A large proportion of scientific studies now rely on software and data as an integral component of the research process. Significant time and resources are committed to the development of research software yet, too often, these valuable assets lie languishing, hidden in the original research paper that presented them. Ensuring the availability of software and data, and directly linking these assets to the research that first introduced them, is a key component in addressing current problems faced by many scientists when attempting to replicate earlier studies. There have been a number of efforts in recent years to develop methodologies for the extraction and classification of software mentions found in full text scholarly documents. In this presentation, we will discuss how large language models can match current SotA approaches to the problem utilising zero-shot methods that require no pre-training.
Pride et al. (Tue,) studied this question.