Automatic Question Answer Generation Using NLP Techniques

EasyChair Preprint 15639

Abstract

The rising demand for scalable educational assessments and efficient question generation exposes the limitations of manual question construction, which is labor-intensive and difficult to scale with style and across domains. As a panacea to this, the "Automatic Question Answer Generator" uses the fine-tuned transformer-based T5 model on the SQuAD and RACE datasets to generate different types of questions: multiplechoice, fill-in-the-blank, factoid, and match-thefollowing. Advanced natural language processing techniques are employed: Keyphrase Extraction through PKE Library and TextRank for fill-in-theblank questions; Named Entity Recognition and Sense2Vec for generating distractors validated by cosine similarity in multiple-choice questions; and finally, for refining match-the-following pairs, using WordNet and BERT. Factoid questions are generated using wh-type structures to target important concepts. The proposed system is a sturdy solution that bolsters contextual relevance, question variety, and its adaptability across domains and thus may possibly rise to the occasion for their dynamic needs associated with modern educational assessment

Keyphrases: Automated Question Generation, Convolutional Neural Networks (CNNs), Natural Language Processing (NLP), RACE Dataset, SQuAD dataset, T5 model, deep learning

@booklet{EasyChair:15639,
  author    = {Keshav Vardhan Narkudi and Hari Haran Juluri and Srujana Inturi},
  title     = {Automatic Question Answer Generation Using NLP Techniques},
  howpublished = {EasyChair Preprint 15639},
  year      = {EasyChair, 2025}}