Filling the Holes in Porous Media: Metadata Assessment and Accessibility in the Digital Porous Media Portal

Abstract

The Digital Porous Media Portal (DPMP) is a repository of 176 datasets and terabytes of data. The applications of its data are scattered across a number of disciplines and areas of ongoing research including but not limited to geothermal engineering and aquifer mapping. Operators of the DPMP intend to make datasets easily accessible and searchable to facilitate data reuse. But understanding to what extent the DPMP is accessible requires knowledge of the metadata not initially available. This paper outlines the development of three metadata curation tools for the DPMP. The first of these replicates a tool used by other repositories to simply assess whether and which metadata fields are filled by each dataset. Recognizing the lack of nuance provided by this tool, another curation tool, backed by a Large Language Model, was created to grade and provide feedback on the content of dataset descriptions instead of merely availability. Having assessed metadata quality, the final tool—an AI chatbot (Rocco) backed by a graph database—was constructed to provide users with the ability to search the repository using natural language queries.

Keywords: data curation, porous media, large language model, graph database, AI, metadata, content assessment

How to Cite: