International Journal of Science, Engineering and TechnologyAuthors: Raj Kumar, Monti Saini, Shilpy Sharma
Abstract: Sentiment analysis is a powerful computational technique used to identify, extract, and interpret subjective information embedded within textual data. It focuses on recognizing and classifying opinions, attitudes, emotions, beliefs, and feelings expressed by individuals through written language. In the educational domain, sentiment analysis plays a crucial role in understanding student behaviour, engagement, motivation, and overall learning experiences. By analysing linguistic features such as word choice, sentence structure, contextual meaning, and sentiment lexicons, sentiment analysis systems can detect a wide range of emotional states, including positive, negative, neutral, and subtle affective expressions. Recent advancements in Natural Language Processing (NLP) have significantly improved sentiment analysis performance through the adoption of transformer-based architectures and Large Language Models (LLMs). These models are trained on massive datasets using deep neural networks, enabling them to capture contextual and semantic nuances more effectively than traditional machine learning approaches. Models such as BERT, RoBERTa, and ELECTRA have demonstrated remarkable success across various NLP tasks, including sentiment classification and emotion detection. However, the practical deployment of such models in educational settings often faces challenges related to limited labelled data, high computational costs, and resource constraints. To address these challenges, this project investigates the application of LLMs for student sentiment analysis within an Education 4.0 framework by examining three primary adaptation strategies: zero-shot learning, N-shot learning, and fine-tuning approaches. The proposed system analyses student feedback to assess sentiment polarity, emotional dimensions, and a composite Learning Quotient (LQ) that reflects engagement, comprehension, motivation, collaboration, and critical thinking. Experimental observations indicate that different adaptation strategies yield significantly varied performance outcomes, highlighting the importance of selecting appropriate modelling techniques based on available resources and application requirements. Overall, the results emphasize the strong potential of LLM-based sentiment analysis systems in enhancing data-driven decision-making, personalized learning, and adaptive educational environments despite existing resource limitations.
DOI: http://doi.org/10.5281/zenodo.21425087
