TechPrufer
Get started
YouTube
YouTube logoYouTube · Feature trail

Misinformation Mitigation on YouTube

6 videos · Aug 2020 – Feb 2022

What this trail covers

YouTube's ongoing efforts to combat misinformation and harmful content have evolved significantly. Initially guided by a '4Rs of Responsibility' framework (Remove, Raise up authoritative sources, Reduce spread), the platform has introduced 'How YouTube Works' for transparency and detailed its policy enforcement mechanisms. This thread now includes the impact of the COVID-19 pandemic on enforcement strategies, where YouTube prioritized rapid removal of potentially harmful content using automated systems, accepting potential over-enforcement in sensitive areas to ensure maximum removal. This led to a significant increase in video removals, with a higher reinstatement rate on appeals due to the reliance on automation.

All YouTube trails
  1. Inside Responsibility: What’s next on our misinfo effortsThis post details YouTube's evolving strategies for combating misinformation. It introduces the '4Rs of Responsibility' (Remove, Raise up authoritative sources, Reduce spread) and discusses the challenges of identifying and mitigating fast-emerging, viral misinformation narratives. The post outlines efforts to leverage targeted classifiers, keywords in multiple languages, and regional analysts to catch new narratives. It also addresses 'data voids' by exploring new types of labels and disclaimers for topics lacking authoritative content. For cross-platform sharing, YouTube is considering disabling share buttons or using interstitial warnings for borderline content. Globally, the focus is on understanding regional nuances, investing in local teams and NGO partnerships, and updating models more frequently to catch hyperlocal misinformation. The post emphasizes building on transparency by sharing ongoing considerations and challenges.
  2. Building greater transparency and accountability with the Violative View RateIntroduced the Violative View Rate (VVR) metric to measure the percentage of views on YouTube that come from violative content. This metric is calculated by sampling videos and having content reviewers identify violations. The post details the methodology and historical VVR data, showing a significant decrease since 2017 due to investments in machine learning. The VVR is intended to provide greater transparency and accountability around content moderation efforts and to identify areas for improvement in protecting viewers from harmful content.
  3. Responsible policy enforcement during Covid-19This post details YouTube's shift to increased reliance on automated systems for content moderation during COVID-19 due to reduced human review capacity. It explains the choice between under-enforcement and over-enforcement, with YouTube opting for the latter to prioritize community safety. This resulted in a doubling of removed videos and a tripling of removals in sensitive areas like violent extremism and child safety, accepting a higher rate of false positives. The post also highlights measures taken to minimize disruption to creators, such as not issuing strikes on content removed without human review (except in high-confidence cases) and dedicating resources to handle an increased volume of appeals, which saw a doubling in reinstatement rates.