International Journal of Education and Development using ICT

International Journal of Education and Development using ICT > Vol. 19, No. 3 (2023)

Revolutionising Educational Assessment: Automated Question Classification using Bloom’s Taxonomy and Deep Learning Techniques – A Case Study on Undergraduate Examination Questions

Kuhaneswaran Banujan, Sabaragamuwa University of Sri Lanka
Samantha Kumara, Sabaragamuwa University of Sri Lanka
Senthan Prasanth, Sabaragamuwa University of Sri Lanka
Nirubikaa Ravikumar, Sabaragamuwa University of Sri Lanka

Abstract
Examinations are one way of evaluating students. To ensure the production of valid exams, frameworks such as Bloom's taxonomy are utilised when preparing questions. Bloom’s taxonomy is a well-known framework that categorises educational objectives into six hierarchical levels of cognitive complexity. However, manually categorising exam questions can be time-consuming and subjective. The extant literature has yet to leverage advanced deep learning methods and state-of-the-art word embedding techniques. This study utilises the effectiveness of Artificial Neural Network (ANN) and Long Short-Term Memory (LSTM) models along with GloVe, BERT and TF-IDF for automating the classification of exam questions according to the revised Bloom’s taxonomy. The study collected various question types from online sources and multiple state universities in Sri Lanka, resulting in a dataset of 16,584 questions labelled manually with the aid of domain experts. The dataset was cleaned using natural language processing techniques. Three models were proposed: ANN+TF-IDF, LSTM+GloVe, and LSTM+BERT. The results of the ANOVA and post hoc pairwise comparisons using Bonferroni correction indicate that the LSTM+BERT model outperformed the other models significantly. The proposed approach provides a reliable and consistent way of evaluating students, and educators can use it to improve their teaching strategies. The findings of this study have important implications for educational institutions and can lead to more effective and efficient evaluations.

Full Text: PDF

Research
Support Tool