The problem is not new, but in recent weeks there has been a lot of misinformation. It must be said that the contestation of the results of the American presidential election on social networks and the violent invasion of the Capitol struck the spirits. Since then, the big platforms have tried to stem this disturbing phenomenon and the debate continues.
In this perspective, a new artificial intelligence system developed by researchers from the University of Sheffield could be of great help. It would identify a user who publishes fake news even before it is massively shared.
79.7% accuracy
To achieve this result, the scientists worked on a sample of over a million tweets from 6,200 profiles. A list of credible sources was then established and it includes media such as the BBC or Reuters. Other portals reputed to be misleading, including Infowars and Disclose.tv, have also been listed.
The 6,200 users were then categorized into two groups: those who post links to disinformation sites more than three times and those who have a habit of sharing information deemed to be trustworthy. This selection was finally completed with criteria related to the vocabulary used in the tweets because the researchers noted that the use of hateful and violent speech often went hand in hand with the sharing of fake news.
The result is ultimately satisfactory since the scientists announce that their algorithm obtains an accuracy of 79.7% in the detection of users who publish misleading content. They are now hoping that their work will be able to help major social networks better fight disinformation.
Note that AI is already heavily used by web giants to moderate content. For its part, the WHO is using machine learning as part of its fight against fake news.
—
