/
Navigation
Chronicles
Browse all articles
Explore
Semantic exploration
Research
Entity momentum
Nexus
Correlations & relationships
Story Arc
Topic evolution
Drift Map
Semantic trajectory animation
Posts
Analysis & commentary
Pulse API
Tech news intelligence API
Browse
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
Concept Search
Semantic similarity search
High Impact Stories
Top coverage by position
Sentiment Analysis
Positive/negative coverage
Anomaly Detection
Unusual coverage patterns
Analysis
Rivalry Report
Compare two entities head-to-head
Semantic Pivots
Narrative discontinuities
Crisis Response
Event recovery patterns
Connected
Search: /
Command: ⌘K
Embeddings: large
TEXXR

Chronicles

The story behind the story

days · browse · Enter similar · o open

Facebook says bullying and harassment takedowns reached 6.3M in Q4 2020, up from 2.8M in Q4 2019, due to increased reviewer capacity and AI improvements

Adi Robertson / The Verge :

The Verge Adi Robertson

Context & Ripple Effects

Facebook's earlier transparency updates showed automated detection taking a larger role in hate-speech enforcement, from 88.8% proactive detection in Q1 2020 to 95% of Q2 violations caught by automated systems. The new bullying-and-harassment figure places increased reviewer capacity alongside that automation push, showing the company scaling enforcement across another policy category.

First-order effects

  • Facebook's moderation operation acted on 6.3M bullying-and-harassment items in Q4 2020, exposing more violating content to removal than a year earlier.
  • Greater reviewer capacity and improved AI become immediate operating inputs to Facebook's enforcement volume, rather than removals being attributable solely to changes in reported content.

Second-order effects

  • Facebook's stated causes mean its removal totals are a less direct indicator of how much bullying and harassment users are posting; capacity and detection improvements can raise the count independently.
  • The company’s transparency reporting increasingly has to be read alongside detection and staffing changes, as it already has for software-identified hate speech.

Third-order effects

  • If Facebook continues to extend the same human-and-automated model across policy areas, content moderation will be organized more around detection-system coverage and reviewer throughput than manual case-by-case review alone.

The trend: Platform moderation is becoming a scaled hybrid of AI detection and human review, with enforcement totals increasingly reflecting the capacity of that system.

Discussion

  • @guyro Guy Rosen on x
    Today we released our Q4 2020 Community Standards Enforcement report. Our goal is to lead the industry in transparency about this work, and we'll continue to share more as part of this effort. https://about.fb.com/... (1/5)
  • @atef @atef on x
    #AI gets better every day. Here's what that means for stopping #HateSpeech. #Facebook https://about.fb.com/...
  • @fbnewsroom @fbnewsroom on x
    AI gets better every day. Here's what that means for stopping hate speech. https://about.fb.com/...
  • @mosseri Adam Mosseri on x
    Today we're sharing our latest Community Standards Enforcement Report, a report we now publish quarterly to track our progress in keeping people safe on Facebook and Instagram. More info in thread below 👇🏼 https://twitter.com/...
  • @kingjeff2 Jeff King on x
    The work is never done but it's the right work to do. So proud of all of our teams working together on this. https://twitter.com/...
  • @evelyndouek Evelyn Douek on x
    This just clearly shows that the take-down/leave-up binary of content moderation is missing the important, far more consequential decisions around algorithmic ranking https://about.fb.com/... https://twitter.com/...