But beware, only if they make less than 20 requests from friends.
In an article published on the blog dedicated to the speeches of the company’s engineers, Facebook unveiled, on April 22, 2020, an artificial intelligence capable of detecting fake accounts. This would be in addition to the abusive behaviour detection filters applied at the time of registration on the platform. The goal of the American company: to observe how real users do to add their friends and base the AI on this model to detect fake accounts.
SybilEdge, Facebook’s new algorithm
Facebook researchers and engineers recently introduced SybilEdge, their new algorithm that could soon save them time in the fight against fake profiles on the social network. With this new tool, Facebook wants to undermine the dissemination of false information, often the only activity of these accounts. The other goal is to protect users from different kinds of attacks.
According to the article, SybeilEdge could detect fake Facebook accounts in less than a week. The AI would be based on their friends list and applications made following registration. These are indicative of a person’s profile. Indeed, people hiding behind fake accounts will look for a target, often depending on a particular topic, and then send him a request for friends. This technique was probably used in the context of the Covid-19 pandemic and more specifically during the dissemination of a conspiracy theory related to the virus.
SybilEdge will then observe the facts: a fake account is more likely to accumulate refusals of requests to add a friend than a person creating his or her real profile. The algorithm would have been designed to categorize two types of users: those who accept false friend requests, and those who accept everyone, if the account seems real. The user who accepts a lot of people, whose fake profiles will allow engineers to know that the issuer of the request can potentially be considered a fake account.
More than 90% accuracy according to the social network’s engineers
According to Facebook, SybilEdge is more than 90% accurate to detect fake accounts that have issued 15 friend requests. The success rate drops to 80% for those who have made only 5. Facebook said of its new tool,“SybilEdge helps us identify malicious individuals quickly and in a way that can be explained and analyzed. In the near future, we plan to explore other ways that can further speed up the detection of abusive accounts and help make decisions with confidence even faster. We plan to achieve this by mixing feature-based models with behavior-based models.”