Transformer-Based Automated Question Prediction and Generation Aligned with Bloom’s Taxonomy for Educational Assessments
No Thumbnail Available
Date
2025-07-30
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Bloom's Taxonomy, a widely recognized framework for predicting cognitive skills, guides the
design and evaluation of educational assessments. Traditional methods often struggle to maintain
an effective balance between Lower-Order Thinking Skills (LOTS) and Higher-Order Thinking
Skills (HOTS), while manually creating exam questions remains time-consuming and inefficient.
To address these challenges, this study presents an automated system for predicting and generating
exam questions aligned with Bloom's Taxonomy, ensuring comprehensive representation across
all cognitive levels. The system leverages transformer-based models, including BERT, Distil
BERT, RoBERTa, T5, BART, DistillBART, XLNet, and BERT2BERT, capitalizing on their deep
contextual and semantic understanding capabilities. Furthermore, the research incorporates a hy
brid prediction model that combines RoBERTa with Long Short-Term Memory (LSTM), Convo
lutional Neural Networks (CNN), and Term Frequency-Inverse Document Frequency (TF-IDF)
features to enhance question prediction accuracy and robustness significantly. The system was
trained and evaluated on a diverse, manually annotated dataset spanning multiple academic disci
plines, ensuring broad applicability. Experimental results demonstrate that the hybrid model
achieved an impressive question prediction accuracy of 96.39% in predicting Bloom's cognitive
levels, effectively handling both LOTS and HOTS complexities. Additionally, the T5-based ques
tion generation model attained a high accuracy of 95.36% alongside a BLEU score of 0.8628,
illustrating its ability to produce contextually relevant and pedagogically diverse questions. This
research contributes to the advancement of educational technology by offering a scalable, efficient,
and intelligent solution for automated assessment design.
Keywords: Bloom's Taxonomy, Automated question generation, Transformer models, BERT,
RoBERTa, T5, BART, BLEU score, Educational Assessments.