A study on learners’ emotion classification based on an improved convolutional neural network algorithm in online teaching and learning

With the rapid advancement of online education, students’ sentiment feedback serves as a pivotal factor in enhancing course quality and refining pedagogical strategies. However, conventional sentiment analysis approaches often struggle with unstructured textual data, limiting their ability to discern the emotional inclinations embedded in student comments precisely. To address this challenge, this study introduces RoBERTa-BiLSTM-TextCNN Network (RBTCN-Net), a novel framework that integrates Robustly Optimized BERT Pretraining Approach (RoBERTa), a Convolutional Neural Network (CNN), a Bidirectional Long Short-Term Memory (Bi-LSTM), and an attention mechanism to classify sentiment in an online learning environment. Specifically, RoBERTa is employed to extract deep semantic representations, CNNs capture localized sentiment features, Bi-LSTM models capture temporal dependencies, and the attention mechanism amplifies critical sentiment-related information, thereby improving classification accuracy and robustness. Experimental evaluations demonstrate that RBTCN-Net surpasses standalone deep learning models in positive and negative sentiment classification across publicly available datasets. The results underscore the framework’s capability to effectively analyze sentiment tendencies in online educational discourse, offering valuable data-driven insights for personalized instruction and course refinement. Beyond enhancing sentiment analysis in digital learning contexts, this study also provides innovative technical solutions and pragmatic pathways for developing intelligent teaching evaluation systems.