Calculating Media Influence

The mosquitoes are fighting—and biting—back. After years of being flooded with insecticides, the tiny, nibbling flies are becoming resistant to our chemical defenses. So scientists are testing an alternative to the chemical sprays: releasing genetically modified (GM) mosquitoes that only produce nonbiting offspring.

It’s a controversial approach. When plans to release GM mosquitoes in Florida first made headlines, they ran into stiff public opposition—even forcing local referendum votes—as fears swirled about unintended consequences harming people and local ecosystems. But, just as with other stories rich with scientific detail and social debates—think COVID-19 vaccines—where people get their news, and what news they trust, matters.

“People might say, ‘The news media don’t have any facts anymore, people don’t read news, they check out social media,’” says Guo, Lei, an associate professor of emerging media studies, who lists her family name first as is customary in her native China. Those assumptions, she’s found, don’t always hold true.

An expert on the interplay between the media—especially social media—and public opinion, Guo uses big data research methods to study trust in journalism and how social media giants like Facebook and Twitter impact democracy. By leveraging computing power, she can sort through thousands—sometimes millions—of articles and posts to better understand mass media’s influence. To figure out where the public was turning to get the buzz on GM mosquitoes, Guo compared online news coverage of the story with the debate on Twitter. With Weirui Wang, an associate professor at Florida International University, she looked at 464 news articles and thousands of tweets, finding Twitter had an “inadequate discussion of risk” when it came to the modified mosquitoes.

While what they called elite (like the New York Times) and emerging (HuffPost) news outlets plunged into reporting on topics as diverse as experimental dangers, cost-effectiveness and ethics, the Twitter debate remained mostly stuck on the potential health impacts. “To obtain a well-rounded perspective about an issue,” wrote Guo and Wang, who published their findings in the Journal of Risk Research in August 2020, “people may still want to access online news sites as their main source of information instead of relying only on Twitter.”

And what’s true for people scrambling to understand GM mosquitoes may also say something about how their views are shaped on other topics, like the coronavirus, gun violence or presidential elections—topics Guo has also studied.

A trained journalist and a founding member of BU’s Faculty of Computing & Data Sciences, Guo notes a common thread between her work as a reporter and as a researcher: “We’re curious,” she says. “The job of a journalist is to gather data, interview people to learn their thoughts. As a researcher, I’m doing the same thing, but the method I use to get an answer is more systematic.”

As well as studying media impact, she’s developing tools to help journalists find stories in the world’s data—and navigate a polarized world.

A Billion Tweets

A decade or two ago, studying media impact was relatively straightforward—if a little laborious. Researchers would watch a handful of core news channels or shows, sift through newspaper articles and poll audiences and readers for their views on different subjects. Then came aggregators and blogs and podcasts and Twitter. Not to mention the proliferation of fake news. Now, researchers can access billions of tweets—500 million new missives are posted on Twitter every day, according to Internet Live Stats—not just for news, but to gauge ever-changing shifts and patterns in public opinion.

“When I started my journey in this field, we still used a lot of traditional methods, for instance manual content analysis,” says Guo, who began studying media effects—an academic term for mass media’s influence—as a graduate student research assistant at the University of Texas at Austin. “But there’s no way for me to go through a billion tweets, so I started to explore computational methods.”

Working with computer scientists, she’s helped design and perfect software that can analyze streams of data. For instance, in a recent study examining the link between how media outlets frame stories on gun violence and the issue’s prominence among midterm election voters, Guo used a computer program to evaluate 42,917 news articles. She was looking to see if there was any difference in influence between articles that just reported on a single incident, one shooting or the impact on one victim— what researchers call an episodic frame—and those stories weighing bigger societal issues and questions, called a thematic frame. Instead of manually sorting all the articles by their angle, as well as by outlet type—conservative, liberal, nonpartisan, mainstream—Guo and her fellow BU researchers taught a computer program to do the work for them.

“We start with a sample of articles and manually go through them and classify them,” says Guo. “Then we use methods in computer science, like machine learning, to train a machine model to classify all of the unlabeled data.”

The program, nicknamed BERT—for Bidirectional Encoder Representations from Transformers—provided them with a much more comprehensive dataset than humans alone could amass. The researchers then combined that information with the results of a general population survey that asked people about their media use habits and views on gun violence.

“We correlate the two types of data to see whether exposure to certain media or certain media’s framing will influence how they think,” says Guo.

They’re techniques she’s also applied to the study of COVID and fake news, tracking differences between pandemic coverage in various countries and the power of misinformation to sway public debate. In one COVID project, Guo and her team analyzed the topics making headlines in countries as diverse as the United States, Egypt, China and Germany. They then compiled all of the information into an interactive world map (covid19.philemerge.com), allowing users to drill down and see how the news focus shifted in each country over the first three months of the pandemic.

The Power of Journalism

By combing through such rich datasets, Guo can often uncover nuances that might have been missed by comparable studies in the past. It also gives her much more robust results. The gun violence study, which was published in 2021 in Mass Communication and Society, found that mainstream media—the New York Times, the Washington Post—can still shift opinion, even on heavily polarized issues. But there was an asterisk: although those legacy outlets could move the views of conservatives by using an episodic frame, thematic stories tended to be a dud in changing opinions.

“People already have their positions, they’re tired of the arguments,” says Guo. “It’s really hard to change their opinions, but if you show them incidents or individual stories, maybe that will. Those kinds of articles have more of an emotional appeal.” On the flip side, seeing an article in the partisan media—even an emotion-packed one—could cement the views of that same conservative, leading them to believe gun violence was less of an important issue. Overall, says Guo, it shows that despite the rise of social media, “journalists do have the power to change people’s minds.”

Human reasoning is still very, very important—we cannot rely on machines for everything. Data science just provides us with a new tool to understand the world.

Guo, Lei

In a new project, she’s applying that same nuanced approach to the life cycle of local news stories after publication, analyzing how articles grounded in fact can become twisted into fiction-filled fake news. In October 2021, Guo and a team of researchers at BU, Temple University and the University of Illinois at Chicago were awarded a $750,000 National Science Foundation grant to use data science techniques— including natural language processing and network analysis—to watch articles as they’re tweeted and retweeted, aggregated and rewritten.

News outlets might have a good handle on the immediate impact and readership of an article, says Guo, but not how the story continues to evolve. And, she says, what happens to a story after it hits the web can determine its impact, and the public’s trust in the media overall: “Journalists don’t have control over their news after they publish it. After several layers of transmission, a good story can become something fake or misleading.”

User-Friendly Data Science

Guo hopes her work can give journalists new insights, but is also working on software and other tools they can use to increase the impact of their reporting. The projects, like her teaching—she leads courses on computational and communication research, communication theories and the design of interactive digital products—aim to democratize access to computational methods. One, just in its early stages, will detect bias in articles—that’s being developed with the technology incubator and experiential learning lab, BU Spark! Another aims to give all reporters access to the computerpowered big data analysis Guo uses in her research.

“We’re developing a user-friendly tool, so they can upload their data—it could be media content, news articles, political speeches—and in a few clicks they will have their results,” says Guo. “If they have tons of data and want to see the topics that emerge, they don’t need to have any computer science skills or knowledge—they can just follow our instructions.”

The open-source program will be available for free online. In a world in which processing and understanding big data have become central to so many jobs, from journalist and social media manager to marketer and user experience designer, the program has the potential to benefit anyone looking for trends or studying sentiment analysis.

But Guo also recognizes computing’s limits. Sometimes, she admits, you just have to work through a problem the old-fashioned way—though perhaps her tools can spare you some shoe leather.

“Human reasoning is still very, very important—we cannot rely on machines for everything,” she says. “Data science just provides us with a new tool to understand the world.”