Research On Answer Extraction Method Of Reading Comprehension

In recent years,reading comprehension has been widely regarded by international and domestic scholars in the field of natural language processing,and has become an important research in the field of artificial intelligence.Reading comprehension is one of the important ports in the college entrance examination papers,which is a research content in the 863 "human intelligence" answer project.College entrance examination reading comprehension has a higher level of challenge compared with the traditional reading comprehension based on the construction of corpus.The main knowledge of the college entrance examination reading comprehension answer is included in the text.Therefore,to achieve the correct answer to read the purpose of understanding the problem,the computer must be in-depth understanding the text and analysis of the questions.The answer selection is based on a text,and try to select the best option from the option set through understanding of the question and options.There are fewer sentences in the text,and the relationship between the information and the options is very subtle,so the answer can’t be drawn directly from the text.How to dig out the relation of the text,the question and the options is the key to solving the problem.Must be on the text and the title of a thorough understanding and comprehensive analysis.In this paper,The object of this study is the view support choice-questions of the college entrance examination,we proposed a answer selection method based on voting theory.First of all,This method uses the semantic core vocabulary matching to extract the view candidate sentences and options candidate sentences,then obtains the option score by the candidate sentence importance,and finally chooses the best option by sorting the options.In order to comprehensively consider the semantic information of the chapter,we propose a method of selecting the answer based on the semantic consistency model,which jointly identifies the relation of a question and relevant sentences and the option.The semantic features of sentence similarity,antisense matching,negation matching and sentence distance are described in the model.we tested these methods on college entrance examination zhenti and moniti.Comparing these methods with the traditional word bag model and the similarity calculation method,which verifies the validity of our methods.