Facebook AI Cracks Down on Fake User Accounts

Users of Facebook’s social media service will be able to breathe a sigh of relief as the company’s AI system seems to be doing an excellent job of finding and removing fake and abusive user profiles. The company’s latest Community Standards Enforcement Report noted that the AI managed to remove 3.2 billion fake accounts between April and September of 2019. The numbers are double the amount the system was able to catch and destroy over the same period last year.

A New Method of Detection

Facebook calls the methodology for its AI Deep Entity Classification (DEC). The company went on to state that, since the implementation of DEC, it has seen a reduction in abusive accounts by over 20%. The social media giant also uses several other methods for the detection of fake accounts. However, DEC is a much better alternative for particularly challenging cases.

Multiple Points of Interest Examined

Typical detection methods utilized by the company focused on low-effort locations that many fakers simply didn’t fill out on fake profiles. Eventually, however, the abusers delved into profile creation and managed to game the cursory inspection. DEC dug further into individual profiles and extracted over 20,000 details that would establish an account as belonging to a real user. The aim of implementing DEC was to reduce the amount of abusing accounts existing on the platform.

Extreme levels of Accuracy

DEC currently uses hundreds more points of detection than other methods. The result is that DEC can maintain a 95% accuracy in detection rates. In addition to DEC, Facebook also uses a language recognition AI that scans posts for abusive language. It is trained in over ninety different languages and thirty dialects. Systems implemented by Facebook also include automatic video detection algorithms that pick up on potentially inappropriate content. By using these AI agents, the social media giant hopes to make a more family-friendly environment for its users and remove profiles that may be dedicated to disruption and hate-speech.

CloudWedge