Meta should implement product and/or operational guideline changes that allow more accurate review of long-form video (e.g. use of algorithms for predicting the timestamp of violation, ensuring proportional review time with length of the video, allowing videos to run 1.5 times or 2 times faster). The Board will consider this implemented when Meta shares its new long-form video moderation procedures with the Board, including metrics for showing improvements in review accuracy for long-form videos.
Our commitment: Our long-form video review optimization processes and protocols remain a priority across our operational teams. We already have a number of established processes geared towards improving video review accuracy and we will continue to iterate on new improvements. We will also evaluate metrics to demonstrate ongoing improvements in review accuracy for long-form videos to the public and to the Board.
Considerations: We currently have operational guidelines in place that govern long form video review. Our protocol seeks to ensure accurate review of long form video while optimizing for the efficiency and wellbeing of our reviewers. Our guidelines include a designated review protocol, explicit minimum review standards including duration as well as video indicators such as violation and reporting timestamps and speed adjustments. Our guidelines also ensure that we apply progressive review on long form video including the assessment of title, description, caption, thumbnails, transcript and comments attached to the video under review.
We have developed new categorical guidance pertaining to live videos and provided reviewers with the corresponding action hierarchy for such instances, still optimizing for both reviewer precision and accuracy with widgets such as “watch this” indicators for potentially violating content as well as warning signs and interstitials. Additionally, we have a designated team that engages with live video reviewers on a recurring basis to surface protocol and tooling improvements. All our video review processes include video specific resiliency features such as muting video, adjustable skip lengths and blurring. We continue to innovate our efforts on video review with solutions such as AI thumbnail generation to flag potentially violating content, increase the accuracy of the existing thumbnails and support reviewer efficiency.
We have rolled out efforts to detect looping videos which are long-form types used by repeat offenders on our platform. We have completed automation to roll this out on Facebook and will scale to Instagram in the second half of 2023.
We continue to improve our products and protocols for long-form and live video review—this remains an ongoing priority area across our enforcement teams. We will also continue to assess the most effective way to use metrics to publicly demonstrate review accuracy for long form videos and share these in future updates.