Understanding content moderation systems: new methods to understand internet governance at scale, over time, and across platforms
Nicolas Pierre Suzor · Edward Elgar Publishing eBooks · 2020
This chapter introduces the ‘Platform Governance Observatory’, a set of tools that continuously collect public social media content and identify content that has been removed. The Observatory, developed at QUT’s Digital Media Research Centre, currently assembles random samples of public posts on YouTube, Twitter, and Instagram. Each post is collected as close as possible to the time it is posted, and its availability is tested 8–20 days later. This infrastructure allows researchers to identify which content has been removed and, depending on the information provided by each particular platform, who was responsible for removing it and why. The Observatory has been running stably since March 2018, and has so far collected usable random samples for hundreds of millions of posts to YouTube (metadata only), Twitter, and Instagram. The Platform Governance Observatory infrastructure enables scholars to undertake in-depth analysis of removal rates and to develop large-scale longitudinal analyses across multiple platforms. With this infrastructure, researchers are able to interrogate the types of content that are being moderated on social media platforms – including, importantly, measures of false negatives and false positives according to each platform’s terms of service. Uniquely, this infrastructure allows scholars to evaluate content moderation in practice and over time, as well as to measure the effects of rule changes, technical interventions, and external events on rates and removal rates of harmful content. This research infrastructure provides an important leap forward in the ongoing task of holding social media platforms to account for their content moderation practices through rigorous independent analysis, moving beyond the high-level statistics released by platforms themselves as part of regular transparency reporting.