LLMs for the Post-Hoc Creation of Provenance

Provenance information is an essential component that facilitates the reproduction of scientific experiments, the assessment of data quality, and other related tasks. However, provenance capture at observation is sometimes difficult, and post-hoc methods are needed. In this paper, we explore the ability of large language models (LLMs) to access and extract provenance information from scientific papers through a set of specially designed prompts. We then identify and suggest the most effective prompt for provenance extraction from papers. Our findings confirm the capability of ChatGPT-4 in accessing and extracting provenance information from biomedical research papers.