Comparison of Approaches to the Extraction of Mathematical Methods from Scientific Texts

Abstract

The processes of extracting and comparing mathematical methods from scientific publications using different approaches—large language models, machine learning based classification method, and probabilistic topic modelling—are discussed. The superiority of the model obtained with probabilistic topic modelling when studying each article separately and of the large language model when studying whole projects is revealed, as well as the significant superiority of combining the results of these two approaches.