南京大学学报(自然科学版) ›› 2019, Vol. 55 ›› Issue (1): 125132.doi: 10.13232/j.cnki.jnju.2019.01.013
顾健伟1,2,曾 诚3,邹恩岑1,陈 扬1,2,沈 艺1,2,陆 悠1,奚雪峰1,2*
Gu Jianwei1,2,Zeng Cheng3,Zou Encen1,Chen Yang1,2,Shen Yi1,2,Lu You1,Xi Xuefeng1,2*
摘要: 机器阅读理解(Machine Reading Comprehension,MRC)一直是自然语言处理(Natural Language Processing,NLP)领域的研究热点和核心问题. 近期,百度开源了一款大型中文阅读理解数据集DuReader,旨在处理现实生活中的RC(Reading Comprehension)问题. 该数据集包含1000 k的文本、200 k的问题和420 k的答案,是目前最大型的中文机器阅读理解数据集,在此数据集上发布的阅读理解任务比以往更具有实际意义,也更有难度. 针对该数据集的阅读理解任务,分析研究了一种结合双向注意力流与自注意力(self-attention)机制实现的神经网络模型. 该模型通过双向注意力流机制来获取query-aware上下文信息表征并进行粒度分级,使用自注意力机制捕捉文本和问题句内的词语依赖关系和句法信息,再通过双向长短期记忆(Long Short-Term Memory,LSTM)网络进行语义信息聚合. 实验结果最终得到相同词数百分比(BLEU-4)为44.7%,重叠单元百分比(Rouge-L)为49.1%,与人类测试平均水平较为接近,证明了该模型的有效性.
中图分类号:
[1] Hermann K M,Kocisky T,Grefenstette E,et al. Teaching machines to read and comprehend ∥ Proceedings of the 28th International Conference on Neural Information Processing Systems. Montreal,Canada:MIT Press,2015:1693-1701. [2] Hill F,Bordes A,Chopra S,et al. The Goldilocks principle:Reading children’s books with explicit memory representations. 2016,arXiv:1511.02301. [3] Chen D Q,Bolton J,Manning C D. A thorough examination of the CNN/daily mail reading comprehension task ∥ Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics. Berlin,Germany:ACL,2016:2359-2367. [4] Rajpurkar P,Zhang J,Lopyrev K,et al. Squad:100,000+questions for machine comprehension of text ∥ Proceedings of 2016 Conference on EmpiricalMethods in Natural Language Processing(EMNLP 2016). Austin,TX,USA:ACL,2016:2383-2392. [5] Cui Y M,Liu T,Chen Z P,et al. Consensus attention-based neural networks for Chinese reading comprehension ∥ Proceedings of COLING 2016,the 26th InternationalConference on Computational Linguistics. Osaka,Japan:TechnicalPapers,2016:1777-1786. [6] He W,Liu K,Lyu Y,et al. DuReader:A Chinese machine reading comprehension dataset from real-world applications. 2018,arXiv:1711.05073. [7] Zhang J B,Zhu X D,Chen Q,et al. Exploring question understanding and adaptation in neural-network-based question answering. 2017,arXiv:1703.04617. [8] Huang H Y,Zhu C G,Shen Y L,et al. Fusionnet:Fusing via fully-aware attention with application to machine comprehension.2018,arXiv:1711.07341. [9] Liu R,Wei W,Mao W G,et al. Phase conductor on multi-layered attentions for machine comprehension. 2018,arXiv:1710.10504. [10] Liu X D,Shen Y L,Duh K,et al. Stochastic answer networks for machine reading comprehension. 2018,arXiv:1712.03556. [11] Bahdanau D,Cho K,Bengio Y. Neural machine translation by jointly learning to align and translate ∥ Proceedings of the 3rd International Conference on Learning Representations(ICLR’2015). San Diego,CA,USA:arXiv:1409.0473,2015. [12] Kim Y. Convolutional neural networks for sentence classification ∥ Proceedings of Conference on Empirical Methods in Natural Language Processing. Stroudsburg,PA,USA:ACL,2014:1746-1751. [13] Li S,Zhao Z,Hu R F,et al. Analogical reasoning on Chinese morphological and semantic relations. 2018,arXiv:1805.06504. [14] Srivastava R K,Greff K,Schmidhuber J. Highway networks. 2015,arXiv:1505. 00387. [15] Hochreiter S,Schmidhuber J. Long short-term memory. Neural Computation,1997:1735-1780. [16] Papineni S,Roukos S,Ward T,et al. Bleu:A method for automatic evaluation of machine translation ∥ Proceedings of the 40th Annual Meeting on Association for Computational Linguistics. Philadelphia,PA,USA:Association for Computational Linguistics Stroudsburg,2002,7:311-318. [17] Lin C Y. ROUGE:A package for automatic evaluation of summaries ∥ Proceedings of the Workshop on Text Summarization Branches Out. Barcelona,Spain:ACL,2004:10. [18] Wang S H,Jiang J. Machine comprehension using match-LSTM and answer pointer. 2016,arXiv:1608.07905. |
[1] | 安 晶, 艾 萍, 徐 森, 刘 聪, 夏建生, 刘大琨. 一种基于一维卷积神经网络的旋转机械智能故障诊断方法[J]. 南京大学学报(自然科学版), 2019, 55(1): 133-142. |
|