An Automatic Grading System for Chinese Primary School Students’ Compositions

The article introduces an automatic grading system for Chinese primary students’ compositions to help teachers assess writing proficiency quickly and accurately. It tested six deep learning traditional methods (CNN, RNN, RCNN, DPCNN, FastText, Transformer), three large language models (ChatGPT, Kimi, Qwen), and a regression model with 12 linguistic features. Results showed CNN performed best, while ChatGPT also delivered strong predictive performance. Though the regression model surpassed the large language models, it remained behind CNN and other deep learning approaches. The article also proves to some extent the feasibility of using ChatGPT to provide AES for primary school students’ compositions.