Research On Summarizing-Question Answering In Prose For Reading Comprehension

Reading comprehension is a branch of the question-answering system and researchers pay much attention to it in recent years.The existing research on the reading comprehension are proposed for English corpus,and the content is relatively simple,relatively fixed form,but the reading comprehension of the college entrance examination is novel,extensive and diverse in the form of problems.Therefore,the task of Chinese reading comprehension for the college entrance examination is very challenging.The thesis,studies summarizing-question answering of reading comprehension in prose in Beijing college entrance examination,aiming at the national project of 863,and the main work is as follows:(1)Problem analysis of summarizing-question for reading comprehension in prose.We analyze the prose of literary reading comprehension,and we find that the highest frequency and high score distribution is the summarizing-question,whose way of asking question is complex and diverse.In order to improve the effect of the machine on such problems,we construct a corpus of summarizing-questions.(2)Way of word association for keywords in problem.The reading comprehension in prose of the college entrance examination is complex,diverse,and the semantic abstractions of words.However,reading materials express rich and implicative content,which leads to the semantic gap between the problem and the answer.Focusing on the problem,we propose the method of word association.Firstly,all of the words in corpus should be clustered by topics through LDA.Then the words are filtered by the part of speech and frequency of them.Secondly,extend the set of words by adding the lexeme-related ones based on semantic similarity of word embedding,so we get words which are related to the subject words.Finally,choose emotional words from the reading material according to the emotional dictionary,adding these emotional words to the extended semantic association word set,which are deemed to the associated word for the current problem.The effectiveness of this method is evaluated by answering the summarizing-questions in reading comprehension.The experimental result show: Using the method of word association in this article to answer the question,we got the F value of the answer questions is 35.11%,which was 5.57% higher than the baseline method.(3)Get the answer-sentence for summarizing-question.The article uses three methods to get answers,respectively,the method of lexical matching and semantic similarity calculation,the sentence similarity calculation method based on word2 Vec and the modeling method of sentence similarity based on CNN.These three methods are used in the experiment of reading comprehension which was sorted by people in prose,and the result displayed: The results of sentence similarity calculation based on word2 Vec are the best,the F value of which is 52.04%.The F value of the combination method of the lexical matching and the semantic similarity calculation method and the modeling method of sentence similarity based on CNN were 50.34% and 49.2%,respectively.(4)System of answer summarizing-question.We combine word association and answer sentence extraction technology of the thesis,and construct the question-answering system of the summing up and summarizing of college entrance examination in prose.The system has a simple interface and clear module features,which can show the practicability of our methods well.