Abstract

Automated response generation is critical in NLP, which improves question-answering (QA) systems. Current models are not fluent, semantically diverse, and accurate in their answers. This research presents a QA system based on Google Flan-T5-Base for question generation and Deepset RoBERTa-Large-SQuAD2 for answer extraction. The system takes text and PDFs as input to produce question-answer pairs. Performance is measured based on fluency, question diversity, semantic diversity, and confidence distribution. Results show high fluency (1.0000), enhanced semantic diversity (0.5047), and high answer relevance (BERTScore 0.9203, METEOR 0.4353). Comparative analysis reveals better coherence and diversity. As a Flask web app, the system pushes the boundaries of NLP-based QA generation.

Authors

D. Veda Valli, Shaik Kaif Mohammad, M. Reshmika, G. Santhosh Naveen Teja
Gayatri Vidya Parishad College of Engineering, India

Keywords

Automated Question Answering, Natural Language Processing (NLP), Question Generation, Answer Extraction, Pretrained Language Models