Natural Language Processing (NLP) based privacy preserving on technology integration using sentiment analysis: A case study on higher educational institutions' elearning systems

Abstract

The rapid integration of eLearning platforms in higher education, accelerated by the COVID-19 pandemic, has created significant opportunities for flexible teaching and learning while simultaneously amplifying critical concerns regarding student data privacy and security. This study investigates the application of Natural Language Processing (NLP) through sentiment analysis to proactively identify and address learners' privacy challenges in digital learning environments, focusing on institutions in Ghana, including the University of Cape Coast, University of Energy and Natural Resources, and Kwame Nkrumah University of Science and Technology. A novel two-tiered hierarchical framework was developed using advanced transformer-based architectures, specifically fine-tuned Bidirectional Encoder Representations from Transformers (BERT) models. The primary sentiment analysis classifier filters user feedback into positive and negative categories, achieving robust performance with an accuracy of 80-91%, precision of 0.850, recall of 0.840, and F1-score of 0.844 on augmented datasets. For responses classified as negative, a dedicated BERT-based privacy classifier identifies specific concerns with an overall accuracy of 74-88%, demonstrating reliable performance in categorising privacy issues into three prevalent types: data breaches/leakage (85% accuracy), identity theft (65% accuracy), and location tracking (85% accuracy).