Researchers at Carnegie Mellon University have cracked the code on fraudulent web activity with an algorithm that identifies fabricated social media posts.
The FRAUDAR algorithm pulls the curtain away from the games played by those who intentionally hijack the internet for financial gain or undermine online reviews.
Christos Faloutsos, a CMU computer science professor, created the algorithm with a team of doctoral students. Their work recently received top honors from the Association for Computing Machinery’s Conference on Knowledge Discovery and Data Mining in San Francisco.
“Where’s there’s money, there is fraud,” Faloutsos said, explaining the need for algorithms like FRAUDAR. “We’re not identifying anything criminal here, but these sorts of frauds can undermine people’s faith in online reviews and behaviors.”
Individuals who infiltrate the internet to undermine social media by posting counterfeit reviews are fraudsters, says Faloutsos. They jeopardize the reputations of individuals and organizations by engaging in fake interactions by posting flattering or unflattering reviews of products and businesses.
Online follower-buying services for Twitter are among the more prevalent culprits, he adds.
While social media companies attempt to flush out and shut down these accounts, the research illustrates that those efforts are widely unsuccessful. The FRAUDAR algorithm, which is available as an open-source code, hopes to assist social media companies like LinkedIn and Facebook in casting a wider net to reel in these imposters and their accounts, he says.
The algorithm is open source for use by social media website engineers who want to keep their sites clean, he says. It's not intended as a consumer product.
In testing the algorithm in 2016, the team ran through a massive Twitter database extracted from the social media platform’s 2009 archives. FRAUDAR fingered more than 4,000 suspicious-looking accounts and found that most of the tweets were still alive on the site, and the accounts hadn’t been suspended.
The researchers also found that among the suspicious accounts 41 percent of the followers and 26 percent of those being followed had targeted follower-buying services.
Perpetrators typically exhibit distinctive online behavior – such as interacting on the web with unusual frequency. This behavior shows up through “graph mining,” which locates patterns in these frequent interactions, the research team's area of expertise.
“Unfortunately, I cannot give too many details,” said Faloutsos who is concerned that too much information might impact the algorithm's success. “There are smarter and dumber fraudsters. Many tend to follow highly popular sites and celebrities.”
Faloutsos has spent more than a decade researching data mining. In 2006, he developed a tool, NetProbe that assists internet auction sites like eBay in identifying fraud.
“Social media companies have every incentive to clean up this activity,” he said. “The big companies will use this and try to block as much of this fraud as possible. We hope that by making this code available as open source, social media platforms can put it to good use.”
The research was supported, in part, by the National Science Foundation.