Facebook shuts down billions of accounts: who’s at risk

Facebook has introduced a new algorithm that automatically deletes all fake accounts. Here's how it works and who it affects

Maybe Facebook has finally found a way to take out fake profiles and trolls of various kinds, and it's an extremely complex method called "Deep Entity Classification" (DEC). The results? Great, according to Facebook: 6.6 billion fake profiles deleted in 2019.

Stunning results, to which are added millions of attempts to create fake profiles nipped in the bud every day, again thanks to DEC. Numbers that, indirectly, also indicate how huge is the industry of fake profiles on social networks. Facebook in primis, but not only Facebook. Facebook has responded to such a big challenge with an equally big algorithm: inside the DEC there are more than 20 thousand parameters analyzed for each profile, in order to be sure that you are not deleting a legitimate profile and, at the same time, to make the algorithm much more difficult to cheat.

Facebook DEC: how it works

The first "anti-fake" algorithms of Facebook took into account a few parameters, such as the amount of friendship requests sent in a certain period of time by the suspect profile. The first thing you do with a fake profile, in fact, is to send rapid-fire requests to users from all over the world. However, it was easy for the fake profile industry to reprogram the strategy by spreading out the friend requests in order to bypass the filter. With DEC, on the other hand, Facebook takes into account not only a profile's network of friends and the friend requests sent, but also the same parameters applied to the contacts of the analyzed profile.

DEC, moreover, also looks at the inside of groups, the profile's daily activity, private messages and many other parameters to be sure that the profile candidate for deletion is really fake.

Facebook DEC: risky behaviors

Facebook claims that the estimated number of fake profiles on the social network today does not exceed 5%, a somewhat physiological value and due to the fact that you can't tighten the grip of the algorithm too much or you risk taking out real profiles, but with suspicious behavior. Last year, for example, the social network took out 900 pages, groups and fake profiles that used very sophisticated methods to spread pro-Trump propaganda: profile photos generated by AI algorithms, a complex network of simulated relationships, interactions, likes and cross-shares and, above all, a command center in Vietnam that managed all this. Such behavior is clearly at risk and, sooner or later, should suffer the axe blow of the new Deep Entity Classification.