Clemson-Frankfurt research helps distinguish social media fake news
CLEMSON, South Carolina — Having difficulty discerning real from fake news on your social media channels?
You aren’t alone. Surveys suggest 75 percent of American adults struggle with distinguishing real news from hoaxes on social media posts.
Research by Christian Janze, a Ph.D. student from Goethe University, Frankfurt, Germany, and Marten Risius, a Clemson University management associate professor, may be of help in removing the question mark as to whether you’re reading legitimate news or a fabrication.
The article “Automatic Detection of Fake News on Social Media Platforms” was recognized as the best paper among 350 submitted at this summer’s Pacific Asia Conference of Information Systems.
“A lot has been written and said about fake news since the 2016 U.S. presidential campaign,” Risius said. “Our explorative study investigates how to automatically identify fake news using information immediately apparent on social media platforms.”
The study examined cognitive, affective and behavioral cues in more than 2,000 news article posts on Facebook from left, right and mainstream media outlets during the 2016 election campaign, as well as responses from the user community.
The articles were fact-checked to determine fake from real. Researchers then used 230 samples of fake news and 230 of real news and applied variables to predict those that were fake. Using their process, researchers were able to determine fake from real with 80 percent accuracy overall when all 460 articles were examined. They trained the algorithm so it could correctly detect 90 percent of the 230 fake stories.
“We looked at the affective, behavioral, visual and cognitive cues found in each of the articles,” Risius said. “Affective cues, for example, might show emotion, such as love or anger, while behavioral is reflected in how many likes or shares there were on the article. Cognitive cues would be things like whether a question mark was used, quotes or exclamation marks.”
Risius said combinations of the intellectual cues found in these posts go a long way in determining whether an article is real or fake. For example, the word count, or using all caps, exclamation marks or question marks in a post, are strong predictors of a story being fake.
“For instance, if a person is quoted in an article, it’s a pretty good indicator the story is real,” he said. “Other indicators of an article being fake are its loudness, number of shares and the responses to those shares. If it’s shared more often and there’s a strong emotionality on how others respond, the likelihood of a story being fake increases.”
Risius said they used a fairly simple process to determine real from fake. He said it begs the question as to why a social media outlet with a multitude of data capabilities wouldn’t flag stories they know to be fake for their users.
“Though they have many resources to determine what is real and isn’t, they may be more inclined to prefer the community engagement and public attention rather than solve an issue over what is real or fake news on their platforms,” he said.