Research On Simile And Personification Identification For Appreciation Question In College Entrance Examination

Reading comprehension task requires analyzing the given context and answering questions related to it.This thesis takes the question of appreciation in the Chinese reading comprehension of the college entrance examination as the background,and analyzes and researches the methods of identifying rhetoric and extracting relevant components,mainly involving metaphor and personification,also designs and completes a system for answering questions in college entrance examination.The main research work is as follows:(1)The question of appreciation analysis in reading comprehension.Through the analysis of various types of questions in literary reading comprehension,it is found that identification of simile in the metaphor and identification of personification in the question of appreciation are examined more frequently in the college entrance examination,and comprehensively answering such questions requires not only identifying rhetoric,but also extracting relevant components.Therefore,the definition and related components of simile and personification are explained respectively,and their differences are analyzed.Moreover,a dataset of simile sentences,personification sentences and relevant components is constructed,annotating the tenor and the vehicle in the simile,annotating the tenor in the personification,and providing a data basis for subsequent research.(2)Simile identification and component extraction.Through studying,it is found that the part of speech features can provide more accurate information for identifying the tenor and the vehicle in the simile.Therefore,the paper proposes a method for the simile identification and component extraction based on part of speech features.The vectorized representation of the words in the sentence is merged with the vectorized representation of the part of speech features,using Bi LSTM to obtain the global features,and adding the CRF layer to the output layer to obtain the optimal annotation sequence of the text.The experiment results show that the proposed method is better than the existing single task method on the open dataset and the F1 value of the simile component extraction is 62.24%,the simile identification is 84.22%.(3)Personification identification and component extraction.Since traditional word vectors cannot represent the polysemy of words,the paper uses a pre-trained language model BERT with strong text feature representation capabilities as the feature acquiring layer of Embedding to achieve deeper text semantic representation.The paper designs a method for the personification identification and component extraction based on character features.This method fuses the features extracted by BERT with character features,inputs the fused vector into the Bi LSTM model for training,and the global optimal annotation sequence is decoded by CRF.Compared with various related methods on the personification dataset,this method has achieved relatively good performance and the F1 value of the personification component extraction is 87.39%,the personification identification is 91.92%.(4)Question and answer system.The construction of the college entrance examination appreciation question answering system uses the methods of simile identification and component extraction,personification identification and component extraction.The system can answer questions through question analysis,rhetoric identification and component extraction,application of templates and language skills.At the same time,the proposed method is applied to the college entrance examination and is proved effective.