Chinese Essay Evaluation Method And System Based On Deep Learning

Due to the problems such as the heavy workload of teachers,monotonous feedback,and subjectivity in the traditional manual correction of essays,many countries around the world have gradually begun to take essay scoring with the help of computer systems instead of manual scoring.Based on deep learning,the evaluation theory and method of students’ Chinese essay from primary and secondary school is studied in this thesis,and an automated evaluation system for primary school students’ Chinese essays is designed.Considering the lack of Chinese essay corpus from primary and secondary school students,essays of primary school students from Xuri Hongwen Co.,Ltd,Fanwen Essay Network and Leleketang Essay Dataset are collected to construct a primary school students’ essay dataset.At the same time,the text containing grammatical error from CGED 2017,NLPCC 2018 and primary school students’ essay dataset are collected to construct a grammatical error text dataset.Aiming at the research of automated essay scoring,a Dynamic Word vector based Automated Essay Scoring model(DW-AES)is designed in this thesis.This model extracts features from essays by ultilizing a convolutional neural network based on dynamic word vectors,where the word vector parameters change with the model training.The last layer(fully connected layer)of this model outputs the grade of the essay(excellent,good,and medium).Firstly,the automated essay scoring process of the DW-AES model is introduced,and the structure and parameters of the DW-AES model is discussed in detail.Then,the influence of different factors on the classification performance of DW-AES is analyzed.The experimental results show that the classification performance of the model proposed in this thesis is better than that of the static word vector model,and better than the traditional machine learning model,fully connected neural network model and Bi LSTM model.The proposed DW-AES model can effectively grade the Chinese essay of primary school students,and the accuracy and f1 score reach to 90.46% and 0.7546 respectively.Aiming at the research of text grammatical error correction,a Dynamic Word vector based Grammatical Error Localization model(DW-GEL)is designed in this thesis,which is suitable for grammatical error localization in Chinese essay written by primary school students.This model adopts the BERT model based on dynamic word vector to convert each word in the text into a one-dimensional vector as the input,and extracts the features of essays via convolutional neural networks.At the same time,the output of the convolutional neural network is connected to the fully connected layer for sequence labeling,so as to complete the localization of text grammatical errors.The process of grammatical error localization in the DW-GEL model is introduced,and the structure and parameters of the DW-GEL model is discussed in detail,as well as the evaluation method.The experimental results show that the performance of the DW-GEL model is better than that of the static word vector model,and better than the fully connected neural network model and the Bi LSTM model,which can accurately identify the location of text grammar errors.The absolute accuracy,3-window relative accuracy,5-window relative accuracy,and 9-window relative accuracy reach to 80.33%,86.48%,87.61%,and 88.52%,respectively.Finally,an Automated Chinese Essay Evaluating System(ACEES)based on the C/S architecture is designd and implemented by merging DW-AES model and DW-GEL model.The front-end interface is based on Qt Designer,and the back-end server is designed with Python.When the user inputs an essay and selects the correction standard for the essay,this system automatically completes essay scoring and grammatical error localization.At the same time,this system generates a personalized report including the essay’s score and modification suggestions to the user.