Research on MOOC Reviews Oriented Sentiment Analysis by Awareness of Emotional Distinctions

It is essential for MOOCs to understand the learner's emotions. This study aims to develop a novel sentiment analysis model to automatically classify the emotions of MOOC learners, which is beneficial to educators and instructors in improving the quality of courses and platform design in MOOCs. Firstly, a robust and interpretable model was proposed, named BB-OAM, which incorporated BERT, attention method, and BiLSTM to extract features of forum reviews as accurately as possible. To capture and differentiate among various affective tendencies more effectively in sentiment analysis, we have taken a breakthrough by introducing an orthogonal attention mechanism to enhance the model's performance for emotion-ambiguous sentences. Compared with models in previous studies including SVM, Tree-CRF, BiRNN, LSTM, and BiLSTM, our method improved the accuracy value of sentiment analysis by 24%, 24%, 19%, 17%, and 9% respectively. Ablation experiments were conducted to systematically evaluate the impact of the orthogonal attention mechanism on sentiment analysis. Through visualization analysis, the model showed higher sensitivity in capturing sentiment-related contents, which further validated the reliability and effectiveness of the proposed method in sentiment classification tasks of MOOC reviews. In essence, this study has great methodological and theoretical insights to help educators and instructors gain a deeper understanding of the actual needs of learners, so as to optimize the efficiency of utilizing MOOC platforms and courses, and to promote effective interactions and collaboration between learners and educators.