MEASURING MISOGYNY: DEPP V HEARD AND THE LIMITS OF ATOMISTIC CONTENT MODERATION

Abstract

Over the past decade, social media companies have come under increasing pressure to make their platforms safer for women. While they have made some changes to their policies and design, their foundational approach to content moderation has remained largely the same. Platforms continue to focus on identifying, assessing and responding to individual pieces of violating content, like overt hate speech, direct threats, and doxxing. This approach is ill-equipped to deal with structural harms like misogyny, which cannot be understood in terms of isolated instances, but instead as part of a continuum. ‘Everyday’ experiences of sexism and misogyny form part of the same dynamic as the more widely recognised, extreme forms of violence, but existing tools for identifying this type of harmful but not prohibited content are extremely limited. This paper presents the preliminary findings of a study that investigates how everyday misogyny manifests on social media platforms, using the online discourse around the $2 trial as a case study. We use a combination of topic modelling and in-depth qualitative analysis of content, informed by the literature on believability and doubt in cases of domestic and sexual violence. We find that everyday misogyny is widespread, and manifests partly in double-standards and double-binds in the expectations imposed on Amber Heard. We aim to use these findings to develop new methods and frameworks for understanding misogyny in aggregate.