Together with
How Facebook Detects Fake Accounts Using AI?
Nov 22, 2022
How Facebook Detects Fake Accounts Using AI?
Abhilasha sandilya
Abhilasha sandilya
I think you can build things, and the world gets better, but with AI, especially, I am really optimistic?. In the next five to ten years, AI is going to deliver so many improvements in the quality of our lives.? -Mark Zuckerberg

Facebook is a platform used for connecting people and developing healthy communities. People here make friends and share things they love. But what if these communities have some fake people disguised as friends- it will make the community more vulnerable.

Fake accounts stand as one of the most challenging threats faced by Facebook. The social giant, in the year 2017-18, introduced a new tool of Artificial Intelligence (AI) to classify and delete millions of fake accounts. The Community Standards Enforcement Report (CSER) and the Securities and Exchange Commission (SEC) filings often question the identification procedure of fake accounts adopted by Facebook.

But before exploring the procedure, it is essential to understand why fake accounts are considered a threat.

Why Are Fake Accounts Considered a Threat to Users?

Every day hundreds of thousands of fake accounts are tried to be created or are created on the Facebook platform by humans and bots. These accounts are credible technical spoofs created by using fake email addresses, synthesized pictures, and highly accurate data that make the accounts appear more authentic.

These accounts pose a significant threat to the platform and its users in most cases. Innocent users often fall prey to such scammers, and these accounts become an easy source of money. These accounts are used by fraudsters to run spam, incite violence, promote terrorism, influence elections, share phishing links, etc. Russian agencies' influence on US midterm election results was one such famous example of the threats of fake accounts. Using technologies like Deepfakes, experts speculated that Facebook's bogus accounts could also play a decisive role in upcoming US elections. And Facebook is using all means to prevent any interference.

How Fake Accounts Impact Facebook?

Facebook generates a large part of its revenue from advertisements. The revenue growth may stay stable as long as the ads are watched, clicked, shared, or liked. Facebook doesn't provide third-party access to calculate the engagement number for the advertisements. With fake accounts looming large, investors doubt whether the ad is reaching the correct audience. Fake accounts may lay inactive for extended periods, producing insufficient data to run generic consumer advertisements on the page. And then again, the fake accounts of bots are of no use to neither Facebook nor marketers. All these factors have affected revenue generation in several ways. Marketers will continue to associate with Facebook and its related products for advertising purposes as long as they get a good investment return. Having false accounts can cause a big concern to go forward for both Facebook and marketers. Facebook, in the year 2018, reported no new users sign-up from January till July from countries like the US and Canada, and the number of unique users joining from countries like Europe fell by one million. Although it didn't cause any loss in revenue, its market value of shares had fallen by $120 billion. Such is the effect users can bring to social media platforms and their businesses.

How Facebook Detects Such Accounts?

Abiding by the set of guidelines by CSER, Facebook identifies fake accounts using both human intelligence and machine learning. It uses hand-coded rules and machine-learning tools to determine whether accounts are genuine or not.

Fake accounts, once recognized, are blocked either at the time of creation or before they are live on the platform. But the real problem arises when these fake accounts pass under the nose of all these defenses and become active. Such accounts are identified by advanced AI tools when malicious activities done by these accounts come under the notice of Facebook's detection system. These accounts are also sometimes reported by active users.

What Are The Tools used For The Procedure?

One can easily deduce the severity of the problem by knowing that alone in the Q1, Q2, & Q3 2019 transparency report by Facebook, it has removed 5.4 billion fake profiles. Fake accounts are of two categories:

  1. Some are accounts of children, pets, businesses, hobbies, etc., which users created instead of creating pages. Facebook categorizes these accounts as misclassified accounts and converts them into pages.
  2. Other accounts are those created by scammers. These accounts are identified and taken down by Facebook.

But disabling accounts means denying the users access to the platform, which is entirely against the norms of Facebook. So, how can weeding be done without harming others? Keeping all these in mind, the social giant came up with several layers of analysis to root out problem accounts. And the tool it uses for this purpose is Deep Entity Classification (DEC).

Facebook deploys the screening of accounts at three different stages:

Stage of Creation: Fake accounts are recognized while they are being created by taking the help of exclusive detection technology. AI immediately blocks these accounts from being created. It processes high amounts of data and makes quick decisions. For example, the technology detects and reports about numerous accounts created from the same IP address, which is then blocked, obstructing their further visit to the social networking site. AI immediately removes these accounts, and hence no harm is caused to active users.

Stage of Activation: Once such accounts sign-up, the artificial learning algorithms lookup for any suspicious actions that can set on the alarm. These algorithms use the knowledge gained from the removal of previous fake accounts to detect such accounts. For example, DeepFace technology recognizes Deepfakes, and the system is alerted. AI traces signals associated with suspicious email addresses or already disabled accounts and blocks the user's account even before appearing on the platform.

Stage of Blocking Registered Accounts: Deep Entity Classification plays a massive role in identifying challenging accounts that are active on Facebook. It beats those who have already learned to play with identification features like numbers of friends, location, age, etc. DEC's AI recognizes such accounts after screening them thoroughly and, if faulty, blocks them.

How DEC AI Works?

Since its implementation, Deep Entity Classification has helped pull down numerous convincing fake accounts. Instead of surface study, the AI scans in-depth features of the accounts, like the details of pages, groups, friends, etc., with which the account interacts. The algorithms screen Their friend's ages, the gender distribution of friends, age groups they interact with, admins of the pages, members of groups, all. The technology takes into consideration as many as 20,000 deep features. The user cannot control these features, and hence a clear snapshot of how a profile behaves appears based on which decisions are taken by AI.

It uses numerous machine learning models established and developed on complex algorithms to identify the account's authenticity. These models are trained on a vast set of simulated and real-time data for higher accuracy. High-precision human tags then scrutinize the segregated accounts that security experts generate. Language-agnostic AI trained in 93 different languages and 30 dialect patterns is also used to identify fake accounts. Accounts such as those that repeatedly send revenue-generating messages or are hacked accounts of real users, an incorrect account that doesn't belong to the character ascribed, or accounts of manipulative scammers are then deemed fake by the algorithms. AI further investigates these and, at last, blocks them with an accuracy rate of 95%.

Facebook has claimed on its Transparency page that DEC has prevented millions of fake accounts from being made every single day in 2019.

Is The Problem Solved?

No, with advancing AI, the intellectual capacity of spammers is mutating as well. According to an estimate by Facebook, irrespective of its efforts, around 5% (approximately 125 million) of the monthly active users are still fake accounts. So it won't be wrong to infer that it's an ever-ending race where technology on both sides competes to be ahead of each other. And to lead, Facebook must keep coming up with upgraded defence systems using AI and ML coupled with human intelligence.

Abhilasha sandilya
Abhilasha sandilya
She is a Postgraduate in English. She has work experience as an English Language Trainer, and Teacher (UK & India), and a Content Writer. She loves doing research and simplifies complexity.
4 articles
Apr 20 joined
More From This Author
Writing Etiquette
May 19, 2020
Related Stories

Write for us

Our writers are independent, remote and growing in numbers. Join our team of enthusiastic authors and begin creating and earning today.

Get Started