Examining the OpenAlex Concepts: A Detailed Case Study of Machine-Derived Classification

Abstract

Machine-learning techniques are becoming increasingly popular in metadata and classification work due to their ability to operate at scale, but insufficient consideration has been given to how effective such techniques truly are against traditional practice. This thesis adopts an approach based on Data Feminism (D'Ignazio & Klein, 2020) to analyze the machine-generated OpenAlex concept hierarchy and its associated machine-learning model in comparison to established classification standards and practices. We find that the OpenAlex concepts differ vastly from a traditional classification system, and that this difference inhibits their effectiveness in some respects, while also offering possible ways to address modern criticisms of classification (Olson, 2001; Mai, 2005). We argue that statistical, data-processing approaches to classification cannot replace the human judgment necessary for classification work, and that, to the extent that they continue to be used in this type of work, automatic techniques should be more informed by information theoretical principles and guided by human expertise.