User Satisfaction Forecasting in Game-Based Educational Technology through Transformer Models

This paper presents a deep learning approach using transformer-based models to predict user satisfaction in online Educational Technology (EdTech) services. Online educational services have become essential tools as technology advances, providing students with an efficient and concise method for acquiring knowledge. In this study, we analyzed 368,690 metadata which were gathered from the Google Play Store over a period of approximately twelve years, from August 21, 2013 to September 3, 2024 focusing on game-based learning EdTech services. We utilized seven different transformer-based models (BERT, XLNet, ALBERT, DistilBERT, DistilRoBERTa, ELECTRA, DeBERTa) to predict user satisfaction. DeBERTa model achieving the highest performance, demonstrating a score of 89.96 in accuracy and attaining an F1-score of 89.41. Additionally, by combining the DeBERTa model with the sentiment score extracted from VADER (Valence Aware Dictionary for sEntiment Reasoning), we reached a score of 90.12 in accuracy and an F1-score of 89.49. Other models also achieved strong performance, with accuracies and F1-scores ranging between 88 and 89, proving their suitability for predicting user satisfaction in online EdTech services. This research identifies the most optimized transformer-based model for predicting online EdTech user satisfaction and presents a framework that demonstrates even better performance through the integration of sentiment scores. Through this study, valuable direction is provided for further research within this area.