Automated Informative Sentence Selection and MCQ Generation using Transformer Models and Supervised Classification

The increasing demand for scalable and efficient assessment tools in education, and the need for automatic tools that can create excellent multiple-choice questions (MCQs) from texts. This study offers automatic (MCQs) for selecting informative sentences. In our research, we have used BERT, SBERT, NER, and supervised classification techniques. It has stages: preprocessing, entity extraction, sentence encoding, supervised classification, and MCQ generation. Starting, gathering domain-specific texts, such as medical and research articles, followed by preprocessing. Subsequently, a pre-trained NER model extracts key entities, which are then prioritized for further processing based on their contextual relevance. Sentence embeddings are generated using SBERT, enabling efficient ranking of sentences through cosine similarity measures. These embeddings capture semantic relationships between sentences, crucial for selecting the most informative ones. A supervised classifier validates the quality of these selected sentences, ensuring their suitability for MCQ generation. The final stage involves the creation of MCQs using a combination of rule-based templates and GPT-2 for generating distractors. The model's efficacy is confirmed by experimental assessments utilizing RMSE, Pearson correlation, and Spearman correlation. The algorithm achieves 89.2% accuracy in creating high-quality multiple-choice questions and 90.5% accuracy in recognizing informative sentences.