Measuring racism and related concepts using computational text-as-data approaches: A systematic literature review

Racism and related concepts such as racist stereotypes and targeted hate speech are increasingly measured using the methodological toolkit of computational social science. While computational text-as-data approaches have many advantages, such as reducing the exposure to disturbing content for human coders or scalability, they also pose challenges for sensitive concepts, such as oversimplification and validity. To shed light on how racism and related concepts are computationally measured in textual data, we provide the first systematic literature review in this area, examining 115 relevant publications. We identify four common measurement pipelines used to study racism and related concepts. We find a wide variety of concepts under study, a strong dominance of social media data, especially from Twitter, and a strong preference for supervised classification models for this task. By critically discussing the current state of research, we identify gaps and provide recommendations for future research.