Context-Controlled Question Generation for Adaptive Learning

In this paper, we introduce Context-Controlled Question Generation model, a novel approach to generate automatic questions which are tailored for adaptive learning environments. Earlier existing methods, they primarily focus on static context and limited controllability, Ours leverages instruction-tuned FLAN-T5 models to dynamically adjust question context, styling, and topical relevance according to the proficiency of the learner. Our approach incorporates advanced techniques such as multi-paragraph input comprehension, adaptive difficulty control, and strategic prompt engineering to generate contextually coherent and pedagogically valuable questions.Through extensive experiments on the SQuAD v2.0 dataset, our model outperforms state-of-the-art models, achieving 8.3% higher METEOR and 6.7% higher BERTScore than traditional T5-Base models. Also, the model demonstrates significant improvements in contextual relevance and cognitive engagement, validated through expert educator evaluations. Our method bridges existing gaps in educational question generation by providing fine-grained control over question characteristics while maintaining high semantic fidelity and diversity.