Research On Automatic Evaluation Of Composition Rhetoric In Elementary School

In the era of artificial intelligence and "Internet +",how to use technology to achieve educational change has become an important research topic.This research focuses on Automatic Essay Evaluation,and explores how to implement automatic essay rhetoric evaluation based on rhetoric features,so as to effectively reduce teachers’ burden and promote the deep integration of education and technology.This article first combines the relevant theories of rhetoric and the definitions and characteristics of rhetoric,and analyzes and extracts the feature set,including vocabulary,sentence pattern and figures of speech.There are 76 predictors in total and 78 index values including two statistic methods were used for vocabulary density and vocabulary density.After determining the evaluation framework,this study tests the five classification algorithms of Logistic Regression,Naive Bayes,Fast Decision Tree,Logical Model Tree,and Random Forest.In terms of test data sources,the relevant data is mainly composed of grade 4 essays for elementary school students.After digital processing,it constitutes a sample of 336 essays.In terms of testing methods,this study divides the rhetoric level into three levels: good,average,and poor.The rhetoric level is judged manually as the dependent variable.The predictive variables are normalized and selected used by chi-square statistics and information gain methods respectively.Next,the article evaluates the performance of the classification method using accuracy rate,recall rate and F1 value,and finds best classification algorithm and feature set.Finally,based on the feature set of information gain screening,this study uses Logical Model Tree to build an automatic rhetoric evaluation model for primary school essay.The performance of model and each rhetoric level is evaluated using accuracy rate,recall rate,F1 value,and adjacent consistency rate.The experimental results show that the model’s consistency rate adjacent reaches 86.08% and we can accept the model performance.For essays with good rhetoric level,the model has well performance: the accuracy rate reaches 0.938,the recall rate and F1 value are above 0.8.