Automated Parsing of Personal Identity Facets for a Collection of Visual Images

Abstract

Collections of digitized, historical images serve as rich primary sources for digital humanities research, though access to these resources has been hindered by inadequate subject metadata. In this study, researchers explored the feasibility of performing subject analysis for a collection of historical images of persons through an automated procedure. Building on previous work that developed a faceted system for representing the identities of persons depicted in 19th century visual images, the present work attempted to automate the process of person and facet parsing for images from the A.S. Williams III Collection at the University of Alabama. A case-based model was built and used to analyze image titles. Compared to a manual control process, the automated model achieved a 95% success rate in parsing persons and an 85% success rate in parsing facets. Errors in parsing were more likely to occur for images of multiple persons, as well as those labeled with incomplete or uncertain names. Findings offer further support for faceted analysis of personal identity in historical materials, and reveal the potentials of automated, text-based methods of enhancing subject access for large visual image collections.