AI-Assisted College Students' English Writing Scoring with Hierarchical LSTM for Enhanced Contextual Understanding and Grading Accuracy

In the realm of college English instruction, assessing students' writing abilities both objectively and effectively presents a variety of ongoing challenges. To address these issues, this paper proposes an AI-supported system for automated English writing evaluation. The system leverages a Hierarchical Long Short-Term Memory (Hierarchical LSTM) model designed to enhance both contextual understanding and accuracy in scoring. By employing advanced natural language processing capabilities, the system automatically evaluates the syntactic, semantic, and logical structure of texts. Through a hierarchical framework, the model thoroughly analyzes the intricate relationships between sentences and paragraphs. During the model's execution, the Hierarchical LSTM first extracts local textual features before enhancing global semantic understanding, ensuring uniform grading standards, even for extended texts with complex sentence structures. In contrast with traditional methods, the AI-assisted system substantially minimizes subjective biases while markedly improving scoring efficiency and precision. Tests conducted on the CET-4 mock test dataset and the Pigai dataset show an average Pearson correlation coefficient of 0.815. This represents an increase of 0.18 and 0.13 compared to baseline models such as BiLSTM and RNN, with average scores of 0.632 and 0.513, respectively, further validating the proposed method's reliability and efficacy.