Emotion Prediction from Online Course Reviews by Using Deep Learning

Abstract

Massive open online courses (MOOCs) emerged as a pivotal solution for distance learning during the COVID-19 pandemic, effectively breaking down barriers related to age, gender, and geography. This study focuses on developing a robust and precise emotion classification model using advanced deep learning techniques, specifically targeting the reviews on Coursera’s online learning platform. Our research dives into key questions surrounding the performance of different deep learning models, particularly comparing the Long Short-Term Memory (LSTM) network with a hybrid model that combines Convolutional Neural Networks (CNN) and LSTM. We hypothesize that this hybrid approach not only enhances the predictive accuracy of emotion analysis but also outperforms traditional supervised learning methods. By analyzing a comprehensive dataset of 140K course reviews, we demonstrate that the hybrid CNN-LSTM model, when coupled with sophisticated word embedding techniques, achieves superior results, reaching a peak accuracy of 93.80%. This work underscores the potential of hybrid models in capturing the complexities of human emotions in educational content, offering valuable insights for improving online learning experiences.