Optimization of English Essay Automatic Scoring Algorithm Based on Deep Learning

With the development of intelligent education technology, the automatic scoring system of English composition plays an increasingly important role in large-scale writing evaluation and personalized teaching feedback. However, existing scoring models generally have problems such as insufficient context understanding ability, single scoring dimension and limited scoring consistency. To this end, this paper proposes a deep learning model based on the fusion of Bi-LSTM and CNN structure, combining language context modeling and local grammatical feature extraction to improve the system’s comprehensive judgment ability on the language quality of students’ compositions. The model was evaluated on two English composition datasets from different sources. The results show that this method is superior to traditional models and mainstream deep learning methods in terms of scoring accuracy, relevance and scoring consistency, and has good adaptability and stability. The study also further analyzes the role of multi-dimensional language feature fusion in promoting model performance and verifies its application potential in real writing scenarios.