Toxicity on Twitter

Toxic communication is at an all-time high in our country. See it in real time.

  • How do you calculate toxicity? What’s this based on?

    Our AI writing assistant uses natural language processing and deep learning to make writing recommendations to business users, including grammar, style, tone, adherence to key terms and core messages, inclusiveness, and healthy communication. We have been doing a lot of thinking about how organizations can combat toxic communications in the workplace and that got us thinking about the Twittersphere.

    We started with our deep learning models that focus on healthy communication, and applied them to a combination of publicly available and internally-labelled data, and then fine-tuned the model on a large dataset of Tweets to make them more used to that style of writing. In our human-labeled data set, we defined “toxic” as aggressive, mean, offensive, racist, sexist or hateful commentary.

  • Why does the export only contain 1,000 tweets?

    It’s sample data! We’re proud to be Twitter developers and wanted to stick to something that would be representative while still sticking to the terms and conditions we agreed to when signing up to build on top of their data stream.

  • How do you define politicos?

    This was more art than science, as there’s no global ranking of follower count on Twitter, and certainly “left-leaning” or “right-leaning” are also open to interpretation. Here’s what we did:

    • Started with a long list of U.S. politicians at the federal level who are active on Twitter
    • Added hundreds of thought leaders, media personalities, former bureaucrats or political strategists, and activists who are active on Twitter and who mostly Tweet about national politics
    • Ordered the list based on each person’s follower count and posting frequency as a proxy for influence and reach on Twitter
    • Sorted people into “left-leaning” or “right-leaning” based on their stated political party affiliation, or as evidenced by Tweets and/or bio
    • Reduced each side to the top 100 based on our definition of influence
  • How do I provide feedback?

    We’d love to hear your suggestions. Email us at