ASLM-AQG: Attention-based Sequence Learning Model for Automatic Question Generation
Chidanand, B. Sujatha, P. Nagamani, N. Leelavathy · 2025
This paper introduces an innovative Attention-based Sequence Learning Model (ASLM) for automatic question generating (AQG) that makes use of sentences extracted from reading comprehension literature. Using sequenceto-sequence learning to encode information at the paragraph and sentence levels, our data-driven approach surpasses the most advanced rule-based systems in automatic evaluations. The model is evaluated on the processed SQuAD dataset, which includes 536 articles and over 100,000 questions generated by crowd workers. To prepare this dataset, we first preprocess it using Stanford Core NLP for tokenization and sentence splitting, followed by converting all text to lowercase. Human evaluations confirm that our model generates more natural and challenging questions, requiring deeper reasoning and syntactic variation. Comparative analysis against competitive baselines demonstrates the model’s superior ability to generate high-quality questions. Furthermore, ANOVA analysis reveals a strong correlation between evaluation metrics and question types, validating the model’s robustness. This work marks a significant advancement in QG for reading comprehension, with potential applications in the educational domain, particularly for generating reading comprehension questions.