TinyBERT-Based Sentiment Analysis for Large-Scale Educational Feedback: A Case Study on Coursera Reviews

Educational platforms often collect extensive user feedback to evaluate and improve their services. Analyzing such feedback manually is time-consuming and inefficient, highlighting the need for automated sentiment analysis. This study aims to develop an efficient and reliable sentiment analysis model using TinyBERT to classify user reviews from the Coursera educational platform. A dataset of 60,000 reviews was categorized into three sentiment classes: positive, neutral, and negative. TinyBERT was fine-tuned to address the imbalanced sentiment distribution and effectively classify reviews. The evaluation results show an overall accuracy of 78%, with an F1-score of 0.90 for the positive class, reflecting the dataset's bias toward positive reviews. However, challenges were encountered in classifying neutral and negative sentiments due to overlapping text features and limited computational resources. Despite these limitations, the study demonstrates TinyBERT's feasibility for large-scale sentiment analysis in resource-constrained environments, providing actionable insights for improving educational services.