Sentiment analysis in digital learning: Comparing Lexical, Traditional machine learning, and deep learning approaches
Mohammed Amraouy, Mohammed Majid Himmi, Mostafa Bellafkih, Jallal Talaghzi, Abdellah Bennane · 2023
Over the last decade, a gradual shift has been observed towards using sentiment analysis approaches to improve the online learning environment. In fact, sentiment analysis can provide us with valuable information about the socio-affective engagement of learners, which is considered a key condition for competences development. Lexical, machine learning, and hybrid approaches have become popular choices in sentiment analysis research and applications. The current study examines the classification accuracy of learners' socio-affective engagement based on their interaction messages using five methods: TextBlob, Naive Bayes, Simple Neural Net, CNN, and LSTM. The experimental results suggest that the LSTM and CNN models perform better, with an accuracy rate on the test set exceeding 0.83.