Metadata Enhancement Using Large Language Models: Improving the Quality of Aggregated Records in the iSamples Project

Abstract

Aggregating resources from diverse data repositories often results in an incomplete ag- gregation despite curation efforts in the digital library space. To address this challenge, in this work we propose a method for automatic metadata assignment that is applicable across different metadata aggregations. Our approach uses zero and few-shot methods which harness the knowledge and reasoning abilities of large language models, while also incorporating the hierarchical structure of the taxonomy. We also illustrate how our approach can be integrated within a digital infrastructure in practical settings.