AN EMPIRICAL EVALUATION OF CONTEXTUAL EMBEDDING-BASED MODELS FOR CLASSIFICATION OF EDUCATIONAL QUESTIONS USING BLOOMS TAXONOMY

The automated classification of examination questions according to Blooms Taxonomy (BT) assists question setters in developing high-quality assessments by accurately categorising questions into cognitive levels. While most previous studies in this area have employed traditional machine learning methods, relatively few have explored deep learning-based approaches. Contextual embeddings, in particular, have shown effectiveness across various natural language processing tasks. This study aims to evaluate a hybrid optimal pre-trained contextual word embedding technique, XLNet,combined with a Convolutional Neural Network (CNN) model tailored for BT- based question classific ation. To this end, the study examines the performance of the proposed XLNet+ CNN model with state-of-the-art models.Experimental results indicate that the XLNet + CNN model achieves performance comparable to existing models. Although it is 0.5% lower in overall accuracy than RoBERTa + CNN, it has 8% higher precision for the higher-order cognitive skill Evaluation category and 4% higher precision for the Analysis Category. Despite slightly lower accuracy, XLNet + CNN demonstrates superior precision and better identification of higher-order cognitive skills, making it more suitable for reliable educational assessment tasks.