When it comes to cyberattacks, social networks have become the main focus of these types of situations, where it is possible to find a large number of accounts created especially to spread malware, spam or phishing links.
In the case of Facebook, this social network has been implementing strong measures to combat all these accounts with malicious activity.
How? Through artificial intelligence, which, in 2019, allowed an average of 2 billion accounts to be deleted by Facebook, only in a quarter, thus cancelling the actions of those who had reached the platform to commit crimes IT mentioned above.
This made Facebook decide to reveal details regarding the machine learning system they use to intercept this destabilizing group from the inside.
Types of fake accounts
In this sense, the blue giant of social networks establishes two types of false accounts: those corresponding to personal profiles for businesses or pets created under the figure of pages, which are easy to cancel for Facebook.
Then there are those considered "violating accounts" referring to personal profiles actively engaged in scam and spam or that in some way violate Facebook's terms and conditions, these being the type of accounts whose removal represents a priority, to prevent them from establishing a network and damaging real user accounts.
To take action on this, Facebook uses hand-coded rules in conjunction with machine learning to stop the false account before it is activated, an action that, if fulfilled, Facebook would go to the next phase in which its new machine learning system called Deep Entity Classification (DEC).
When the DEC enters into action, it has a large number of labels generated by low-precision machines as a result of a combination of learning rules and models with the ability to estimate the falsehood or truthfulness of a user.
This data is used to train a neural network, which is then tuned with a lot of high-precision data labelled by hand, provided by people from all over the world who are familiar with the local dialect.
Thus, in the end, the classification system acquires the ability to distinguish between 4 types of false profiles:
Illegitimate accounts not representative of the person.
Legitimate accounts hacked by attackers.
According to statements offered by Facebook, since its application, the DEC has significantly reduced the number of fake accounts on the platform keeping them at 5% concerning the number of real active users.