Automated Research Paper Summarization with Multiple Model and Accessibility Enhancements

Abstract

The number of research papers published each year is growing at an overwhelming pace, making it difficult for students, researchers, and professionals to keep up with new knowledge. Existing summarization tools can help, but most of them rely on large models like GPT, Pegasus, or BERT, which need powerful hardware and constant internet access. This limits their use, especially in low-resource or offline environments. This work introduces a novel framework for Automated Research Paper Summarization that employs a multi-model hybrid pipeline, integrating both extractive and abstractive strategies. Unlike resource intensive models, this approach emphasizes lightweight architectures, enabling efficient performance even in low-resource settings while preserving summary quality. To further enhance usability, the system includes keyword extraction modules that highlight central concepts and accessibility features such as text-to-speech, supporting users with visual or cognitive challenges. A distinctive feature of this framework is its section-wise summarization output, which mirrors the logical flow of research papers allowing users to quickly access context, methodology, findings, or conclusions as needed. System performance is assessed through standard metrics like ROUGE and BLEU, complemented by qualitative evaluations of readability, informativeness, and coherence. By avoiding full dependence on large, pre-built models such as GPT or Pegasus, this work prioritizes component level innovation, offline functionality, and greater privacy, making it adaptable across diverse use cases. The study advances the field of scientific summarization by offering a practical, modular, and accessible tool that supports knowledge discovery and management in research intensive domains.