Sentiment feedback for English writing using BERT-LSTM hybrid modelling

Addressing the challenges of subjectivity and delayed feedback in English writing sentiment analysis, particularly for non-native academic contexts, this study proposes an automated framework based on a hybrid bidirectional encoder representations from transformers-long short-term memory model. The model integrates Bert's contextual encoding capabilities with long short-term memory's sequential modelling strengths, leveraging attention mechanisms for three-dimensional granular sentiment analysis (tendency-intensity-object). Evaluated on a 2020-2023 subset of the international English language testing system writing dataset, the model achieves a sentiment classification accuracy of 91.5% (F1-score 0.89), outperforming baseline models by 12.3%. Educational applicability testing showed an 87.2% teacher approval for reducing feedback workload, confirming its efficacy in supporting writing pedagogy decisions. Key innovations include a multi-task with domain adaptation and a visualised feedback system, establishing a new paradigm for intelligent educational tool with real-time intervention capabilities.