A new machine learning algorithm analyzes user behavior to detect anomalies. More than 6.6 billion false accounts were detected in the past year.
At Facebook, the fight against fraud and disinformation is increasingly done thanks to artificial intelligence. According to US media reports, the social network has developed a new machine learning model capable of detecting fake accounts. Last year, this software would have removed more than 6.6 billion false identities on the platform. It's enormous.
To achieve this, this artificial intelligence examines more than 20,000 “deep features” of each account (“deep features”). It is not only raw data entered by the user, but above all behavioral data: with whom does it communicate? How frequently ? How many invitations does he send? In which groups is he active? how does he behave in these groups? etc.
The complexity of this model allows Facebook to outstrip the vast majority of fraudsters, at least for now. Because the latter do not remain swinging arms and perpetually try to circumvent the protections of the social network. Facebook currently estimates that about 5% of its 2.98 billion active users are fake people. And every day, it continues to block millions of attempts to create fake accounts. An endless fight.