How Facebook is Using AI to Identify Fake Accounts

Facebook has released details about how it’s using machine learning to identify fake accounts on its platform.

Fake accounts are not uncommon on social media pages.

Last year, Facebook removed an average of roughly 2 billion accounts every quarter. Aside from spreading spam, fraudsters use these accounts to distribute phishing links as well as malware.

The social media company divided fake accounts on its platform into two categories — user-misclassified and violating accounts.

The latter refers to personal profiles for businesses or pets that are means to be Pages. So, they are relatively harmless, and Facebook converts them to Pages.

On the other hand, Violating accounts are personal profiles that violate the platform’s terms of service and engage in spamming and scamming. These are the accounts that Facebook is removing from its platform.

To handle the challenge, the tech company uses hand-coded rules and machine learning. So, Facebook can block a fake account before it’s created or before it becomes active.

But what happens when the fake account has already gone live? That’s where the new machine learning system, Deep Entity Classification (DEC), comes in.

Now, Facebook is releasing details about how DEC works.

Using Deep Entity Classification to Spot Fake Accounts

DEC relies on what it calls deep features to differentiate real Facebook users from the fake ones.

It’s a connection pattern across the network that provides a snapshot of how each profile behaves. Examples of deep features include the average age or gender distribution of the user’s friends.

Facebook uses over 20,000 deep features to classify each account.

It begins with a large number of low precision machine-generated labels. However, the classification soon extends to a small batch of high precision hand-labeled data from users across the world.

Thanks to the final classification process, Facebook would be able to identify one of four types of fake profiles. These include:

• Accounts that are not representative of the user
• Real profiles that attackers have taken over
• Spammers who send revenue-generating messages repeatedly
• Scammers who manipulate users into divulging personal information

According to Facebook, DEC has helped maintain a low volume of fake accounts on its platform since its inception. Currently, it’s about 5 percent of monthly active users.

With the U.S. presidential election approaching, Facebook is endeavoring to fight manipulation on its platform.

Earlier in the year, the social media giant introduced a new policy to ban deepfake videos. The DEC implementation contributes significantly to the ongoing effort.

Read More: Facebook Offers People $5 in Exchange for Voice Recordings