Abstract

This research aims to develop a robust sentiment analysis model by fine-tuning pretrained language models such as BERT, DistilBERT, and MobileBERT on diverse datasets, specifically movie and restaurant reviews. To enhance model performance, ensemble techniques are employed, combining the strengths of multiple models. Leveraging the knowledge gained from movie and restaurant reviews, the study applies transfer learning to predict sentiments in educational content reviews, demonstrating the effectiveness of this approach. The developed sentiment analysis model is applied to analyze comments on educational vlogs, providing insights into viewer sentiment and engagement. This analysis generates detailed reports that highlight key themes, positive and negative feedback trends, and actionable insights to help educational content creators improve their content. The performance of individual pretrained models and their ensemble is assessed on both the training datasets (movie and restaurant reviews) and the target dataset (educational reviews). Additionally, the study compares the ensemble model's performance with baseline sentiment analysis models to demonstrate the advantages of the ensemble and transfer learning approach. The practical implications of using sentiment analysis for educational vloggers are explored, including how it can enhance content strategy, viewer engagement, and content quality.