Abstract.
We present an educational framework that automates the generation and assessments of questionnaires without domain-specific and language constraints by leveraging Natural Language Processing (NLP) and advanced Language Models (LMs) to internalize Bloom’s Taxonomy. Our framework categorizes questions into three distinct difficulty levels, addressing the core challenge of transferring the structured knowledge of cognitive and learning levels into LMs. We hypothesize that these difficulty levels can be effectively represented by grouping Bloom’s categories, facilitating the model’s understanding and generation of appropriate questions. To test this hypothesis, we generated multiple-choice and open-ended questions, evaluating their syntactic construction and semantic integrity. Our experiments demonstrate a robust alignment between the proposed difficulty levels and the generated questions compared to our baseline. The framework consistently produces questions that are semantically accurate, syntactically precise, and contextually relevant. Additionally, the automatic evaluation model for open-ended questions provides accurate scores and feedback on student responses, further supporting effective self-assessment. These findings underscore the potential of our approach to enhance the learning process by effectively transferring the structured knowledge of Bloom’s Taxonomy into LMs. This enables the generation of high-quality, difficulty-leveled questions without domain-specific constraints, thus promoting efficient and adaptive learning experiences.