基于问题分解的多跳机器阅读理解模型

计算机工程与科学 ›› 2022, Vol. 44 ›› Issue (08): 1506-1513.

基于问题分解的多跳机器阅读理解模型

周展朝1,2，刘茂福1,2，胡慧君1,2

（1.武汉科技大学计算机科学与技术学院，湖北武汉 430065;
2.智能信息处理与实时工业系统湖北省重点实验室，湖北武汉 430065）

收稿日期:2020-07-28 修回日期:2021-01-04 接受日期:2022-08-25 出版日期:2022-08-25 发布日期:2022-08-25
基金资助:
国家社会科学基金（11&ZD189）；湖北省教育厅人文社会科学研究项目（17Y018）

A multi-hop reading comprehension method based on question decomposition

ZHOU Zhan-zhao1,2,LIU Mao-fu1,2,HU Hui-jun1,2

(1.School of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065;
2.Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System,Wuhan 430065,China)

Received:2020-07-28 Revised:2021-01-04 Accepted:2022-08-25 Online:2022-08-25 Published:2022-08-25

摘要/Abstract

摘要： 多跳机器阅读理解是自然语言处理领域最困难的任务之一，需要在多个段落之间进行推理。多跳机器阅读理解任务中的复杂问题一般由多个简单问题融合而成，可以通过分解复杂问题使模型更好地理解问题。因此，针对复杂多跳问题，提出了一种基于问题分解的多跳阅读理解模型。该模型首先将多跳问题分解为多个单跳问题，然后利用单跳阅读理解模型对其进行求解。将问题分解视作一个阅读理解任务：多跳问题是问题分解的上下文，而包含问题答案的证据段落则是问题。阅读理解任务捕捉了多跳问题和证据段落之间的交互语义信息，可以指导多跳问题中单跳问题的抽取。所提模型的BLEU值和Rouge-L值分别为71.48%和79.29%。实验结果表明,该模型对多跳机器阅读理解是有效的。

关键词: 机器阅读理解, 推理, 多跳问题, 问题分解

Abstract: Multi-hop machine reading comprehension is one of the most difficult tasks in the field of natural language processing, which requires reasoning between multiple paragraphs. The complex question in multi-hop machine reading comprehension task is usually composed of several simple questions and decomposing the complex question can make the model better understand the question itself. Therefore, for the complex multi-hop question, a multi-hop reading comprehension model based on question decomposition is proposed. The multi-hop question is first decomposed into several single-hop questions, and then the single-hop reading comprehension model is used to solve them. The question decomposition is regarded as a reading comprehension task: the multi-hop question is the context of the question decomposition, while the evidence paragraph containing the answer to the question is the question. Machine reading comprehension task captures the interactive semantic information between multi-hop question and evidence paragraph, which can guide the extraction of single-hop questions in multi-hop questions. The values of the proposed model on BLEU and Rouge-L are 71.48% and 79.29%, respectively. Experimental results show that this model is effective for multi-hop machine reading comprehension.

Key words: machine reading comprehension, reasoning, multi-hop question, question decomposition ,

周展朝, 刘茂福, 胡慧君, . 基于问题分解的多跳机器阅读理解模型[J]. 计算机工程与科学, 2022, 44(08): 1506-1513.

ZHOU Zhan-zhao, LIU Mao-fu, HU Hui-jun, . A multi-hop reading comprehension method based on question decomposition[J]. Computer Engineering & Science, 2022, 44(08): 1506-1513.

参考文献［19］

［1］	Talmor A,Berant J.The web as a knowledge-base for answering complex questions［J］.arXiv:1803.06643,2018.
［2］	Min S,Zhong V,Zettlemoyer L,et al.Multi-hop reading comprehension through question decomposition and rescoring［C］∥Proc of the 57th Annual Meeting of the Association for Computational Linguistics,2019:6097-6109.
［3］	Rajpurkar P,Zhang J,Lopyrev K,et al.SQuAD:100 000+questions for machine comprehension of text［J］.arXiv:1606.05250,2016.
［4］	Wang S,Jiang J.Machine comprehension using Match-LSTM and answer pointer［C］∥Proc of the 5th International Conference on Learning Representations,2017:1351-1362.
［5］	Seo M,Kembhavi A,Farhadi A,et al.Bi-directional attention flow for machine comprehension［C］∥Proc of the 5th International Conference on Learning Representations,2017:1370-1383.
［6］	Yu A W,Dohan D,Luong M T,et al.QANet:Combining local convolution with global self-attention for reading comprehension［C］∥Proc of the 6th International Conference on Learning Representations,2018:1220-1236.
［7］	Wang Y,Liu K,Liu J,et al.Multi-passage machine reading comprehension with cross-passage answer verification［C］∥Proc of the 56th Annual Meeting of the Association for Computational Linguistics,2018:1870-1879.
［8］	Devlin J,Chang M W,Lee K,et al.BERT:Pre-training of deep bidirectional transformers for language understanding［J］.arXiv:1810.04805,2018.
［9］	Sun Y,Wang S,Li Y,et al.ERNIE:Enhanced representation through knowledge integration［J］.arXiv:1904.09223,2019.
［10］	Ding M,Zhou C,Chen Q,et al.Cognitive graph for multi-hop reading comprehension at scale［C］∥Proc of the 57th Annual Meeting of the Association for Computational Linguistics,2019:2694-2703.
［11］	Qiu L,Xiao Y,Qu Y,et al.Dynamically fused graph network for multi-hop reasoning［C］∥Proc of the 57th Annual Meeting of the Association for Computational Linguistics,2019:6140-6150.
［12］	Kundu S,Khot T,Sabharwal A,et al.Exploiting explicit paths for multi-hop reading comprehension［C］∥Proc of the 57th Annual Meeting of the Association for Computational Linguistics,2019:2737-2747.
［13］	McCann B,Keskar N S,Xiong C,et al.The natural language decathlon:Multitask learning as question answering［J］.arXiv preprint arXiv:1806.08730,2018.
［14］	Levy O,Seo M,Choi E,et al.Zero-shot relation extraction via reading comprehension［C］∥Proc of the 21st Conference on Computational Natural Language Learning,2017:333-342.
［15］	Li X,Feng J,Meng Y,et al.A unified MRC framework for named entity recognition［C］∥Proc of the 58th Annual Meeting of the Association for Computational Linguistics,2020:5849-5859.
［16］	Li X,Yin F,Sun Z,et al.Entity-relation extraction as multi-turn question answering［C］∥Proc of the 57th Annual Meeting of the Association for Computational Linguistics,2019:1340-1350.
［17］	Wu W,Wang F,Yuan A,et al.Coreference resolution as query-based span prediction［C］∥Proc of the 58th Annual Meeting of the Association for Computational Linguistics,2020:6953-6963.
［18］	Tan Hong-ye, Li Xuan-ying,Liu Bei.Reading comprehension based on external knowledge and hierarchical discourse representation ［J］.Journal of Chinese Information Proces- sing,2020,34(4):85-91.(in Chinese)
［19］	Zhang Dong,Li Shou-shan,Wang Jing-jing.Semi-supervised question classification with jointly learning question and answer representations ［J］.Journal of Chinese Information Processing,2017,31(1):1-7.(in Chinese)
［20］	Chen D,Fisch A,Weston J,et al.Reading wikipedia to answer open-domain questions［C］∥Proc of the 55th Annual Meeting of the Association for Computational Linguistics,2017:1870-1879.
［21］	Vaswani A,Shazeer N,Parmar N,et al.Attention is all you need［C］∥Proc of the 31st International Conference on Neural Information Processing Systems,2017:5998-6008.
［22］	Yang Z,Qi P,Zhang S,et al.HotpotQA:A dataset for diverse,explainable multi-hop question answering［C］∥Proc of the 2018 Conference on Empirical Methods in Natural Language Processing,2018:2369-2380.
［23］	Papineni K,Roukos S,Ward T,et al.BLEU:A method for automatic evaluation of machine translation［C］∥Proc of the 40th Annual Meeting of the Association for Computational Linguistics,2002:311-318.
［24］	Lin C Y.ROUGE:A package for automatic evaluation of summaries［C］∥Proc of workshop on Text Summarization Branches Out,2004:74-81.
	附中文参考文献：
［18］	谭红叶，李宣影，刘蓓.基于外部知识和层级篇章表示的阅读理解方法［J］.中文信息学报,2020,34(4)：85-91.
［19］	张栋，李寿山，王晶晶.基于问题与答案联合表示学习的半监督问题分类方法［J］.中文信息学报,2017,31(1)：1-7.

[1]	张莉丽, 王睿伯, 王晓东, 张文喆. 自治故障管理系统推理规则的智能学习技术[J]. 计算机工程与科学, 2023, 45(05): 772-781.
[2]	张怡睿宸, 李云峰, 顾旭阳, 纪淑娟. 一种融合行为与结构特征推理的造假群组检测算法[J]. 计算机工程与科学, 2021, 43(05): 926-935.
[3]	龙腾1,2，许智武3. 基于限界约束的安全相关性质的推理证明[J]. 计算机工程与科学, 2017, 39(04): 717-724.
[4]	包甜甜1，王岩磊2，李猛1，谢新连1，张薇1. 基于证据推理的多指标评价系统设计与实现[J]. J4, 2016, 38(06): 1269-1274.
[5]	苏依拉，吉亚图，窦葆媛. 基于蒙古语课程领域语义Web的推理与检索方法的研究[J]. J4, 2016, 38(02): 376-385.
[6]	申蔓蔓1，乐晓波1，周恺卿2. 一种新的基于直觉模糊Petri网的模糊推理算法[J]. J4, 2015, 37(02): 354-358.
[7]	孙晓玲，王宁. 基于直觉模糊Petri网的故障诊断[J]. J4, 2014, 36(09): 1736-1741.
[8]	张弛1，乐晓波1，周恺卿2，莫礼平3. 采用差分进化算法优化模糊Petri网参数[J]. J4, 2014, 36(06): 1095-1100.
[9]	崔纪锋1,2，杜红悦2，张勇1，邢春晓1，薛本新2. 基于细分部分整体关系建立影像文件目录体系[J]. J4, 2013, 35(6): 1-6.
[10]	陈海亮，雷琳，周石琳. 一种基于空间规则度的谱图划分提取舰船编队方法[J]. J4, 2012, 34(5): 116-120.
[11]	甘茂华. 一种具有二次诊断的生殖疾病诊断专家系统[J]. J4, 2012, 34(5): 184-189.
[12]	徐小来1,2，朱华勇1，贺中武2，王伟2，牛轶峰1. 基于UKF的自适应模糊推理神经网络[J]. J4, 2012, 34(4): 82-87.
[13]	孙晓玲，王宁，梁艳. 应用带标识的模糊Petri网的模糊推理[J]. J4, 2012, 34(3): 152-157.
[14]	郭冬梅，李朝品. 基于检索决策树的螨种分类鉴定专家系统[J]. J4, 2012, 34(3): 192-195.
[15]	曹晓翀，王珊珊，吴刚. 虚拟计算环境中的上下文建模与推理方法研究[J]. J4, 2012, 34(2): 111-115.