基于多篇章多答案的阅读理解系统  

Machine Reading Comprehension for Multi-document and Multi-answer

作　　者：刘家骅 韦琬 陈灏 杜彦涛 LIU Jiahua;WEI Wan;CHEN Hao;DU Yantao(Department of Computer Science and Technology,Tsinghua University,Beijing 100084,China;Naturali Ltd,Beijing 100080,China)

机构地区：[1]清华大学计算机系,北京100084 [2]北京奇点机智科技有限公司,北京100080

出　　处：《中文信息学报》2018年第11期103-111,共9页Journal of Chinese Information Processing

摘　　要：机器阅读理解任务一直是自然语言处理领域的重要问题。2018机器阅读理解技术竞赛提供了一个基于真实场景的大规模中文阅读理解数据集,对中文阅读理解系统提出了很大的挑战。为了应对这些挑战,我们在数据预处理、特征表示、模型选择、损失函数的设定和训练目标的选择等方面基于以往的工作做出了对应的设计和改进,构建出一个最先进的中文阅读理解系统。我们的系统在正式测试集ROUGE-L和BLEU-4上分别达到了63.38和59.23,在105支提交最终结果的队伍里面取得了第一名。Machine Reading Comprehension(MRC)has become a popular issue in Natural Language Processing(NLP).The 2018 NLP Challenge on Machine Reading Comprehension provides a large-scale application-oriented dataset for Chinese Machine Reading Comprehension,which is much more challenging than previous Chinese MRC dataset.To cope with those challenges,we present a system with improvements in all aspects,including preprocessing strategy,feature expression,model design,loss function and training criterion.Our system achieves 63.38 in ROUGE-L score and 59.23 in BLEU-4 score on the final test set,ranking first among 105 participating teams.

关 键 词：机器阅读理解 问答系统 深度循环神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]