Mapping Scientific Literature with Large Language Models and Topic Modeling

This study presents a large language model (LLM)-based framework for mapping scientific literature through topic modeling, applied to over 1,500 engineering articles from PNAS. The two-stage classification process achieves a 75.9% accuracy in categorizing articles semantically, outperforming traditional models in topic diversity and coherence metrics. This approach facilitates the identification of latent thematic connections, providing a novel method for researchers to navigate and analyze fragmented scientific domains.