Updated:03/05/2020 01: 30h
Facebook is the largest social network in the world, it has nothing more and nothing less than 2,449 million active users per month, these are not the number of accounts it hosts, but the number of users who logged in or participated for thirty days, regardless of the 1.3 billion users that use Facebook Messenger.
The popularity of the social network has become the target of cybercriminals attacks, data theft, electoral influence, impersonation and much more, after all human beings have done nothing but move all the world’s problems real to virtual
Facebook has been experimenting with artificial intelligence and machine learning technologies for a long time to be able to tackle the effects caused by accounts that violate its usage policies. The manual control of almost 2.5 billion accounts is virtually impossible, not to mention the millions of accounts that are created daily on the social network. One of the most common infractions on Facebook is the creation of fake accounts.
These can be classified into two categories, those that do not belong to a real person and have been identified as such, for example, a pet account cannot be a Facebook user, but rather a page. In these cases, Facebook automatically converts them into pages as they are value-added accounts, but not natural persons. The second type is the most worrying, are those fake accounts that are created to violate Facebook’s terms of service in order to obtain some revenue, such as those dedicated to spam or scams, such as recent cases of cryptocurrency sales.
Facebook deactivated around 2,000 million false accounts between January and March 2019, of which 99.7% occurred proactively, that is, before any user reported them. The way to deal with these types of accounts is divided into three levels. The first is before the account itself is created, through the forms that the user has to complete, many accounts that could be potentially false are discarded, if there is suspicion, Facebook asks for more information.
Asking for the phone number is a great barrier for the creators of fake accounts because generating fake phone numbers is simply expensive. The next level is when the account is created, and before it is active it is deleted. In this phase most of the fake accounts fall, before coming into contact with another user. And finally, there are the accounts that surpass all the controls and manage to be active within the social network, these suppose around 5% of monthly active accounts, that is, 122 million accounts, three times the population of Spain. The latter are often created manually by transgressors and subsequently detected using machine learning.
The new machine learning process that Facebook has been implementing for the last two years to detect these accounts by tracking behavior patterns is called Deep Entity Classification (DEC). Basically this system analyzes the graph of the user, or what is the same, their interactions and connections, instead of staying only in the direct features of the fake account, which are easier to overcome, check up to 10,000 points direct and indirect environment to an account The labeling of these behaviors occurs in two ways, automatically for large volumes and low precision, and human for low volumes and high precision, and Both are combined in two phases to obtain the best result.
Facebook is in a constant race against the creators of fake accounts, which also use technology to overcome all the technological barriers that the social network puts. Deep Entity Classification is the next step that promises to improve with time learning of users and ensure that Facebook accounts They are 100% authentic.