摘要：

语句填补类选择题是高考语文阅读理解中题型之一,是自然语言处理研究的热点,其中题干信息和答案的关系非常隐蔽,无法从篇章中直接选出答案。为此,针对语句填补选择题提出基于长短时记忆网络模型的语句填补答案选择方法。使用神经网络中的LSTM模型对篇章中每个句子和选项进行分布式表示,通过向量直接拼接和按位相乘融合篇章和选项之间的语义信息,实现对语句填补类选择题的解答。在新闻语料和全国各省近10年高考题和模拟题上的实验结果,验证了该方法的有效性。

关键词: 机器阅读理解, 神经网络, 语句填补, 分布式语句表示, 答案选择

Abstract:

The sentence filling is one of the important choice-questions in the reading comprehension based on the college entrance examination,and has also become a hot research topic in the field of natural language processing.Due to the relationship between the information and the options is very subtle,the answer can not be drawn directly form text.Aiming at this problem,this paper applies Long Short-Term Memory(LSTM) model of deep learning to answer the choice-questions of sentence filing.The models carry out the distributed sentence representation for every sentence of document and option,by two means of concatenate and multiply which efficiently combine the semantic information of discourse and options.Experiments are carried out on news corpus and college entrance examination Zhenti and Moniti in the recent ten years,which have verified the validity of the proposed method.

Key words: machine Reading Comprehension(RC), neural network, sentence filling, distributed sentence representation, answer selection

中图分类号: 

• TP18