A Study of Scoring English Tests Using an Automatic Scoring Model Incorporating Semantics

The automatic essay scoring (AES) model enables automatic analysis and scoring of texts, which is an essential element in education. This paper designed an AES model to extract text features from lexical, semantic, and thematic aspects. The semantic features were obtained by a convolutional neural network (CNN) and long short-term memory (LSTM) neural network. The thematic features were obtained using term frequency-inverse document frequency (TD-IDF). Then, the neural network-based AES model was used to realize the automatic scoring of English texts. The experiment on the Kaggle ASAP dataset found that the model worked best when using all features for scoring, and its quadratic weighted Kappa and Pearson correlation coefficients were 0.8163 and 0.8535, respectively, which outperformed the baseline model. The experimental results demonstrate that the AES model is more reliable for automatic scoring of English texts than the baseline model. The model can have a better application in practice.