American scientists have developed an algorithm that can recognize the messages of Internet trolls and bots. Artificial intelligence (AI) can identify features of content-content that are typical of such messages.
the”Content-centric functions of the system,” the authors write in Science Advances, able to work without human intervention, however, the evolving tactics and strategies of trolls are demanding that the AI is constantly studied.
a Team of Mazama Alizadeh from Princeton University (USA) noted that their study was based on a specific, particularly common type of post in social networks. In total, they analyzed 7.2 million messages – as trolls and normal users.
After training the system on data set AI learned how to define texts for the trolls: the algorithm takes into account the volume of posts, their content, and keywords.
the Tests showed that the AI can not only find the posts of trolls, but also calculate them in other social networks.
According to scientists, the easiest way to discover Venezuelan Troll harder Chinese, and Russian.
search Algorithms on the basis of content, say the authors definitely give a chance to counter the flood of fake messages on different platforms. “The algorithm can be used to assess in real time how many trolls on the platform and what they’re saying,” says study co-author Jacob Shapiro.
the developers of the system acknowledge that “Troll detection is not perfect, but it can force them to be more creative or even to stop their campaign.”
