Research On The Appreciation-question-answer Based On The Figures Of Speech Recognition

Machine reading comprehension requires a machine to understand and analyze the contents of a text on the basis of a given article,and to answer the questions it set up,which involves a number of Natural Language Processing technologies.In this thesis,the reading materials is a literary text which from Beijing College Entrance Examination,the goal is to solve the question of appreciation,carry out a research on the solution of the appreciation-question based on the figures of speech recognition,the two core technology is the method to recognize parallelism and metaphor/simile,the main research work is as follows:(1)Problem analysis in appreciation.This thesis analyzes the requirements of the Chinese College Entrance Examination for the ability of appreciation and evaluation,and realizes that the recognition of the figures of speech is the breakthrough point for the automatic appreciation.Subsequently,the question of appreciation are analyzed which from Chinese College Entrance Examination,the statistics result show the parallelism and metaphor be use frequently,establish the parallelism and metaphor/simile recognition method as the core of the appreciation question solving.For the convenience of the study,the analogy definition and characteristics of parallelism and metaphor/simile are described,collecting 2000 parallelism and 4535 metaphor/simile,statistics of 88 kinds of trigger words from metaphor/simile.(2)Parallelism recognition.In this thesis,according to the characteristics of the similar syntactic structure and content relevance in parallelism,we design the fusion of convolutional neural network and structure similarity calculation method to recognition parallelism.In parallel sentences dataset,compared with other methods,the proposed method F1 value is 91.7%,this method was applied in the Beijing College Entrance Examination,can correctly identify parallelism needed by appreciation-question,and also proves the validity of the method.(3)Metaphor/Simile recognition.Based on the semantic similarity between sentences,sentences contains words and have emotional tendencies,we design of CNN_C method to recognition metaphor/simile,and the experimental result is compared with SVM,RNN,maximum entropy method,CNN_C method F1 value of 83%,which simile sentence F1 value is 89%,metaphor is 50%.The CNN_C method was applied in the Beijing College Entrance Examination can correctly identify parallelism needed by appreciation-question.(4)Construction of essay reading comprehension question and answer system.For appreciation-question,a problem solving process based on the recognition of the rhetorical is designed.First of all,design a appreciation-question recognition method based on keyword matching;secondly,establishment the knowledge base of language skills and answer template;finally,the parallelism and metaphor/simile recognition method,appreciation-question recognition method,the knowledge base of language skills and answer template is combined to construct the question and answer system.This system is used to solve the appreciation-question from Beijing College Entrance Examination,the result is not bad.