Application of Recurrent Neural Networks Based on Attentional Mechanisms in Classification Error Correction in English Teaching

Abstract

Abstract The way teachers correct errors in English teaching can cause problems such as psychological pressure, and deep learning technology offers the possibility of automatic error correction. In this paper, the final states of left and right texts are computed by constructing two attention mechanisms, target word-independent and related, and the merged obtained vectors are inputted into the RNN model for grammatical error recognition. In collocation error recognition, a Rank-based word candidate set ranking method is added, and error correction for verb usage is semantically encoded using RNN. The study was analyzed and tested in terms of grammar, collocation, and verbs. The ATT-RNN model accuracy is 3.62 percentage points higher than CAMB, and the difference in recall and F 0.5-value is not more than 0.5 percentage points, which indicates that the algorithmic model has some research value.