Exploring the utility of metadata record graphs and network analysis for metadata quality evaluation and augmentation

Abstract

Our study explores the possible uses and effectiveness of network analysis, including Metadata Record Graphs, for evaluating collections of metadata records at scale. We present the results of an experiment applying these methods to records in the University of North Texas (UNT) Digital Library and two sub-collections of different compositions: the UNT Scholarly Works collection, which functions as an institutional repository, and a collection of architectural slide images. The data includes count- and value-based statistics with network metrics for every Dublin Core element in each set. The study finds that network analysis provides useful information that supplements other metrics, for example by identifying records that are completely unconnected to other items through the subject, creator, or other field values. Additionally, network density may help managers identify collections or records that could benefit from enhancement. We also discuss the constraints of these metrics and suggest possible future applications.