Arab Researchers Use Artificial Intelligence in Bid to Thwart Fake News

ABU DHABI—Monther Aldwairi sees the pervasive and much-discussed issue of fake news with the kind of problem-solving simplicity that only a computer engineer could.

“Our computers already filter for dangerous viruses,” he says. “Why don’t they do that for fake news too?” It was this question that launched a collaboration between himself and like-minded researchers in Jordan.

Aldwairi, an associate professor in the College of Technological Innovation at Zayed University, in Abu Dhabi, isn’t trying to solve the fake news problem by eradicating it; instead, he wants to help readers to be more discerning about what they read.

His aim is to ensure that the Internet users of tomorrow are alerted if they’re reading a webpage that’s likely to contain falsified information. Crucially, he and his collaborators want to do this in both English and Arabic.

To do this, the researchers are using their field’s trendiest tool—artificial intelligence. (See a related article, “Artificial Intelligence Helps Computers Leap Forward in Reading Arabic.”)

A Problem on Social Media

Fake news—broadly defined by academics as low quality news with deliberately false information usually designed with the intent to encourage social division—is nothing new; it’s been around for centuries. But most observers say it’s exploded in recent years, thanks largely to social media.

That’s a significant problem, says Aldwairi, when you realize that a growing number of people rely on Twitter and Facebook as their main source for news. For example, a 2019 study by Ofcom, the United Kingdom’s communications regulator, showed that at least half of British adults used social media to keep up with the news.

Journalists and experts have previously alleged that some countries, chiefly Russia, are coordinating fake news efforts on social media to disrupt and sway foreign elections in their favor.

“Look at the polarization this creates in the U.K. and E.U.,” says Aldwairi. “We also have our fair share of fake news here in the region, though, mainly around Syria.”

His colleagues agree that digital fake news has taken hold in the Middle East.

“This is a global problem. It’s literally every day that I see people forwarding untrustworthy stories on social media,” says Moath Jarrah, an associate professor of computer engineering at the Jordan University of Science and Technology who is working with Aldwairi to produce a fake news detector. “That’s why it’s so important to develop a tool so that people don’t necessarily believe everything they ever read on the Internet.”

Using Data to Detect Fakery

To do this, the researchers created an algorithm that hunts for certain traits in text, such as an excessive use of exclamation marks, entire words written in capital letters, and certain key words.

“We also have data about how long people stay on a given page. If lots of people are immediately leaving, then it’s suggestive of fake news because they were probably baited to go there,” explains Aldwairi.

The algorithm uses the frequency and severity of these patterns to judge the likelihood that a social media post or even news article has been falsified.

The researchers were able to verify their “fake news checker” algorithm against a database containing over 12,000 samples of Internet posts, which have already been pre-labeled by humans as either fake or real. The results showed that the software was accurate 99.4 percent of the time.

[Enjoying this article? Subscribe to our free newsletter.]

But there’s a problem. In Aldwairi’s line of work, a sample size of 12,000 is small. The researchers want to test their algorithm on more than 100 million data points, and it’s unrealistic that humans will ever be able to label that many posts as real or fake, for them to check their results against.

“It requires a continuous learning process,”says Jarrah, “and in order to improve accuracy, you have to give it as many sources of news as possible to learn from.”

How Machine Learning Can Help

To solve this quandary, they’re using a process known as “machine learning” in which computers, not humans, will label a social media post or news article as real or fake. This will end up with false positives, but that’s where the earlier algorithm comes in.

“Our algorithm should hopefully filter out these false positives,” says Aldwairi.

Additionally, the corrected false positives will feed back into the machine learning to help it understand its mistakes and improve.

The end result of all of this would be a piece of software that gives a rating to the Internet’s vast network of websites on a sliding scale and notifies users of this score when they visit a given page.

“We’re talking about less than a year for it to be a downloadable product,” says Jarrah.

While the machine learning is taking place, the researchers are hunting for suitable datasets in Arabic to teach their algorithm to work in both languages. “We hope to finish that this year too,” says Jarrah.