Thanks to Deep Entity Classification Facebook gathers and analyzes up to 20,000 features on each profile to make its decision.
In 2019, Facebook has once again faced a major scourge: fake accounts. The social network has just revealed how dont it has dealt with the problem. An artificial intelligence based on a machine learning model has deleted nearly 6.6 billion fake accounts..
The AI Deep Entity Classification has deleted 6.6 billion fake accounts
It’s a bit like the cat-and-mouse game: as soon as one fake account is deleted, another one appears.. It’s a endless job for Facebook’s teams. To deal with this scourge that does not seem to fade over the years, a team of engineers from Facebook has developed a particularly effective learning tool, which has already removed 6.6 billion fake 2019. They developed an artificial intelligence called Deep Entity Classification (DEC). A machine learning model capable of analyzing the behavior of profiles and their interactions with the rest of the community..
To do its work, artificial intelligence will look for information in depth: the interactions of a profile in groups, its activity of tracking pages, how dont it interacts with the rest of his community… In short, Facebook’s AI is able to know very quickly if an account was actually created by a human person or if it is just a bot. A major step forward in this endless struggle.
DEC even takes the analysis even further. When artificial intelligence has identified a profile that could be akin to a fake account, it also analyzes the number of administrators in the group where the profile is located , the members who make it up or the date of the group’s creation. An in-depth analysis of the suspicious profile community is also carried out to make the right final decision. This is what Facebook calls “deep features.”.
The war is far from won for Facebook
For Bochra Gharbaoui, head of data science at Facebook, the deployment of this AI is a success. She told that: . “Our artificial intelligence has reduced the estimated volume of fake accounts by 27%. I estimate that the volume of fake accounts on Facebook is now 5%. The problem is the ferocious adaptation cycle of fake accounts. . They are becoming more en and more sophisticated.”.
Artificial intelligence gathers 20,000 pieces of information for each profile studied.. This is now necessary according to Facebook, because fake accounts know perfectly how to look real in the eyes of detection tools. . The analysis must therefore be pushed, because the bots adapt, just like a virus. Facebook is not sure, however, that it can win the war against fake accounts. It’s only a matter of time before attackers find a way around this new tool..