How it Works

Using Machine Learning to Reduce Toxicity Online

Perspective uses machine learning models to identify abusive comments. The models score a phrase based on the perceived impact the text may have in a conversation. Developers and publishers can use this score to give feedback to commenters, help moderators more easily review comments, or help readers filter out “toxic We define toxicity as a rude, disrespectful, or unreasonable comment that is likely to make someone leave a discussion.” language.

Perspective models provide scores for several different attributes. In addition to the flagship Toxicity attribute, here are some of the other attributes Perspective can provide scores for:

  • Severe Toxicity
  • Insult
  • Profanity
  • Identity attack
  • Threat
  • Sexually explicit

To learn more about our ongoing research and experimental models, visit our Developers site.

Language availability

Perspective API is free and available to use in English, Spanish, French, German, Portuguese, Italian and Russian. The team is constantly developing models to support new languages.

To learn more about international publishers and platforms using Perspective API, check out our case studies. To learn more about languages in development, visit the Developers site.

Applying the model
Perspective in Action

One of Perspective API’s most common uses is content moderation. Developers at platforms like DISQUS and publications like The Wall Street Journal augment their existing content moderation systems with Perspective, creating custom combinations of attributes to suit the needs of their platforms.

Perspective API is meant to make content moderation less of a burden on individuals and organizations, but is not meant to completely replace the work of human decision-makers. Perspective API helps developers and moderators make better decisions at scale, allowing healthy dialogue to flourish.

Read more about how our partners use Perspective on our Case Studies page.

How to customize Perspective to your needs

Choose the combination of attributes you’d like to use
Choose the combination of attributes you’d like to use
Choose which language you’d like to support
Choose which language you’d like to support
Determine what actions should be triggered based on the output
Determine what actions should be triggered based on the output

Looking to learn more? Visit our Developers site for more technical information.

Go to developers site