On August 26, Facebook published details of its new system called the TIES (Temporal Interaction EmbeddingS) system. The company has developed this system to improve the platform’s detection of fake profiles and misleading information, adding to Facebook’s existing approaches and refining the company’s overall effort in eliminating bad actors from Facebook’s platforms.
According to Facebook, its new TIES research will provide incremental improvements to Facebook’s detection process. According to the social media giant, billions of users rely on Facebook products as well as services to connect with their loved ones, build communities, and run their businesses. But the rise of fake profiles has introduced many integrity challenges. Entities on Facebook like profiles, posts, groups, and pages are not static and they interact with one another over time. Facebook says that this can reveal a lot about their nature, for instance, inauthentic accounts and posts containing misinformation elicit different types of reactions from users than do normal profiles and posts.
The TIES system of Facebook can highlight common behaviors linked to inauthentic entities by monitoring various interactions and activities with each post on the platform. It is a deep learning, app-agnostic, scalable framework for embedding sequences of entity interactions, Facebook wrote. Entities on social media platforms generate numerous interactions, and Facebook’s TIES system can better match those behaviors to clear signals.
The company trained its TIES model on 2.5 million accounts (with 80/20 real/fake) and 130,000 posts. Nearly 10% of those 130K Facebook posts were labeled as misleading information. The model was able to accurately identify bad actors within the platform’s graph. Although users would not note any real difference on a small scale, definitive patterns emerge on a larger scope. This will ultimately help Facebook to improve its detection as well as enforcement efforts.
During the last few months, the company has been under significant pressure to do more in combating misleading information. The company has taken several actions to tackle misinformation. In April of this year, the company reported that nearly 5% of Facebook’s user base is made up of fake accounts which do not sound so bad. However, considering the 2.7 billion users of Facebook, 5% equates to over 135 million active fake accounts on Facebook’s platform.
The company notes in the TIES documentation that at the scale of Facebook, even a couple of percentage point improvement indicates a significant number of fake profiles being detected.
Read next: Facebook is worried about the new privacy changes in iOS 14 and warns developers about impending doom for the targeted ad business
According to Facebook, its new TIES research will provide incremental improvements to Facebook’s detection process. According to the social media giant, billions of users rely on Facebook products as well as services to connect with their loved ones, build communities, and run their businesses. But the rise of fake profiles has introduced many integrity challenges. Entities on Facebook like profiles, posts, groups, and pages are not static and they interact with one another over time. Facebook says that this can reveal a lot about their nature, for instance, inauthentic accounts and posts containing misinformation elicit different types of reactions from users than do normal profiles and posts.
The TIES system of Facebook can highlight common behaviors linked to inauthentic entities by monitoring various interactions and activities with each post on the platform. It is a deep learning, app-agnostic, scalable framework for embedding sequences of entity interactions, Facebook wrote. Entities on social media platforms generate numerous interactions, and Facebook’s TIES system can better match those behaviors to clear signals.
The company trained its TIES model on 2.5 million accounts (with 80/20 real/fake) and 130,000 posts. Nearly 10% of those 130K Facebook posts were labeled as misleading information. The model was able to accurately identify bad actors within the platform’s graph. Although users would not note any real difference on a small scale, definitive patterns emerge on a larger scope. This will ultimately help Facebook to improve its detection as well as enforcement efforts.
During the last few months, the company has been under significant pressure to do more in combating misleading information. The company has taken several actions to tackle misinformation. In April of this year, the company reported that nearly 5% of Facebook’s user base is made up of fake accounts which do not sound so bad. However, considering the 2.7 billion users of Facebook, 5% equates to over 135 million active fake accounts on Facebook’s platform.
The company notes in the TIES documentation that at the scale of Facebook, even a couple of percentage point improvement indicates a significant number of fake profiles being detected.
Read next: Facebook is worried about the new privacy changes in iOS 14 and warns developers about impending doom for the targeted ad business
