Research On The Identification Of Rhetorical Devices In Elementary School Composition Based On Deep Learning

With the continuous development of artificial intelligence technology,the use of technological means to change "education" has become a hot research topic.To solve the problems of flexible use of rhetoric,objective evaluation and improvement of feedback efficiency in primary school composition,this study is designed with the theme of automatic identification of rhetorical techniques in primary school composition,and explores and researches the deep learning model for judging rhetoric identification.At the same time,it promotes the deep integration of natural language processing technology and education.This paper first analyzes the concepts and characteristics of primary school rhetoric devices,and selects parallelism,metaphor,anthropomorphic and quotations as the research content,and determines specific rhetorical identification schemes based on their rhetorical characteristics.Secondly,the basic concepts and algorithm principles involved in the process of model construction are explained.At the same time,a total of 39,000 pieces of parallelism,metaphor and anthropomorphic data are crawled from the primary school composition website,after a series of data classification,structure checking and sorting and cleaning operations,I finally got 25,000 data of parallelism,metaphor and anthropomorphic,and 50,000 pieces of quotations database.Thirdly,perform distributed feature extraction for text segmentation to generate word vectors,and use convolutional neural networks to implement text classification models to obtain the accuracy,recall and F1 value of the model,and analyze and predict model results.At the same time,increase the comparison study with the recognition effect of the machine learning model,the effect of different embedding layers on the convolutional neural network model,and the analyze of model performance.Finally,the bidirectional multi-angle matching model is used to conduct matching and recognition research on citation rhetoric,and the algorithm process and principle of the model are discussed.When the accuracy,recall and F1 value of model matching are obtained,the model results are compared and analyzed.Experimental verification and comparison results show that the text classification model constructed by the convolutional neural network has a recognition effect of parallelism,metaphor and anthropomorphic rhetoric due to the machine learning method.Its accuracy rate can reach 90.4%,and the recall rate and F1 value are both 0.8above,the performance effect of the model is better.At the same time,the accuracy rate of the two-way multi-angle matching model used in the rhetorical method can reach85.3%,and the recall rate and F1 value are both above 0.8.Therefore,this study has certain practical application value and significance for the automatic recognition of elementary school composition rhetoric.