Before the US elections: how can we identify "fake news"?

While the battle between the contenders for the US presidency is tighter than ever, researchers from Ben-Gurion University of the Negev have developed a method that will help verify facts in the amount of information that passes through social networks

"Fake news" is a well-known problem, which grows closer to the existence of election systems, due to the influence on the electorate through misinformation and conspiracy theories. While the battle between the contenders for the US presidency is closer than ever, researchers from Ben-Gurion University of the Negev have developed a method that will help verify facts in the amount of information that passes through social networks. The principles of the method were published at the leading international conference Kdd.

The dynamic nature and amount of information passing through the Internet makes the act of identifying false information in real time an extremely difficult task. The research group of Dr. Nir Greenberg and Prof. Rami Pozis From the Department of Software and Information Systems Engineering at Ben-Gurion University of the Negev, they developed a method using machine learning to detect fake information sources.

The method is based on active machine learning that is based on the characteristics of the content that is distributed, the users that distribute it and the network structure of the consumers of the false content. The researchers examined the success of the method both retroactively and over time, and in effect simulated the way in which human fact checkers are required to identify false information.

For example, it was found that monitoring fake news sites, rather than individual articles or posts, can significantly reduce the burden on fact checkers and produce reliable results over time. The model also knew how to differentiate between temporary textual terms compared to terms known over time. It was also found that labeling popular sources may refer to the most common lies in the general population, but it is problematic for identifying new "fake" sources that often arise in certain communities where there is a demand for this type of content.

"The problem today with the spread of fake news is that fact checkers are overwhelmed. They can't check everything, but the extent of their coverage in a sea of ​​social media content is unclear. Furthermore, we don't know if fact checkers manage to get to the most important content to check, which has led us to develop a machine learning approach that can help fact checkers direct their attention better and increase their productivity," explains Dr. Greenberg.

Sources of "fake news" tend to appear and disappear quickly over the years, so maintaining lists of these sites involves a lot of cost and work. Their system considers the flow of information on social networks and the audience's appetite for this kind of information. The models developed so far are based on the common approach of identifying those who share misinformation. Now, the researchers have shown that their method can maintain the same level of accuracy (80% in PR-AUC measure) in identifying fake sources, while requiring less than a quarter of the cost of testing all sources.

"The system will never replace the human fact-checkers, but the method can flood new sources that seemingly require the attention of the fact-checkers today," noted Prof. Pozis. "Now, social networks have another tool that will make it possible to fight misinformation."

The research group included: Maor Reuven from the Department of Software Engineering and Information Systems at Ben-Gurion University of the Negev and the independent researcher, Dr. Lisa Friedland.