Researchers at Cornell University conducted a study using a mixture of Google’s Perspective API and human intelligence to test predictions of when conversations might go sour. They analyzed 1,270 conversations in Wikipedia editors’ forums that had begun civilly enough but devolved into hostility (they used Perspective API to measure “toxicity”). They lined up these conversations with ones on the same topic that went well and found that when the first comment in an exchange used direct questions or started sentences with “you,” the conversation was significantly more likely to go awry. Comments
...more
