Identify, obtain, explore

using NLP to link article and journal records in the NHM library catalogue

Abstract

This paper addresses the critical need to link related records in library catalogues, particularly for aggregated volumes and serial articles in physical collections, to enhance user access and discoverability. The Natural History Museum (NHM) Library, facing a significant backlog of unlinked article-level records within physical journal holdings, developed a semi-automated solution. We outline a two-stage pipeline utilizing natural language processing (NLP) and record linkage techniques to automate the matching of article (child) records to journal (parent) records. This approach, leveraging title similarity and metadata extraction, achieved 60% accuracy against a previously-linked dataset. We discuss the challenge article level records can present and how such computational methods can optimise metadata workflows, allowing human effort to be redirected to more complex tasks and improving access across vast collections with limited resources.