Abstract
Abstract This study introduces a dynamic optimization technique for the min-df (minimum document frequency) parameter in the GreedSum algorithm, aimed at improving extractive summarization. We applied geometric, percentile-based, and clustering-based methods on a dataset of 17,038 scientific articles from arXiv and PubMed.The results indicate a 2% improvement in ROUGE-1 F-measure compared to fixed min-df settings, achieving a peak F-measure of 45%. These findings demonstrate the potential of dynamic min-df tuning for better summa-rization, with future research directed at integrating deep learning models to further enhance accuracy and efficiency.