Development of an automated question generation system from multimedia content using text-to-text transfer transformer with Bloom taxonomy classification
Abstrak
The object of this study is the assessment process in digital learning environments, where multimedia educational materials such as textual documents and instructional images and require efficient methods for automatic question generation. The main problem investigated is the difficulty in generating high-quality, relevant questions from various multimedia learning materials (text and images). A multimedia question-generation system is proposed that integrates the text-to-text transfer transformer (T5) model with classification based on Bloom's taxonomy. A preprocessing workflow has been created that extracts and combines textual representations from text and images using optical character recognition (OCR) for data tokenization and performs named entity recognition (NER). The question generator application can generate various question types, including multiple-choice, short-answer, and essay questions. These questions are classified according to Bloom's taxonomy. The generated questions were evaluated using bilingual evaluation understudy (BLEU) and recall-oriented understudy for gisting evaluation (ROUGE). Experimental results demonstrated strong performance, with average scores of BLEU-1 = 0.86, BLEU-2 = 0.79, ROUGE-1 = 0.88, and ROUGE-2 = 0.81. Evaluation scores indicate that the multimodal quiz generator application produces high-quality and contextually relevant questions. Evaluation scores show similarities between reference questions and generated questions, with structured questions receiving higher scores than essay questions. The system allows its use in education and intelligent tutoring systems to produce measurable, efficient assessments. The method proposed in this study is limited to multimedia input consisting of text and images
Identifier
- DOI: https://doi.org/10.15587/1729-4061.2026.355867
- OpenAlex: https://openalex.org/W7166854242