Research of Automatic Scoring of Essays Based on Data Augmentation
Weiqin Guo, Yong Yang, Ge Ren · 2024
The article discusses the research on automated essay scoring based on data augmentation, focusing on the use of data augmentation methods to improve the performance of automated essay scoring systems. In previous studies, researchers have also applied data augmentation methods to the field of automated essay scoring and achieved certain results. However, the past research still has certain defects, such as the problem of the deviation between the representation vectors of the augmented text and the original text labels. This study calculates the cosine similarity between the representation vectors of the augmented text and the original text, and uses the obtained cosine similarity to screen the augmented text, in order to solve the problem of the deviation between the augmented text and the original labels. This study conducts data augmentation on the corpus of automated essay scoring and screens the augmented data to improve the training effect, reducing the negative impact of biased data on model training, and ultimately achieved a QWK (Quadratic Weighted Kappa) score of 80.21 on the ASAP (Automated Student Assessment Prize) automated essay scoring dataset. In model construction, long convolution is adopted to improve performance, and the effect is verified in the experiments, confirming the role of long convolution in the task of automated essay scoring.