Improving Sentiment Analysis in Online Course Reviews with BERT and Transformer Attention Mechanism

Abstract In the field of text mining, sentiment analysis has grown significantly in importance for understanding user reactions. This knowledge contributes to improving several features of given goods or services. Considering its adequacy, sentiment analysis can play a significant role in the educational field, where student input is essential for enhancing curricula and recommending courses, resources, and other elements. Likewise, because of the higher accuracy and decreased complexity, transfer learning models are increasingly being used in a variety of domains, which suggests that the educational sector will substantially benefit from incorporating them into its related domains. To leverage the advantages of both sentiment analysis and transfer learning techniques, this study suggests a Transfer Learning Model (BERT-base-uncased) based on BERT (Bidirectional Encoder Representations from Transformers) using the Transformer Attention Mechanism to analyse the sentiment of student feedback from various online courses. Two applications of the suggested model have been demonstrated- one with a sequence classifier and another without it. The result shows a significant accuracy of 76% with a high recall value with the sequence classifier.