Using racial discourse communities to audit personalization algorithms

Abstract

Abstract Personalization algorithms are the information undercurrent of the digital age. They learn users’ behaviors and tailor content to individual interests and predicted tastes. These algorithms, in turn, categorize and represent these users back to society—culturally, politically, and racially. Researchers audit personalization algorithms to critique the ways bias is perpetuated within these systems. Yet, research examining the relationship between personalization algorithms and racial bias has not yet contended with the complexities of conceptualizing race. This article argues for the use of racialized discourse communities within algorithm audits, providing a way to audit algorithms that accounts for both the historical and cultural influences of race and its measurement online.