A human-centred deep learning approach facilitating design pedagogues to frame creative questions

Creative questions are a major component of examination in design education for testing creative aptitude. During this process of framing creative questions, examiners remain ever-inquisitive to know whether questions framed by them really capture features of creative questions. Our objective is to explore whether technology can support examiners in situations like these. This paper investigates features of creative questions through mixed-method research techniques. A model is proposed based on DL algorithms that can find out inherent creativity factors in questions and identify whether a question is creative. This process of identifying creative questions triggers decision-making of examiners by which they update their questions based on the outcome of the DL-based system. This model is implemented using bidirectional encoder representations using transformers (BERT), and long short-term memory (LSTM) method for identifying creativity in questions, and their performance is compared. Results highlight that BERT overrules LSTM mechanism, showing 99.99% and 81.006% accuracy, respectively. Inter-rater reliability between the model and examiner’s opinion shows higher agreement (α = 0.96) in categorizing creative questions, and comparison among baselines builds trust in the model. A significant contribution of this research is to capture creative features in a question and categorize whether a question is creative in design education. This model highlights human–machine collaboration and promotes examiners' decision-making process to frame effective questions. It attempts to reduce uncertainty of examiners and assists in quick decisions to include creativity features in their questions by providing feedback on whether a question is creative.