Using AI To Spot Fake News Online

That fake news is hugely influential, and hugely damaging, to the smooth running of society is perhaps beyond dispute, but there hasn’t really been a firm strategy for tackling it yet.

Whilst there has been considerable publicity about how automated technologies are helping to spread fake news, researchers from Ben-Gurion University of the Negev and University of Washington believe AI technologies could also help to stop it.  They’ve developed an algorithm to help detect fake accounts on the most common social networks.

The work, which was published recently in Social Network Analysis and Mining, revolves around the belief that fake accounts tend to have improbable links to other users on these social networks.

“With recent disturbing news about failures to safeguard user privacy, and targeted use of social media by Russia to influence elections, rooting out fake users has never been of greater importance,” the authors explain. “We tested our algorithm on simulated and real-world data sets on 10 different social networks and it performed well on both.”

The algorithm consists of two main parts.  The first creates a classifier that is able to estimate the likelihood of a link existing between any two users.  The second part then generates a new set of meta-features based upon this classification.  These meta-features are then used to spot the fake profiles.

“Overall, the results demonstrated that in a real-life friendship scenario we can detect people who have the strongest friendship ties as well as malicious users, even on Twitter,” the authors say. “Our method outperforms other anomaly detection methods and we believe that it has considerable potential for a wide range of applications particularly in the cyber-security arena.

It’s an interesting project, and whilst it’s certainly not the first to attempt to classify fake accounts online, it is an example of how the field is evolving, especially as many fake news accounts are actually manned by humans. You can learn more about the project via the video below.