USF-led researchers create algorithm to promote trustworthy, diverse content

A team of researchers led by the University of South Florida may have found a solution to the rampant spread of misinformation on social media platforms, while protecting free speech and encouraging varying viewpoints.

The research team consisted of computer scientists, physicists and social scientists from USF, Indiana University and Dartmouth College. Together, they sought to determine how social media platforms can ensure users receive reliable news.

The results of their study were published Feb. 3 in the academic journal Nature Human Behaviour. The research focused on the recommendation algorithms social media platforms employ to prioritize the content users consume. While those algorithms measure engagement through the number of interactions and pageviews, the researchers concentrated on the reliability score of the news sources and the political diversity of their audience.

“These algorithms, unfortunately, do not have access to signals that tell us if the news source is trustworthy or not …,” said Giovanni Luca Ciampaglia, assistant professor of computer science and engineering at USF. “Because algorithms select what is relevant from what has been shared by our friends, if we somehow end up in what is called an echo chamber, an algorithm will simply see something that is popular in a group of people that already agree.”

Giovanni Luca Ciampaglia, assistant professor of computer science and engineering at USF. Photo provided.

Ciampaglia, who led the study, explained that the algorithms and the echo chambers they create lead to the audience becoming increasingly vulnerable to misinformation and disinformation (information intentionally designed to mislead). Meanwhile, the algorithm only recognizes the continued engagement, quality notwithstanding, which further amplifies the misinformation.

So, the research team created a new algorithm.

Their algorithm uses data on web browsing activity and the self-reported partisanship of 6,890 people. The data, meant to reflect the sex, race and political diversity throughout the country, was provided by the online polling organization YouGov.

The researchers also utilized the reliability scores of 3,765 widely shared news sources based on the NewGuard Reliability Index. The index rates sources according to several journalistic criteria, including editorial responsibility, accountability and transparency.

Ciampaglia said the team hoped to promote content that was engaging but engaging to a diverse audience.

“We saw the first thing is that popularity does not predict quality,” he said. “We also saw the sources that were read by more diverse crowds – meaning crowds that included both people from across the aisle, so to speak – tended to be more reliable.”

The researchers tested those findings through other, well-known algorithms through computer science and confirmed that incorporating more diversity led to more reliable recommendations. Perhaps most importantly, their algorithm still produced relevant content that engaged a diverse audience.

Ciampaglia said the algorithm was particularly useful for those most susceptible to misinformation. He added that the findings held for both conservative and liberal news sources, meaning the algorithm did not penalize one side of the political spectrum to benefit another.

“Which is very important for social media platforms because, in a certain sense, these platforms try not to have a bias towards a particular publisher,” he said.

Ciampaglia said the researchers were surprised to find that a post’s popularity had little to no correlation with its quality. He said their surprise was due to the old notion of the “wisdom of the crowd.”

He explained that in the marketplace of ideas, many people believe the most popular and accepted views are typically the most reliable. Regardless of who puts forth the information and consumes it, Ciampaglia said the thought was “the good stuff bubbles up.” He found that is no longer the case in the digital age.

He believes it is now critical that news sources intentionally speak to a diverse audience, regardless of their beliefs.

“A more diverse set of readers will keep you more in check, so to speak,” said Ciampaglia.

Ciampaglia recognizes the challenge of convincing social media platforms to change algorithms that have led to great financial success. He said the assumptions that platforms were essentially neutral and would aggregate information from a wide variety of sources to show what is most relevant and reliable have proven false.

The ongoing practice of simply maximizing engagement leads to consequences for users, and social media corporations struggle to keep misinformation from running rampant. Some companies are now trying to mitigate misinformation while maintaining current levels of engagement, which Ciampaglia said is a tough proposition.

Ciampaglia said he is moderately optimistic that social media platforms will explore innovative ways to combat the spread of misinformation. He noted Facebook’s recent reversal from a state of denial to committing resources to address the problem.

“Artificial Intelligence and machine learning can help,” he said. “But we claim that in a certain sense it could also help to incorporate our findings and that this could help them do a better job.”