Facebook’s efforts to minimize the prevalence of violating content have shown promising results

Facebook's efforts to minimize the prevalence of violating content have shown promising results

KEY POINTS

● Prevalence, defined as the estimated number of views showing violating content divided by the total content views, serves as a crucial metric for evaluating FB’s performance in upholding its policies and minimizing the impact of violations on its user base.

● Notably, the prevalence rate for hate speech experienced a substantial improvement over the years. From approximately 10-11 in 2020 Q3, the prevalence decreased significantly to 2 by 2023 Q4.

● From 2019 Q4 to 2023 Q4, the total number of actions taken on FB and Instagram for adult nudity & sexual activity violations has increased gradually between 2017 and 2023.

Table of Contents

Facebook (FB) actively monitors and assesses the prevalence of violating content on its platform to ensure a safe and enjoyable experience for its users. Prevalence, defined as the estimated number of views showing violating content divided by the total content views, serves as a crucial metric for evaluating FB’s performance in upholding its policies and minimizing the impact of violations on its user base. In 2019 Q1, the prevalence of content violating FB standards related to adult nudity & sexual activity reached an all-time high, with an average of 12-14 violations per 10,000 content views. However, this figure witnessed a significant decline over the subsequent years. The prevalence dropped to approximately 7 violations per 10,000 views in 2023 Q4. The prevalence rate for bullying and harassment exhibited a gradual decrease from 10 in 2021 Q3 to 8 by 2023 Q4. This downward trend reflects FB’s ongoing efforts to combat such harmful behaviours on its platform.

Fig: 1

Notably, the prevalence rate for hate speech experienced a substantial improvement over the years. From approximately 10-11 in 2020 Q3, the prevalence decreased significantly to 2 by 2023 Q4, indicating successful mitigation of hate speech content on Facebook. The prevalence rate of incitement and violence also showed improvement from Q4 of 2017, when it was relatively high at about 16–19. The prevalence has slightly increased in recent years despite a sharp decline after 2019, underscoring the continued difficulty in effectively addressing violent and graphic content (fig: 1).

Facebook’s efforts to minimize the prevalence of violating content have shown promising results across various categories. The decline in prevalence rates for adult nudity & sexual activity, bullying and harassment, hate speech, and violence and incitement signifies FB’s commitment to maintaining a safer and more inclusive online environment. However, challenges remain in sustaining these improvements and effectively addressing emerging content moderation issues. Continued vigilance and proactive measures will be essential in reducing the prevalence of harmful content on the platform.

Fig: 2

Facebook (FB) measures the number of pieces of content (such as posts, photos, videos, or comments) or accounts that it takes action on for violating FB standards. This metric reflects the scale of enforcement activity, which may include removing content, applying warnings to disturbing photos or videos, or disabling accounts. Actions escalated to law enforcement are not additionally counted. While tempting, interpreting the content-actioned metric as an indicator of FB’s effectiveness in finding violations or their impact on the community should be done cautiously. The volume of content actioned is only part of the picture, failing to reflect the detection time or the frequency of user exposure to violations.

Adult nudity and sexual activity have emerged as one of the most actively moderated areas, showing considerable fluctuation but generally increasing from 20.8 million actions in Q4 2017 to 55 million in Q4 2023. This suggests both improved detection capabilities and possibly increased prevalence of such content. Bullying and harassment enforcement has also grown substantially, from virtually no actions in 2017 to consistently handling 12-17 million cases quarterly by 2023. This reflects increased attention to user safety and improved ability to detect more nuanced forms of harmful content. Child safety enforcement has undergone significant changes. While early quarters showed minimal actions, by 2021, Meta began explicitly tracking and acting on child endangerment in multiple categories. The sexual exploitation category, in particular, has shown consistent attention, with actions ranging from 17-35 million per quarter in recent years. This indicates both enhanced detection systems and a strong priority on child protection. Dangerous organizations, including terrorism and organized hate groups, show steady enforcement with some quarterly variations. Terrorism-related actions have generally ranged between 5-18 million per quarter in recent years, while organized hate group enforcement has typically remained under 5 million actions per quarter.

Hate speech moderation has shown one of the most dramatic increases, growing from 1.6 million actions in Q4 2017 to a peak of 41.3 million in Q2 2021 before stabilizing around 15-28 million actions quarterly in recent periods. This reflects both improved automated detection capabilities and increased focus on addressing hate speech globally. Regulated goods enforcement, covering drugs and firearms, has remained relatively stable at lower volumes compared to other categories, typically ranging from 1-10 million actions per quarter, which indicates either a relatively consistent prevalence of such content or stable detection capabilities in these areas. Spam enforcement shows interesting patterns, with generally high volumes ranging from 413 million to 2.9 billion actions per quarter, though showing a slight downward trend in recent years. This could suggest either improved preventive measures reducing spam or shifts in how spam is classified and handled. The same trend can be seen for fake accounts as well. The introduction of violence and violent incitement as a separate category in 2021, with consistent enforcement levels around 15-24 million actions quarterly, demonstrates Meta’s evolving content moderation framework and growing attention to various forms of harmful content (fig: 2).

The trends in enforcement activity on Facebook and Instagram reveal a dynamic landscape of efforts to uphold platform standards and ensure user safety. While some violation categories exhibit declining enforcement actions, others show fluctuations or increasing trends, highlighting the evolving nature of content moderation challenges. Continued vigilance and adaptability will be crucial in addressing emerging threats and maintaining a safe online environment for all users.

References

  1. Prevalence. (n.d.). Meta. Retrieved November 3, 2024, from https://transparency.meta.com/en-gb/policies/improving/prevalence-metric/
  2. Community Standards Enforcement Report. (n.d.). Meta. Retrieved November 3, 2024, from https://transparency.meta.com/reports/community-standards-enforcement/
  3. Integrity, G. R. V. O., & Meta. (2021, November 1). Hate speech prevalence has dropped by almost 50% on Facebook. Meta. https://about.fb.com/news/2021/10/hate-speech-prevalence-dropped-facebook/

 



About Author



 

Pankaj Chowdhury is a former Research Assistant at the International Economic Association. He holds a Master’s degree in Demography & Biostatistics from the International Institute for Population Sciences and a Bachelor’s degree in Statistics from Visva-Bharati University. His primary research interests focus on exploring new dimensions of computational social science and digital demography.

Disclaimer: The views expressed in this article are those of the author and do not necessarily reflect the views of 360 Analytika.

Acknowledgement: The author extends his gratitude to the Facebook for providing data support.

This article is posted by Sahil Shekh, Editor at 360 Analytika.

You May Like This