Machine learning in the library: Developing an inter-departmental core solution to manage data

Abstract

The Oklahoma State University Archives identified the need for an updated, comprehensive inventory of its digital assets to guide the development of digital preservation priorities. Creating it was complicated by sparse records, limited manpower and dependence on fading institutional memory as well as poor data management. A strategic planning process was launched to address these deficiencies. Machine learning (ML) was identified as a promising tool to minimise the labour-intensive process of sorting artefacts and identifying records that needed to be augmented, cleaned or eliminated from the collection. A pilot project to explore the effectiveness of using ML to curate a high-value archival collection was implemented. This paper describes the nature of ML, its promise and limitations for use in archives, and the outcomes of the pilot project. In particular, the pilot project showed promising results in the application of facial recognition techniques. Collaboration with interested colleagues in other departments suggests that ML can be widely applied to projects throughout the library.