Research on the Application of Neural Network Classification Model in English Grammar Error Correction

Abstract

The field of English grammar error correction has developed rapidly in recent years, and many excellent research results have been obtained. However, errors are generally viewed as a whole for revision, which cannot be subdivided to identify the specific error types of sentences, lacks interpretability, and requires a lot of manpower to label the data. To address this, the study uses the CoreNLP tool to output the input text as a basic form of each word, and encodes and decodes the linguistic text in the corpus using the CNN-based Seq2Seq model. And word embedding and left-right GRU operations are performed on left-right text and target words in turn, and the merged new vector is obtained using two attention mechanism operations. It is fed into the MLP classifier for classification and error correction. To further improve the model performance, two model optimizers, stochastic gradient descent (SGD) and Adam, are used to tune the parameters of the classification prediction model. The study thus constructs an English grammar classification and error correction model based on a neural network classification method. The experimental analysis shows that the average accuracy of the model constructed in the study is 96.83%, the average recall rate is 78.93%, and the average overall evaluation F0.5 is 92.67%, which can effectively and accurately classify and correct the grammatical errors that occur in the process of English language learning.