【摘要】 机器阅读理解要求机器根据给定的上下文回答问题，但中文成语特殊的结构对机器阅读理解提出了独特的挑战。以往的成语语义表示方法不能很好地获取到成语准确的语义表示，该文提出一种基于成语语义释义来构建辅助数据集帮助模型更好地理解成语语义的方法，在此基础上，使用BERT作为预训练模型，结合双向匹配策略建模完形填空题型中填空部分和候选项之间的关系表征，然后再进一步获取其与文档的关系表征。该文模型ID-BOPM(Idiom Dictionary and Blank-Option-Passage Matching)在大规模中文成语完形填空数据集ChID上进行了实验。结果表明，该方法有助于模型更好地理解中文成语语义，取得较好的效果。更多还原

【Abstract】 Machine reading comprehension requires the machine to answer questions according to the given context. To address the in unique challenge to machine reading comprehension of Chinese idioms, This paper proposes a ID-BOPM(Idiom Dictionary and Blank-Option-Passage Matching) model. It first builds an auxiliary data set based on the semantic interpretation of idioms to help the model better understand the semantics of idioms. On this basis, BERT is used as a pre training model and a two-way matching strategy is adopted to model the representation of the relationship between the blank filling part and candidates in the cloze test, and the whole reading passage. Tested on a large-scale Chinese idiom cloze data set ChID, this method is demonstrated to achieve good results.更多还原