Research Of Discourse Title Selection For Chinese Reading Comprehension Of College Entrance Examination

Discourse Title Selection of reading comprehension is one of the types of the College Entrance Examination on Chinese.It is one of the tasks in the field of artificial intelligence.Discourse Title Selection consists of one article,one question,and the question is followed by four candidate titles in which only one is the most suitable.The title can accurately express the meaning of the article,and has a distinctive structure.Summarizing information about the article and analyzing the title structure is the key to solve the problem.In this work,we propose a TS-LSTM/CNN(Title Selection-LSTM/CNN)model which based on convolutional neural networks and recurrent neural network.This model combines the merits on extracting linguistic information from both convolutional neural networks and recurrent neural network.The method excavates relevant information from the article and the title to select the best answer.The end-to-end neural network model can automatically learn features,but the results of the experiment are not very obvious because of data sparsity and problem complexity.On the basis of this,analyze the features of the question types,then proposed a correlation analysis model based on title and discourse main points to solve the problem.This model constructs a correlation matrix of title and the discourse main points by analyzes the correlation between title and the discourse main points.On this foundation,together with the title structure features to select the best answer.Because there is no reading comprehension data sets for discourse title selection question.Firstly,we collect the college entrance examination questions and test questions in the period from 2004 to 2016.Besides,form a new TS-LCSTS(Title Selection-LCSTS)dataset from the LCSTS data set by screening and integrating data.Finally,we have tested on the two data sets,which verifies the validity of the method.