Automated English Essay Scoring Based on Machine Learning Algorithms
Shutong Yang · 2023
Many English learners need to improve their English writing skills because of the lack of writing practice, which is a common phenomenon, especially in China. However, it is impractical for English teachers to grade and provide comprehensive feedback for essays due to the limited time. In this paper, an automated essay scoring method is proposed to solve this issue. The 3000 features are ordered by term frequency from the whole Kaggle dataset. Furthermore, 10 handcrafted features of evaluating an essay are selected based on experience, and then 2-degree polynomial form of the handcrafted features is prepared for comparison. After data preprocessing, Natural Language Processing (NLP) and other machine learning models are considered in this paper. By using the mean-square error and mean average percentage error as the criteria, the model comparison is conducted in the base 3000 features, 10 handcrafted features with base 3000 features, and 66 polynomial features with base 3000 features. Then, Linear Regression, Ridge Regression, Gradient Boosting Regression, Random Forest Regression or XGboost Regression with the best performance can be found. The results demonstrate that XGboost and Ridge Regression provide the top two performance among the algorithms used on this dataset. The polynomial feature importance can be obtained from the algorithm with the best performance. These polynomial features can be the guidance for students to improve their writing skills. In conclusion, the proposed methods can be effective for English essay scoring.