8ba1\u7b97\u96c6\u5408\u7684ROCC\u503c\u3002 S(G) = R \u03b5 + O(G) \u2022 R(\u03b5 + C(H)) (6) \u5982\u5f0f6\u4e2d\u6240\u793a\uff0cR\u4e3a\u96c6\u5408\u4e0e\u9009\u9879\u6539\u5199\u53e5\u7684relevance\u5f97\u5206\uff0cO\u4e3a\u96c6\u5408\u7684overlap\u503c\uff0cC(H)\u4e3a\u96c6\u5408\u5bf9 \u9009\u9879\u6539\u5199\u53e5\u7684coverage\u503c\uff0c\u4e3a\u907f\u514d\u8ba1\u7b97\u4e2d\u51fa\u73b0\u5206\u5b50\u6216\u5206\u6bcd\u4e3a0\u7684\u60c5\u51b5\uff0c\u6dfb\u52a0\u03b5\u4f5c\u4e3a\u5e73\u6ed1\u9879\uff0c\u5b9e\u9a8c \u4e2d\u8bbe\u03b5\u503c\u4e3a1\u3002\u4e4b\u540e\uff0c\u9009ROCC\u5f97\u5206\u6700\u5927\u96c6\u5408\u4f5c\u4e3a\u6700\u7ec8\u7684\u5019\u9009\u53e5\u96c6\u5408E 2 \u3002 4 \u7b54 \u7b54 \u7b54\u9898 \u9898 \u9898\u6a21 \u6a21 \u6a21\u578b \u578b \u578b \u5f97\u5230\u5019\u9009\u53e5\u96c6\u5408\u540e\uff0c\u5c06\u5176\u53e5\u5b50\u62fc\u63a5\u4e3a\u6587\u7ae0C\u3002\u4e4b\u540e\u540c\u95ee\u9898Q\uff0c\u9009\u9879O i \u4e00\u8d77\u4f5c\u4e3a\u7b54\u9898\u6a21\u578b\u7684 \u8f93\u5165\u3002 A i = f (C, Q, O i )", |
| "eq_num": "(7)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "L(A t |C, Q) = \u2212 log exp(W T \u2022 A t ) m j=1 exp(W T \u2022 A j ) (8) \u5f0f7\u4e2d\uff0cf (.)\u8868\u793a\u6a21\u578b\u7f16\u7801\u8fc7\u7a0b\uff0c\u6240\u5f97A i \u2208 R d \u4e3a\u6587\u7ae0\uff0c\u95ee\u9898\uff0c\u9009\u9879\u7684\u6700\u7ec8\u8868\u793a\uff0c\u5176\u4e2dd\u4e3a\u6a21\u578b\u7ef4 \u5ea6\u3002\u5f0f8\u4e2d\uff0cW \u2208 R d\u00d74 \u4e3a\u53c2\u6570\u77e9\u9635\uff0cA t \u4e3a\u95ee\u9898\u7684\u6b63\u786e\u9009\u9879\u3002 5 \u6570 \u6570 \u6570\u636e \u636e \u636e\u96c6 \u96c6 \u96c6 5.1 \u5019 \u5019 \u5019\u9009 \u9009 \u9009\u53e5 \u53e5 \u53e5\u6570 \u6570 \u6570\u636e \u636e \u636e\u96c6 \u96c6 \u96c6 \u9ad8 \u9ad8 \u9ad8\u8003 \u8003 \u8003\u5019 \u5019 \u5019\u9009 \u9009 \u9009\u53e5 \u53e5 \u53e5\u6570 \u6570 \u6570\u636e \u636e \u636e\u96c6 \u96c6 \u96c6\uff1a\u7531\u4e8e\u7f3a\u5c11\u4e2d\u6587\u9605\u8bfb\u7406\u89e3\u5019\u9009\u53e5\u8bed\u6599\uff0c\u672c\u6587\u4ece\u6570\u636e\u96c6\u4e2d\u968f\u673a\u62bd\u53d6500\u9053 \u9898,\u5bf9\u6bcf\u4e2a\u9009\u9879\u4eba\u5de5\u6807\u6ce8\u5176\u5019\u9009\u53e5\u3002\u6807\u6ce8\u89c4\u5219\u4e3a\uff1a\u5bf9\u5e94\u6bcf\u4e2a\u9009\u9879\uff0c\u6587\u7ae0\u4e2d\u4e0e\u5224\u65ad\u5176\u6b63\u8bef\u6709\u5173\u53e5\u5b50 \u6807\u6ce8\u4e3a1\uff0c\u53cd\u4e4b\uff0c\u6807\u6ce8\u4e3a0\u3002\u4e3a\u786e\u4fdd\u6570\u636e\u6807\u6ce8\u8d28\u91cf\uff0c\u672c\u6587\u91c7\u53d6\u4ea4\u53c9\u9a8c\u8bc1\u7684\u6807\u6ce8\u65b9\u5f0f\uff1a\u5c06\u6570\u636e\u4e8c\u7b49 \u5206\uff0c\u7531\u56db\u4e2a\u540c\u5b66\u4e24\u4e24\u4e00\u7ec4\u8fdb\u884c\u6807\u6ce8\uff0c\u5404\u7ec4\u5185\u540c\u5b66\u6807\u6ce8\u7684\u6570\u636e\u76f8\u540c\u3002\u6807\u6ce8\u540e\u4e24\u7ec4\u540c\u5b66\u4ea4\u6362\u8fdb\u884c\u4e24 \u8f6e\u6821\u9a8c\uff0c\u9488\u5bf9\u6807\u6ce8\u7ed3\u679c\u4e2d\u4e0d\u4e00\u81f4\u6570\u636e\uff0c\u7531\u4ef2\u88c1\u8005\u4ef2\u88c1\u8fdb\u884c\u7b2c\u4e09\u8f6e\u6821\u9a8c\uff0c\u5254\u9664\u65e0\u6cd5\u786e\u5b9a\u7684\u6570\u636e\uff0c \u82e5\u65e0\u5f02\u8bae\uff0c\u7ecf\u4e09\u8f6e\u9a8c\u8bc1\u540e\uff0c\u5c06\u6240\u5f97\u6807\u6ce8\u7ed3\u679c\u786e\u5b9a\u4e3a\u6700\u7ec8\u5019\u9009\u53e5\u96c6\u5408\uff0c\u5305\u542b45\uff0c311\u53e5\u5bf9\u3002\u5176\u4e2d\u8bad \u7ec3\u96c6\uff0c\u9a8c\u8bc1\u96c6\uff0c\u6d4b\u8bd5\u96c6\u5305\u542b\u6570\u636e\u91cf\u5206\u522b\u4e3a\uff1a36,254\uff0c4,528\uff0c4,529\u3002 RACE\u5019 \u5019 \u5019\u9009 \u9009 \u9009\u53e5 \u53e5 \u53e5\u6570 \u6570 \u6570\u636e \u636e \u636e\u96c6 \u96c6 \u96c6\uff1a\u672c\u6587\u91c7\u7528Wang et al. (2019)\u6807\u6ce8\u7684500\u9053RACE mid-challenge\u90e8\u5206 \u5019\u9009\u53e5\u5bf9\uff0c\u517134\uff0c736\u53e5\u5bf9\uff0c\u5176\u4e2d\u8bad\u7ec3\u96c6\uff0c\u9a8c\u8bc1\u96c6\uff0c\u6d4b\u8bd5\u96c6\u5206\u522b\u4e3a27,790\uff0c3,473\uff0c3,473 \u3002\u7531\u4e8e \u521d\uff0c\u9ad8\u4e2d\u8bd5\u9898\u96be\u5ea6\u6709\u6240\u533a\u522b\uff0c\u5728\u9a8c\u8bc1\u5019\u9009\u53e5\u62bd\u53d6\u5bf9\u7b54\u9898\u7684\u5f71\u54cd\u65f6\uff0c\u672c\u6587\u4ec5\u4f7f\u7528RACE\u6570\u636e\u96c6\u4e2d \u7684\u521d\u4e2d\u90e8\u5206\u8fdb\u884c\u6d4b\u8bd5\u3002 5.2 \u9605 \u9605 \u9605\u8bfb \u8bfb \u8bfb\u7406 \u7406 \u7406\u89e3 \u89e3 \u89e3\u591a \u591a \u591a\u9879 \u9879 \u9879\u9009 \u9009 \u9009\u62e9 \u62e9 \u62e9\u9898 \u9898 \u9898\u6570 \u6570 \u6570\u636e \u636e \u636e\u96c6 \u96c6 \u96c6 \u672c \u6587 \u91c7 \u7528RACE\u6570 \u636e \u96c6 \u4e2dmid-challenge\u90e8 \u5206 \u8fdb \u884c \u5b9e \u9a8c \uff0c \u5171 \u6536 \u96c618\uff0c364\u9053 \u95ee \u9898 \uff0c \u63098\uff1a1\uff1a1\u65b9\u5f0f\u5c06\u6570\u636e\u5212\u5206\u7ed9\u8bad\u7ec3\u3001\u9a8c\u8bc1\u3001\u548c\u6d4b\u8bd5\u96c6\uff1b\u6b64\u5916\u672c\u6587\u540c\u65f6\u6536\u96c6\u4e862005-2019\u5e74\u9ad8\u8003 \u8bed\u6587\u9605\u8bfb\u7406\u89e3\u9009\u62e9\u9898\u51717886\u9053\uff0c\u4e0eRACE\u91c7\u7528\u540c\u6837\u65b9\u5f0f\u5212\u5206\u3002 6 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u8bbe \u8bbe \u8bbe\u8ba1 \u8ba1 \u8ba1\u4e0e \u4e0e \u4e0e\u7ed3 \u7ed3 \u7ed3\u679c \u679c \u679c\u5206 \u5206 \u5206\u6790 \u6790 \u6790 6.1 \u6a21 \u6a21 \u6a21\u578b \u578b \u578b\u8bc4 \u8bc4 \u8bc4\u4ef7 \u4ef7 \u4ef7\u6307 \u6307 \u6307\u6807 \u6807 \u6807 \u5019 \u5019 \u5019\u9009 \u9009 \u9009\u53e5 \u53e5 \u53e5\u62bd \u62bd \u62bd\u53d6 \u53d6 \u53d6\u8bc4 \u8bc4 \u8bc4\u4ef7 \u4ef7 \u4ef7\u6307 \u6307 \u6307\u6807 \u6807 \u6807\uff1a \uff1a \uff1a \u5b9e\u9a8c\u91c7\u7528F1\u503c\u3001P(\u7cbe\u786e\u7387)\u3001R(\u53ec\u56de\u7387)\u6765\u8bc4\u4f30\u5019\u9009\u53e5\u62bd\u53d6\u6548\u679c\uff0c \u8ba1\u7b97\u516c\u5f0f\u5982\u4e0b\uff1a P = T P T P + F P \u00d7 100% P = T P T P + F N \u00d7 100% F 1 = 2 \u00d7 P \u00d7 R P + R \u00d7 100% (9) \u7b54 \u7b54 \u7b54\u9898 \u9898 \u9898\u6a21 \u6a21 \u6a21\u578b \u578b \u578b\u8bc4 \u8bc4 \u8bc4\u4ef7 \u4ef7 \u4ef7\u6307 \u6307 \u6307\u6807 \u6807 \u6807:\u5bf9\u4e8e\u7b54\u9898\u90e8\u5206\uff0c\u91c7\u7528accuracy\u4f5c\u4e3a\u6a21\u578b\u6027\u80fd\u8bc4\u4ef7\u6307\u6807\u3002 6.2 \u53c2 \u53c2 \u53c2\u6570 \u6570 \u6570\u8bbe \u8bbe \u8bbe\u7f6e \u7f6e \u7f6e \u9488\u5bf9\u4e0d\u540c\u6570\u636e\u96c6\u7684\u5b9e\u9a8c\u53c2\u6570\u8bbe\u7f6e\u5982\u88681\u6240\u793a\u3002 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c DataSet", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "A large annotated corpus for learning natural language inference", |
| "authors": [ |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Samuel", |
| "suffix": "" |
| }, |
| { |
| "first": "Gabor", |
| "middle": [], |
| "last": "Bowman", |
| "suffix": "" |
| }, |
| { |
| "first": "Christopher", |
| "middle": [], |
| "last": "Angeli", |
| "suffix": "" |
| }, |
| { |
| "first": "Christopher", |
| "middle": [ |
| "D" |
| ], |
| "last": "Potts", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Manning", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "632--642", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning. 2015. A large annotated corpus for learning natural language inference. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 632-642, Lisbon, Portugal, September. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Bert: Pre-training of deep bidirectional transformers for language understanding", |
| "authors": [ |
| { |
| "first": "Jacob", |
| "middle": [], |
| "last": "Devlin", |
| "suffix": "" |
| }, |
| { |
| "first": "Ming-Wei", |
| "middle": [], |
| "last": "Chang", |
| "suffix": "" |
| }, |
| { |
| "first": "Kenton", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| }, |
| { |
| "first": "Kristina", |
| "middle": [], |
| "last": "Toutanova", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Learning to search in long documents using document structure", |
| "authors": [ |
| { |
| "first": "Mor", |
| "middle": [], |
| "last": "Geva", |
| "suffix": "" |
| }, |
| { |
| "first": "Jonathan", |
| "middle": [], |
| "last": "Berant", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Proceedings of the 27th International Conference on Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "161--176", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Mor Geva and Jonathan Berant. 2018. Learning to search in long documents using document structure. In Proceedings of the 27th International Conference on Computational Linguistics, pages 161-176, Santa Fe, New Mexico, USA, August. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "An end-to-end model for question answering over knowledge base with cross-attention combining global knowledge", |
| "authors": [ |
| { |
| "first": "Yanchao", |
| "middle": [], |
| "last": "Hao", |
| "suffix": "" |
| }, |
| { |
| "first": "Yuanzhe", |
| "middle": [], |
| "last": "Zhang", |
| "suffix": "" |
| }, |
| { |
| "first": "Liu", |
| "middle": [], |
| "last": "Kang", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics", |
| "volume": "1", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Yanchao Hao, Yuanzhe Zhang, Liu Kang, Shizhu He, and Jun Zhao. 2017. An end-to-end model for question answering over knowledge base with cross-attention combining global knowledge. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Defining textual entailment", |
| "authors": [ |
| { |
| "first": "Daniel", |
| "middle": [ |
| "Z" |
| ], |
| "last": "Korman", |
| "suffix": "" |
| }, |
| { |
| "first": "Eric", |
| "middle": [], |
| "last": "Mack", |
| "suffix": "" |
| }, |
| { |
| "first": "Jacob", |
| "middle": [], |
| "last": "Jett", |
| "suffix": "" |
| }, |
| { |
| "first": "Allen", |
| "middle": [ |
| "H" |
| ], |
| "last": "Renear", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Journal of the Association for Information Science Technology", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Daniel Z. Korman, Eric Mack, Jacob Jett, and Allen H. Renear. 2018. Defining textual entailment. Journal of the Association for Information Science Technology.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Race: Large-scale reading comprehension dataset from examinations", |
| "authors": [ |
| { |
| "first": "Guokun", |
| "middle": [], |
| "last": "Lai", |
| "suffix": "" |
| }, |
| { |
| "first": "Qizhe", |
| "middle": [], |
| "last": "Xie", |
| "suffix": "" |
| }, |
| { |
| "first": "Hanxiao", |
| "middle": [], |
| "last": "Liu", |
| "suffix": "" |
| }, |
| { |
| "first": "Yiming", |
| "middle": [], |
| "last": "Yang", |
| "suffix": "" |
| }, |
| { |
| "first": "Eduard", |
| "middle": [], |
| "last": "Hovy", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Guokun Lai, Qizhe Xie, Hanxiao Liu, Yiming Yang, and Eduard Hovy. 2017. Race: Large-scale reading comprehension dataset from examinations.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Focal loss for dense object detection", |
| "authors": [ |
| { |
| "first": "Priya", |
| "middle": [], |
| "last": "Tsung Yi Lin", |
| "suffix": "" |
| }, |
| { |
| "first": "Ross", |
| "middle": [], |
| "last": "Goyal", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Girshick", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "2017 IEEE International Conference on Computer Vision (ICCV)", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Tsung Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollar. 2017. Focal loss for dense object detection. In 2017 IEEE International Conference on Computer Vision (ICCV).", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Neural network-based question answering over knowledge graphs on word and character level", |
| "authors": [ |
| { |
| "first": "Denis", |
| "middle": [], |
| "last": "Lukovnikov", |
| "suffix": "" |
| }, |
| { |
| "first": "Asja", |
| "middle": [], |
| "last": "Fischer", |
| "suffix": "" |
| }, |
| { |
| "first": "Jens", |
| "middle": [], |
| "last": "Lehmann", |
| "suffix": "" |
| }, |
| { |
| "first": "Sren", |
| "middle": [], |
| "last": "Auer", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "International World Wide Web Conference", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Denis Lukovnikov, Asja Fischer, Jens Lehmann, and Sren Auer. 2017. Neural network-based question answering over knowledge graphs on word and character level. In International World Wide Web Conference 2017.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Efficient and robust question answering from minimal context over documents", |
| "authors": [ |
| { |
| "first": "Sewon", |
| "middle": [], |
| "last": "Min", |
| "suffix": "" |
| }, |
| { |
| "first": "Victor", |
| "middle": [], |
| "last": "Zhong", |
| "suffix": "" |
| }, |
| { |
| "first": "Richard", |
| "middle": [], |
| "last": "Socher", |
| "suffix": "" |
| }, |
| { |
| "first": "Caiming", |
| "middle": [], |
| "last": "Xiong", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Sewon Min, Victor Zhong, Richard Socher, and Caiming Xiong. 2018. Efficient and robust question answering from minimal context over documents.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "A corpus and cloze evaluation for deeper understanding of commonsense stories", |
| "authors": [ |
| { |
| "first": "Nasrin", |
| "middle": [], |
| "last": "Mostafazadeh", |
| "suffix": "" |
| }, |
| { |
| "first": "Nathanael", |
| "middle": [], |
| "last": "Chambers", |
| "suffix": "" |
| }, |
| { |
| "first": "Xiaodong", |
| "middle": [], |
| "last": "He", |
| "suffix": "" |
| }, |
| { |
| "first": "Devi", |
| "middle": [], |
| "last": "Parikh", |
| "suffix": "" |
| }, |
| { |
| "first": "Dhruv", |
| "middle": [], |
| "last": "Batra", |
| "suffix": "" |
| }, |
| { |
| "first": "Lucy", |
| "middle": [], |
| "last": "Vanderwende", |
| "suffix": "" |
| }, |
| { |
| "first": "Pushmeet", |
| "middle": [], |
| "last": "Kohli", |
| "suffix": "" |
| }, |
| { |
| "first": "James", |
| "middle": [], |
| "last": "Allen", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", |
| "volume": "", |
| "issue": "", |
| "pages": "839--849", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Nasrin Mostafazadeh, Nathanael Chambers, Xiaodong He, Devi Parikh, Dhruv Batra, Lucy Vanderwende, Pushmeet Kohli, and James Allen. 2016. A corpus and cloze evaluation for deeper understanding of commonsense stories. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 839-849, San Diego, California, June. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Option comparison network for multiple-choice reading comprehension", |
| "authors": [ |
| { |
| "first": "Qiu", |
| "middle": [], |
| "last": "Ran", |
| "suffix": "" |
| }, |
| { |
| "first": "Peng", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "Weiwei", |
| "middle": [], |
| "last": "Hu", |
| "suffix": "" |
| }, |
| { |
| "first": "Jie", |
| "middle": [], |
| "last": "Zhou", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Qiu Ran, Peng Li, Weiwei Hu, and Jie Zhou. 2019. Option comparison network for multiple-choice reading comprehension.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "The probabilistic relevance framework: Bm25 and beyond", |
| "authors": [ |
| { |
| "first": "E", |
| "middle": [], |
| "last": "Stephen", |
| "suffix": "" |
| }, |
| { |
| "first": "Hugo", |
| "middle": [], |
| "last": "Robertson", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Zaragoza", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "Foundations Trends R in Information Retrieval", |
| "volume": "3", |
| "issue": "4", |
| "pages": "333--389", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Stephen E. Robertson and Hugo Zaragoza. 2009. The probabilistic relevance framework: Bm25 and beyond. Foundations Trends R in Information Retrieval, 3(4):333-389.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Conceptnet 5.5: An open multilingual graph of general knowledge", |
| "authors": [ |
| { |
| "first": "Robyn", |
| "middle": [], |
| "last": "Speer", |
| "suffix": "" |
| }, |
| { |
| "first": "Joshua", |
| "middle": [], |
| "last": "Chin", |
| "suffix": "" |
| }, |
| { |
| "first": "Catherine", |
| "middle": [], |
| "last": "Havasi", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Robyn Speer, Joshua Chin, and Catherine Havasi. 2016. Conceptnet 5.5: An open multilingual graph of general knowledge.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "Learning to rank answers on large online QA collections", |
| "authors": [ |
| { |
| "first": "Mihai", |
| "middle": [], |
| "last": "Surdeanu", |
| "suffix": "" |
| }, |
| { |
| "first": "Massimiliano", |
| "middle": [], |
| "last": "Ciaramita", |
| "suffix": "" |
| }, |
| { |
| "first": "Hugo", |
| "middle": [], |
| "last": "Zaragoza", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "Proceedings of ACL-08: HLT", |
| "volume": "", |
| "issue": "", |
| "pages": "719--727", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Mihai Surdeanu, Massimiliano Ciaramita, and Hugo Zaragoza. 2008. Learning to rank answers on large online QA collections. In Proceedings of ACL-08: HLT, pages 719-727, Columbus, Ohio, June. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "Multi-matching network for multiple choice reading comprehension", |
| "authors": [ |
| { |
| "first": "Min", |
| "middle": [], |
| "last": "Tang", |
| "suffix": "" |
| }, |
| { |
| "first": "Jiaran", |
| "middle": [], |
| "last": "Cai", |
| "suffix": "" |
| }, |
| { |
| "first": "Hankz", |
| "middle": [], |
| "last": "Zhuo", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Proceedings of the AAAI Conference on Artificial Intelligence", |
| "volume": "33", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Min Tang, Jiaran Cai, and Hankz Zhuo. 2019. Multi-matching network for multiple choice reading comprehension. Proceedings of the AAAI Conference on Artificial Intelligence, 33:7088-7095, 07.", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Repurposing entailment for multi-hop question answering tasks", |
| "authors": [ |
| { |
| "first": "Harsh", |
| "middle": [], |
| "last": "Trivedi", |
| "suffix": "" |
| }, |
| { |
| "first": "Heeyoung", |
| "middle": [], |
| "last": "Kwon", |
| "suffix": "" |
| }, |
| { |
| "first": "Tushar", |
| "middle": [], |
| "last": "Khot", |
| "suffix": "" |
| }, |
| { |
| "first": "Ashish", |
| "middle": [], |
| "last": "Sabharwal", |
| "suffix": "" |
| }, |
| { |
| "first": "Niranjan", |
| "middle": [], |
| "last": "Balasubramanian", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Harsh Trivedi, Heeyoung Kwon, Tushar Khot, Ashish Sabharwal, and Niranjan Balasubramanian. 2019. Repurposing entailment for multi-hop question answering tasks.", |
| "links": null |
| }, |
| "BIBREF16": { |
| "ref_id": "b16", |
| "title": "A co-matching model for multi-choice reading comprehension", |
| "authors": [ |
| { |
| "first": "Shuohang", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "Mo", |
| "middle": [], |
| "last": "Yu", |
| "suffix": "" |
| }, |
| { |
| "first": "Jing", |
| "middle": [], |
| "last": "Jiang", |
| "suffix": "" |
| }, |
| { |
| "first": "Shiyu", |
| "middle": [], |
| "last": "Chang", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics", |
| "volume": "2", |
| "issue": "", |
| "pages": "746--751", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Shuohang Wang, Mo Yu, Jing Jiang, and Shiyu Chang. 2018. A co-matching model for multi-choice read- ing comprehension. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 746-751, Melbourne, Australia, July. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF17": { |
| "ref_id": "b17", |
| "title": "Evidence sentence extraction for machine reading comprehension", |
| "authors": [ |
| { |
| "first": "Hai", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "Dian", |
| "middle": [], |
| "last": "Yu", |
| "suffix": "" |
| }, |
| { |
| "first": "Kai", |
| "middle": [], |
| "last": "Sun", |
| "suffix": "" |
| }, |
| { |
| "first": "Jianshu", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "Dan", |
| "middle": [], |
| "last": "Roth", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL)", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hai Wang, Dian Yu, Kai Sun, Jianshu Chen, and Dan Roth. 2019. Evidence sentence extraction for machine reading comprehension. In Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL).", |
| "links": null |
| }, |
| "BIBREF18": { |
| "ref_id": "b18", |
| "title": "A broad-coverage challenge corpus for sentence understanding through inference", |
| "authors": [ |
| { |
| "first": "Adina", |
| "middle": [], |
| "last": "Williams", |
| "suffix": "" |
| }, |
| { |
| "first": "Nikita", |
| "middle": [], |
| "last": "Nangia", |
| "suffix": "" |
| }, |
| { |
| "first": "Samuel", |
| "middle": [], |
| "last": "Bowman", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", |
| "volume": "1", |
| "issue": "", |
| "pages": "1112--1122", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Adina Williams, Nikita Nangia, and Samuel Bowman. 2018. A broad-coverage challenge corpus for sen- tence understanding through inference. In Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 1112-1122, New Orleans, Louisiana, June. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF19": { |
| "ref_id": "b19", |
| "title": "Quick and (not so) dirty: Unsupervised selection of justification sentences for multi-hop question answering", |
| "authors": [ |
| { |
| "first": "Vikas", |
| "middle": [], |
| "last": "Yadav", |
| "suffix": "" |
| }, |
| { |
| "first": "Steven", |
| "middle": [], |
| "last": "Bethard", |
| "suffix": "" |
| }, |
| { |
| "first": "Mihai", |
| "middle": [], |
| "last": "Surdeanu", |
| "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", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Vikas Yadav, Steven Bethard, and Mihai Surdeanu. 2019. Quick and (not so) dirty: Unsupervised selec- tion of justification sentences for multi-hop question answering. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP).", |
| "links": null |
| }, |
| "BIBREF20": { |
| "ref_id": "b20", |
| "title": "Dual co-matching network for multi-choice reading comprehension", |
| "authors": [ |
| { |
| "first": "Shuailiang", |
| "middle": [], |
| "last": "Zhang", |
| "suffix": "" |
| }, |
| { |
| "first": "Hai", |
| "middle": [], |
| "last": "Zhao", |
| "suffix": "" |
| }, |
| { |
| "first": "Yuwei", |
| "middle": [], |
| "last": "Wu", |
| "suffix": "" |
| }, |
| { |
| "first": "Zhuosheng", |
| "middle": [], |
| "last": "Zhang", |
| "suffix": "" |
| }, |
| { |
| "first": "Xi", |
| "middle": [], |
| "last": "Zhou", |
| "suffix": "" |
| }, |
| { |
| "first": "Xiang", |
| "middle": [], |
| "last": "Zhou", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Shuailiang Zhang, Hai Zhao, Yuwei Wu, Zhuosheng Zhang, Xi Zhou, and Xiang Zhou. 2019. Dual co-matching network for multi-choice reading comprehension.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "TABREF2": { |
| "content": "<table><tr><td colspan=\"11\">\u7840\u8fdb\u884c\u6539\u8fdb\uff0c\u7ed3\u679c\u8868\u660e\u7ed3\u5408RFTR(\u672c\u6587\u65b9\u6cd5)\u540e\uff0c\u6a21\u578b\u6548\u679c\u5728P\u503c\u4e0a\u63d0\u53475.41\u4e2a\u767e\u5206\u70b9\uff0cR\u503c</td></tr><tr><td colspan=\"11\">\u63d0\u53471.99\u4e2a\u767e\u5206\u70b9\uff0cF1\u503c3.44\u4e2a\u767e\u5206\u70b9\u3002\u5bf9\u4e8eRACE\u6570\u636e\u96c6\u57fa\u7ebf\u6a21\u578b\u4e2dBERT-wwm\u53d6\u5f97\u6700\u4f18\u6548</td></tr><tr><td colspan=\"11\">\u679c\uff0c\u6545\u5728\u6b64\u57fa\u7840\u4e0a\u7ed3\u5408RFTR\u540e\uff0c\u6548\u679c\u63d0\u5347\u4e863.95\u4e2a\u767e\u5206\u70b9\u3002\u4ee5\u4e0a\u6240\u8ff0\u9a8c\u8bc1\u4e86\u6240\u63d0\u65b9\u6cd5\u7684\u4f18\u8d8a</td></tr><tr><td>\u6027\u3002</td><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td/><td>\u9ad8\u8003</td><td/><td/><td/><td colspan=\"3\">RACE</td><td/></tr><tr><td>Baseline Model</td><td/><td colspan=\"9\">P(%) R(%) F1(%) P(%) R(%) F1(%)</td></tr><tr><td>ALBERT-base</td><td/><td>81.50 46.13</td><td colspan=\"2\">58.91</td><td/><td colspan=\"3\">76.34 59.08</td><td colspan=\"2\">66.61</td></tr><tr><td>BERT-base</td><td/><td>73.59 61.84</td><td colspan=\"2\">67.21</td><td/><td colspan=\"3\">77.30 61.10</td><td colspan=\"2\">68.25</td></tr><tr><td>BERT-wwm</td><td/><td>73.78 63.84</td><td colspan=\"2\">68.45</td><td/><td colspan=\"3\">78.03 60.78</td><td colspan=\"2\">68.33</td></tr><tr><td colspan=\"3\">BERT-wwm+OR BERT-wwm+OR+FL 76.94 65.81 77.29 65.34 \u56fe 4. \u9012\u5f52\u62bd\u53d6\u4e2dK\u503c\u53d8\u5316</td><td colspan=\"2\">70.81 70.94</td><td colspan=\"6\">83.10 63.77 82.58 64.10 \u56fe 5. \u7b5b\u9009\u6a21\u5757\u4e2dm\u503c\u53d8\u5316 72.16 72.18</td></tr><tr><td colspan=\"11\">RFTR-ROCC \u5931\uff1b\u800c\u5f53\u8df3\u6570\u4e3a4\u65f6\u53ec\u56de\u7387\u4e0b\u964d\uff0c\u8bf4\u660e\u8df3\u6570\u8fc7\u591a\u4e5f\u4f1a\u5f15\u5165\u4e00\u5b9a\u7684\u566a\u58f0\u3002\u7531\u56fe5\u53ef\u77e5\uff0c\u5728\u9ad8\u8003\u6570\u636e 76.86 65.87 70.94 81.40 64.85 72.19</td></tr><tr><td colspan=\"11\">RFTR \u53caRACE\u4e2dm\u503c\u5206\u522b\u4e3a4\u548c2\u65f6\u7cbe\u786e\u7387\u8fbe\u5230\u6700\u4f18\uff0c\u8bf4\u660eROCC\u53ef\u6709\u6548\u7b5b\u9009\u5197\u4f59\u4fe1\u606f\uff0c\u6700\u5927\u9650\u5ea6\u5730 79.19 65.83 71.89 84.36 63.23 72.28</td></tr><tr><td colspan=\"8\">\u6574\u5408\u76f8\u5173\u4fe1\u606f,\u4ece\u800c\u5254\u9664\u4e00\u90e8\u5206\u65e0\u5173\u53e5\u6216\u5197\u4f59\u53e5\u3002 \u8868 2. \u57fa\u4e8e\u9ad8\u8003\u8bed\u6587\u548cRACE\u7684\u5019\u9009\u53e5\u62bd\u53d6\u7ed3\u679c 6.3.5 \u9519 \u9519 \u9519\u8bef \u8bef \u8bef\u62bd \u62bd \u62bd\u53d6 \u53d6 \u53d6\u793a \u793a \u793a\u4f8b \u4f8b \u4f8b\u5206 \u5206 \u5206\u6790 \u6790 \u6790</td><td/><td/></tr><tr><td colspan=\"10\">6.3.2 \u5019 \u5019 \u5019\u9009 \u9009 \u9009\u53e5 \u53e5 \u53e5\u62bd \u62bd \u62bd\u53d6 \u53d6 \u53d6\u6d88 \u6d88 \u6d88\u878d \u878d \u878d\u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c \u672c\u6587\u9009\u53d6\u4e86\u6d4b\u8bd5\u96c6\u4e2d50\u6761\u9519\u8bef\u6570\u636e\u8fdb\u884c\u4e86\u5206\u6790\uff0c\u88684\u4e3a\u5217\u4e3e\u7684\u9519\u8bef\u6570\u636e\u3002</td></tr><tr><td colspan=\"11\">\u4e3a\u8fdb\u4e00\u6b65\u7814\u7a76\u6240\u63d0\u65b9\u6848\u5bf9\u5b9e\u9a8c\u7ed3\u679c\u7684\u5f71\u54cd\uff0c\u5728\u9ad8\u8003\u53caRACE\u5019\u9009\u53e5\u6570\u636e\u96c6\u4e0a\u5206\u522b\u8fdb\u884c \u5019\u9009\u53e5 \u9009\u9879 \u9884\u6d4b\u7ed3\u679c \u771f\u5b9e\u7ed3\u679c</td></tr><tr><td colspan=\"11\">\u4e86\u6d88\u878d\u5b9e\u9a8c\u3002\u5982\u88682\u6240\u793a,\u6539\u5199\u9009\u9879\u540e(\u5373\u8868\u4e2dOR)\u5728\u4e24\u6570\u636e\u96c6\u4e0a\u6a21\u578bF1\u503c\u76f8\u6bd4\u57fa\u7ebf\u5206\u522b\u63d0 \u8709 \u8709 \u8709\u8763 \u8763 \u8763\u8fd9\u79cd\u751f\u7269\u5927\u591a\u6570\u65f6\u95f4\u751f\u6d3b\u5728\u6c34\u91cc\uff0c\u4ee5\u85fb\u7c7b \u8709 \u8709 \u8709\u8763 \u8763 \u8763\u6709 \u6709 \u6709\u7fc5 \u7fc5 \u7fc5\u540e\u5373\u5347\u7a7a\u98de\u884c\u3002\u867d\u7136\u98de\u884c\u65f6\u95f4\u4e0d\u957f\uff0c 0 1</td></tr><tr><td colspan=\"11\">\u53472.36\u53ca3.83\u4e2a\u767e\u5206\u70b9\uff0c\u5e76\u4e14P\u503c\u548cR\u503c\u4e5f\u5747\u6709\u63d0\u5347\uff0c\u8868\u660e\u6539\u5199\u9009\u9879\u4f7f\u4fe1\u606f\u66f4\u5b8c\u6574\uff0c\u8bed\u4e49\u66f4\u901a \u4e3a\u98df\uff0c\u5f53\u5b83 \u5b83 \u5b83\u4eec \u4eec \u4eec\u51c6\u5907\u597d\u7e41 \u7e41 \u7e41\u6b96 \u6b96 \u6b96\uff0c\u4fbf\u722c\u51fa\u6c34\u9762\uff0c\u5728\u6c34 \u4f46\u7531\u6b64\u5b9e\u73b0\u4e86\u751f \u751f \u751f\u547d \u547d \u547d\u7684 \u7684 \u7684\u5ef6 \u5ef6 \u5ef6\u7eed \u7eed \u7eed\u3002</td></tr><tr><td colspan=\"11\">\u987a\uff0c\u8fd9\u5bf9\u6a21\u578b\u7684\u8bed\u4e49\u5b66\u4e60\u6709\u5f88\u5927\u5e2e\u52a9\uff1b\u4e4b\u540e\u9488\u5bf9\u6570\u636e\u96c6\u4e2d\u6b63\u8d1f\u6837\u672c\u4e0d\u5747\u8861\u73b0\u8c61\uff0c\u4f7f\u7528Focal \u8fb9\u7684\u690d\u7269\u4e0a\u8715\u76ae\uff0c\u6210\u4e3a\u6709 \u6709 \u6709\u7fc5 \u7fc5 \u7fc5\u7684 \u7684 \u7684\u6210 \u6210 \u6210\u866b \u866b \u866b\u3002</td></tr><tr><td colspan=\"11\">Loss\u8fdb\u4e00\u6b65\u4f7f\u6a21\u578b\u6548\u679c\u5728F1\u503c\u4e0a\u5206\u522b\u63d0\u53470.13\u548c0.02\u4e2a\u767e\u5206\u70b9\uff0c\u5176\u4e2d\u5bf9\u4e8e\u9ad8\u8003R\u503c\u63d0\u53470.47\u4e2a\u767e 0 1</td></tr><tr><td colspan=\"11\">\u5206\u70b9\uff0c\u8868\u660e\u66f4\u6362\u635f\u5931\u51fd\u6570\u540e\uff0c\u6a21\u578b\u5bf9\u6b63\u6837\u672c\u5b66\u4e60\u7684\u504f\u5411\u6027\u589e\u5f3a\uff1b\u4f7f\u7528TF-IDF\u76f8\u4f3c\u5ea6\u8ba1\u7b97\u62bd\u53d6\u9700</td></tr><tr><td colspan=\"11\">\u591a\u6b65\u63a8\u7406\u95ee\u9898\u7684\u5019\u9009\u53e5\u4f7f\u6a21\u578b\u53ec\u56de\u7387\u5206\u522b\u63d0\u5347\u4e860.06\u548c0.75\uff0c\u8868\u660e\u8be5\u65b9\u6848\u53ef\u4ee5\u6709\u6548\u7f13\u89e3\u591a\u6b65\u63a8\u7406</td></tr><tr><td colspan=\"11\">\u95ee\u9898\u7684\u4fe1\u606f\u635f\u5931\uff1b\u6700\u540e\uff0c\u4f7f\u7528ROCC\u4ece\u5019\u9009\u53e5\u96c6\u4e4b\u95f4\u7684\u5197\u4f59\u5ea6\uff0c\u5019\u9009\u53e5\u96c6\u5bf9\u9009\u9879\u4fe1\u606f\u8986\u76d6\u7387\u548c \u4ece\u5370\u5237\u7684\u57fa\u672c\u9700\u6c42\u6765\u770b\uff0c\u6392\u5b57\u673a\u7684\u5b57\u5e93\u901a\u5e38\u8981 \u5019\u9009\u53e5\u4e0e\u9009\u9879\u76f8\u5173\u6027\u4e09\u65b9\u9762\u8003\u8651\uff0c\u8fdb\u4e00\u6b65\u5bf9\u7ed3\u679c\u8fdb\u884c\u7b5b\u9009\uff0c\u5728\u4e24\u6570\u636e\u96c6\u4e0aP\u503c\u5206\u522b\u63d0\u53472.33\u4e2a\u767e \u65367000\u591a\u5b57\u3002\u800c\u4ece\u4e00\u822c\u4e66\u62a5\u7684\u9700\u6c42\u6765\u8bf4\uff0c\u5b57 \u5b57 \u5b57\u4f53 \u4f53 \u4f53</td></tr><tr><td>\u5206\u70b9\u548c2.96\u4e2a\u767e\u5206\u70b9\u3002 \u5c31 \u5c31</td><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>6.3.3 \u5019 \u5019 \u5019\u9009 \u9009 \u9009\u53e5 \u53e5 \u53e5\u62bd \u62bd \u62bd\u53d6 \u53d6 \u53d6\u6548 \u6548 \u6548\u679c \u679c \u679c\u9a8c \u9a8c \u9a8c\u8bc1 \u8bc1 \u8bc1</td><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"11\">\u672c\u6587\u5728\u9ad8\u8003\u9605\u8bfb\u7406\u89e3\u9009\u62e9\u9898\u4e0eRACE\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u9a8c\u8bc1\uff0c\u5c06\u62bd\u51fa\u7684\u5019\u9009\u53e5\u62fc\u63a5\u4f5c\u4e3a\u65b0\u6587\u7ae0</td></tr><tr><td colspan=\"11\">\u8f93\u5165\u6a21\u578b\uff0c\u6548\u679c\u5982\u88683\u6240\u793a\uff0c\u5176\u4e2dEV(RFTR)\u8868\u793a\u4f7f\u7528\u5019\u9009\u53e5\u4f5c\u4e3a\u6587\u7ae0\u7684\u65b9\u6cd5\uff0c\u8be5\u5b9e\u9a8c\u7ed3\u679c\u8bc1\u660e</td></tr><tr><td>\u4e86\u5019\u9009\u53e5\u62bd\u53d6\u7684\u6709\u6548\u6027\u3002</td><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td/><td colspan=\"2\">\u9ad8 \u9ad8 \u9ad8\u8003 \u8003 \u8003</td><td/><td/><td colspan=\"3\">RACE</td><td/></tr><tr><td>\u6a21 \u6a21 \u6a21\u578b \u578b \u578b</td><td/><td>dev(%)</td><td colspan=\"2\">test(%)</td><td/><td colspan=\"2\">dev(%)</td><td colspan=\"3\">test(%)</td></tr><tr><td>BERT(base)</td><td/><td>32.61</td><td>30.33</td><td/><td/><td>55.91</td><td/><td colspan=\"2\">57.66</td></tr><tr><td>BERT+EV(RFTR)</td><td/><td>36.29</td><td>31.34</td><td/><td/><td>59.51</td><td/><td colspan=\"2\">59.87</td></tr><tr><td>co-matching</td><td/><td>35.27</td><td>32.36</td><td/><td/><td>47.54</td><td/><td colspan=\"2\">42.22</td></tr><tr><td colspan=\"11\">epoch max length batch learning rate K m co-matching+EV(RFTR) 36.29 35.39 50.65 46.40</td></tr><tr><td>\u9ad8\u8003 RACE ALBERT \u5019\u9009\u53e5\u62bd\u53d6 ALBERT+EV(RFTR)</td><td>3 3</td><td>128 29.06 128 32.11</td><td>28.05 29.57</td><td colspan=\"2\">32 32</td><td>65.26 68.92</td><td>3e-5 3e-5</td><td colspan=\"2\">67.08 69.30</td><td>3 3</td><td>4 2</td></tr><tr><td>\u9ad8\u8003 RACE DCMN+EV(RFTR) DCMN \u7b54\u9898\u6a21\u578b</td><td>6 3</td><td>450 30.83 320 32.64</td><td>30.24 32.12</td><td colspan=\"2\">40 32</td><td>48.16 50.53</td><td>1e-5 5e-5</td><td colspan=\"2\">49.53 52.19</td><td>----</td></tr><tr><td colspan=\"9\">\u8868 1. \u6a21\u578b\u53c2\u6570\u8bbe\u7f6e \u8868 3. \u9ad8\u8003\u8bed\u6587\u4e0eRACE\u6570\u636e\u96c6\u7b54\u9898\u6a21\u578b\u5bf9\u6bd4\u7ed3\u679c</td><td/></tr><tr><td/><td/><td>\u54cd \u54cd</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"11\">\u8f83 \u8f83 \u4e3a\u6bd4\u8f83\u5b9e\u9a8c\u4e2dTF-IDF\u9012\u5f52\u62bd\u53d6\u6a21\u5757(\u89c13.3\u8282)\u4e2dK\u503c\u53ca\u5019\u9009\u53e5\u7b5b\u9009\u6a21\u5757(\u89c13.4\u8282)\u5168\u7ec4\u5408</td></tr><tr><td colspan=\"11\">\u88682\u4e2d\u5c55\u793a\u5404\u6a21\u578b\u5728\u9ad8\u8003\u53caRACE\u5019\u9009\u53e5\u6570\u636e\u96c6\u4e0a\u7684\u6548\u679c\uff0c\u5bf9\u4e8e\u9ad8\u8003\u5019\u9009\u53e5\u6570\u636e\u96c6\uff0c\u4ece\u8868 \u4e2dm\u503c\u5bf9\u5019\u9009\u53e5\u62bd\u53d6\u7684\u5f71\u54cd\uff0c\u672c\u6587\u8fdb\u884c\u4e86\u53c2\u6570\u5bf9\u6bd4\u5b9e\u9a8c\u3002\u5b9e\u9a8c\u7ed3\u679c\u5982\u56fe4,5\u6240\u793a\uff1a</td></tr><tr><td colspan=\"11\">\u4e2d\u53ef\u770b\u51faBERTwwm\u7684P\u503c\uff0cR\u503c\u53caF1\u503c\u5747\u9ad8\u4e8eBERT-base\uff1bALBERT-base\u62bd\u53d6\u5019\u9009\u53e5\u867dP\u503c \u7531\u56fe4\u53ef\u77e5\uff0c\u5728\u9ad8\u8003\u6570\u636e\u53caRACE\u6570\u636e\u96c6\u4e2d\uff0c\u968f\u7740\u8df3\u6570\u7684\u589e\u52a0\uff0c\u5019\u9009\u53e5\u7684\u53ec\u56de\u7387\u9010\u6e10\u63d0</td></tr><tr><td colspan=\"11\">\u8f83\u9ad8\uff0c\u4f46R\u503c\u76f8\u6bd4\u4e8eBERTwwm\u4f4e17.71\u4e2a\u767e\u5206\u70b9\u3002\u6545\u9488\u5bf9\u9ad8\u8003\u6570\u636e\u96c6\uff0c\u672c\u6587\u4ee5BERTwwm\u4e3a\u57fa \u9ad8\uff0c\u5f53K\u4e3a3\u65f6\u53ec\u56de\u7387\u4e0eF1\u503c\u8fbe\u5230\u6700\u4f18\uff0c\u8868\u660e\u5f53\u8df3\u6570\u4e3a3\u65f6\u53ef\u6709\u6548\u7f13\u89e3\u591a\u6b65\u63a8\u7406\u95ee\u9898\u7684\u4fe1\u606f\u635f</td></tr></table>", |
| "type_str": "table", |
| "html": null, |
| "num": null, |
| "text": "\u65e0\u8bba\u6211\u4eec\u5982\u4f55\u770b\u5f85\u9c81\u8fc5\uff0c\u5982\u4f55\u8bc4\u4ef7\u9c81\u8fc5\u5148\u751f\u7684 \u6bd5\u751f\u4e4b\u95ee\u548c\u4ed6\u4e3a\u6b64\u6240\u505a\u7684\u4e00\u5207\uff0c\u73b0 \u73b0 \u73b0\u5728 \u5728 \u5728\uff0c \uff0c \uff0c\u6211 \u6211 \u6211\u4eec \u4eec \u4eec\u90fd \u90fd \u90fd \u4f9d \u4f9d \u4f9d\u7136 \u7136 \u7136\u5f97 \u5f97 \u5f97\u548c \u548c \u548c\u4ed6 \u4ed6 \u4ed6\u4e00 \u4e00 \u4e00\u8d77 \u8d77 \u8d77,\u627f \u627f \u627f\u53d7 \u53d7 \u53d7\u4e00 \u4e00 \u4e00\u4e2a \u4e2a \u4e2a\u5404 \u5404 \u5404\u4eba \u4eba \u4eba\u5fc3 \u5fc3 \u5fc3\u5e95 \u5e95 \u5e95\u8bda \u8bda \u8bda\u4fe1 \u4fe1 \u4fe1\u4e0e \u4e0e \u4e0e\u7231 \u7231 \u7231 \u90fd \u90fd \u90fd\u5c1a \u5c1a \u5c1a\u6709 \u6709 \u6709\u4e0d \u4e0d \u4e0d\u8db3 \u8db3 \u8db3\u7684 \u7684 \u7684\u65f6 \u65f6 \u65f6\u4ee3 \u4ee3 \u4ee3\u3002 \u9c81 \u9c81 \u9c81\u8fc5 \u8fc5 \u8fc5\u7684\u65f6\u4ee3\u8fc7\u53bb\u4e86\uff0c\u4f46 \u4f46 \u4f46\u90a3 \u90a3 \u90a3\u4e2a \u4e2a \u4e2a\u65f6 \u65f6 \u65f6\u4ee3 \u4ee3 \u4ee3\u7684 \u7684 \u7684\u56fd \u56fd \u56fd\u6c11 \u6c11 \u6c11\u52a3 \u52a3 \u52a3\u6839 \u6839 \u6839\u6027 \u6027 \u6027 \u4eca \u4eca \u4eca\u5929 \u5929 \u5929\u4f9d \u4f9d \u4f9d\u7136 \u7136 \u7136\u5b58 \u5b58 \u5b58\u5728 \u5728 \u5728\uff0c\u4e3a\u6b64\u6211\u4eec\u8981\u547c\u5524\u9c81\u8fc5\uff0c\u4e0d \u4e0d \u4e0d\u8981 \u8981 \u8981\u6f20 \u6f20 \u6f20 \u89c6 \u89c6 \u89c6\u9c81 \u9c81 \u9c81\u8fc5 \u8fc5 \u8fc5\u7684 \u7684 \u7684\u5b58 \u5b58 \u5b58\u5728 \u5728 \u5728\u3002 \u5c31\u6709 \u6709 \u6709\u4e66 \u4e66 \u4e66\u7248 \u7248 \u7248\u5b8b \u5b8b \u5b8b\u3001 \u3001 \u3001\u62a5 \u62a5 \u62a5\u7248 \u7248 \u7248\u5b8b \u5b8b \u5b8b\u3001 \u3001 \u3001\u6807 \u6807 \u6807\u9898 \u9898 \u9898\u5b8b \u5b8b \u5b8b\u3001 \u3001 \u3001\u4eff \u4eff \u4eff\u5b8b \u5b8b \u5b8b\u3001 \u3001 \u3001\u6977 \u6977 \u6977\u4f53 \u4f53 \u4f53\u3001 \u3001 \u3001 \u9ed1 \u9ed1 \u9ed1\u4f53 \u4f53 \u4f53...\u7b49 \u7b49 \u7b49\u5341 \u5341 \u5341\u591a \u591a \u591a\u79cd \u79cd \u79cd\u3002 \u56e0\u5b57 \u5b57 \u5b57\u5f62 \u5f62 \u5f62\u5b57 \u5b57 \u5b57\u4f53 \u4f53 \u4f53\u7684\u5236\u7ea6\uff0c\u6c49\u5b57\u6392\u7248\u7e41\u590d\u3002 0 1" |
| } |
| } |
| } |
| } |
