Design And Implementation Of Chinese Composition Intelligent Evaluation System Based On Deep Learning

Composition can not only improve personal literary literacy,but also occupy a high position in talent selection.However,due to the complexity of Chinese composition understanding and the subjectivity of evaluation in manual evaluation,a mature intelligent evaluation system is increasingly needed to automatically extract the corresponding evaluation indicators to help users get rid of the tedious work.This paper starts from the detailed rules of college entrance examination composition,mainly from the development level of composition evaluation,aiming at the literary talent and content,identifies and judges figurative sentences and parallelism sentences,and designs an intelligent evaluation scheme of Chinese composition based on deep learning aiming at the problems of imperfect evaluation dimension and incomplete test content of current Chinese composition intelligent evaluation system.Based on the implementation of this method,the evaluation results are further optimized.By constructing a neural network model,the metaphorical,parallelism and beautiful sentence components of the composition can be detected.At the same time,we use the extracted text features to calculate the overall score of the composition.Then,using the identified rhetorical sentences and composition scores,the final evaluation results are obtained through the automatic generation model of comments.Finally,the correctness and effectiveness of the method are verified by experiments.In order to solve the problem that deep learning can't get the semantic of the article quickly,bi-lstm model with multiple inputs is used in the input information of figurative sentence recognition.Not only the sentence text after word segmentation,but also the part of speech of the sentence are used as the input of the article.According to the different input,the pre-processing layer encodes the sentence text by embedding,constructs the vector matrix by one hot for the part of speech,and obtains the final input data by splicing the two,which not only ensures the long-distance correlation characteristics of the sentence text,but also integrates the part of speech characteristics of the sentence.Finally,we collect the metaphorical data in the dictionary of using metaphors and the study of metaphorical semantics,test the extracted data set,and compare with other algorithms to verify the accuracy of the model.In the aspect of engineering,this paper realizes the Chinese composition intelligent evaluation system,and provides some auxiliary functions to improve the practicability of the system.For the article input,the image recognition interface is used to facilitate the user to operate.In addition,the system modules such as Article recommendation,article collection and permission setting are provided,which have been deployed and trial run in the server.