Some Reddit users just love to disagree, new AI-powered troll-spotting algorithm finds

In today’s fractured online landscape, it is harder than ever to identify harmful actors such as trolls and misinformation spreaders.

Often, efforts to spot malicious accounts focus on analysing what they say. However, our latest research suggests we should be paying more attention to what they do – and how they do it.

We have developed a way to identify potentially harmful online actors based solely on their behavioural patterns – the way they interact with others – rather than the content they share. We presented our results at the recent ACM Web Conference, and were awarded Best Paper.