Big Data as a Force for Good:
UVM’s White Hat Number Crunchers
UVM Professors Peter Dodds (left) and Chris Danforth. Photo by David Seaver.
From creating new forms of life to measuring the intangible, scientists in the UVM Complex Systems Center live in the realm of uncharted territory. The topics they tackle are, well, complex, so they have developed a suite of equally complex-sounding virtual instruments—the Hedonometer, the Teletherm, and the Lexicocalorimeter, to name a few—to translate massive amounts of data into useful information that will help people around the globe live longer, happier, and healthier lives.
Their groundbreaking discoveries are putting UVM Complex Systems on the map and attracting the interest and philanthropic support of major partners like AMD, Google, and the Massachusetts Mutual Life Insurance Company (MassMutual).
With help from the supercomputers at the Vermont Advanced Computing Center, researchers are quantifying the changes in language used around mental health, gauging the physical health of populations by state, charting the use of proverbs in books and other written materials, reducing race and gender bias in patient medical records, and much more.
From creating new forms of life to measuring the intangible, scientists in the UVM Complex Systems Center live in the realm of uncharted territory. The topics they tackle are, well, complex, so they have developed a suite of equally complex-sounding virtual instruments—the Hedonometer, the Teletherm, and the Lexicocalorimeter, to name a few—to translate massive amounts of data into useful information that will help people around the globe live longer, happier, and healthier lives.
Their groundbreaking discoveries are putting UVM Complex Systems on the map and attracting the interest and philanthropic support of major partners like AMD, Google, and the Massachusetts Mutual Life Insurance Company (MassMutual).
With help from the supercomputers at the Vermont Advanced Computing Center, researchers are quantifying the changes in language used around mental health, gauging the physical health of populations by state, charting the use of proverbs in books and other written materials, reducing race and gender bias in patient medical records, and much more.
What's in a Word?
When’s the last time the country was in a really good mood? UVM professors Chris Danforth and Peter Dodds can tell you the exact date: December 25, 2008 was the happiest day on Twitter since the pair began keeping record.
Danforth and Dodds, both trained mathematicians and computer scientists, are co-directors of UVM’s Computational Story Lab at the Vermont Complex Systems Center. They lead the research group responsible for the Hedonometer, an instrument that accurately quantifies happiness based on Twitter activity.
When Danforth and Dodds first created the Hedonometer, their team surveyed speakers of dozens of different languages, asking them to rate thousands of words on a scale of happiest to saddest. They struck an agreement with Twitter to sample a random 10 percent of daily tweets, and the social media platform has since provided a limitless treasure trove of real-time data, known as “digital traces,” for researchers to study. Each day the software analyzes roughly 50 million Twitter posts, assigning a value to words, phrases, and emojis based on the survey results. Averaged together, these values can measure the nation’s happiness in a single day or reveal patterns in the national consciousness over time.
“It won’t come as a surprise to most people that Christmas is always the happiest day of the year,” says Danforth. “Twitter is flooded with highly rated words like ‘merry,’ ‘happy,’ and ‘family.’ And ‘Christmas’ itself, of course.”
Since they began tracking tweets in 2009, Danforth and Dodds have found that global and national bad news—mass shootings, natural disasters, terrorist attacks—often correlates with a drop in happiness, though typically only for a day. Data from the Hedonometer reveals that May 31, 2020, five days after the killing of George Floyd, was the saddest day the team has ever recorded.
Danforth and Dodds hope that what they learn from social media about the public’s collective well-being may give clues about how to help people live happier and healthier lives.
Better Health is a Walk in the Park
UVM researchers are interested in investigating all of the variables that contribute to better mental and physical well-being. In a landmark study that brought together multiple teams, including the MassMutual Center of Excellence in Complex Systems and Data Science and the Gund Institute for Environment, UVM scientists used Twitter posts and geolocation data to measure the happiness effects of city parks in the 25 largest U.S. cities.
They found that city-dwellers communing with nature experience an elevation in mood similar to that recorded in the general population on Christmas day. Ensconced in the verdant calm of an urban park, people wrote far happier tweets, using words like “beautiful,” “fun,” “enjoying,” and “amazing.” The powerful happiness benefit from city parks lingered for hours after individuals left the location. It was present across all seasons, months, weeks, days, and times of the day—not just weekends and summer holidays—and was highest in parks over 100 acres in area. Danforth says greater immersion confers greater benefits.
“Being in nature offers restorative benefits not available for purchase in a store, or downloadable on a screen. However, not all parks appear to be equal when it comes to happiness. The ability to immerse yourself in larger, greener natural areas had a greater effect than smaller paved city parks.”
Public parks are the primary source of nature for millions of people living in cities, but until now, nature’s effect on humans has been nearly impossible to quantify on this scale. Given the findings in this study, the researchers argue that city planners should work to protect these green spaces and expand access to urban nature.
“Computational social science has so much potential to address pressing issues. Understanding how to help people live longer, live better…what more could we hope for?” Chris Danforth
Signals in the Noise
Instruments born of the Computational Story Lab have shown that they can gauge public mental health in response to trending topics and events.
But what other stories can they tease out of the data we share publicly, and what will these stories portend?
“So much information can be gleaned from the types of words we use,” says Danforth. “We’re interested in finding out what is predictable, both on a population scale and in individuals.”
Story Lab researchers have been able to apply machine learning to successfully detect depression in individuals based on clues from their social media posts. One study found that Instagram users suffering from depression post photos whose colors are cooler and darker than those of non-depressed people. The computer’s rate of detection was 70 percent, 28 percent higher than that of general-practice doctors diagnosing depression in person.
Research has revealed that social media activity can serve as a bellwether for all manner of sociocultural phenomena. Danforth and Dodd’s latest invention, the StoryWrangler, uses a daily sampling of Twitter posts to analyze the popularity of words and ideas across 150 different languages. The new tool acts as a powerful sensor for understanding, in real time and in any location around the globe, what happened, how people reacted, and what might come next.
In addition to giving us insights into pop culture—trends in fashion, the popularity of celebrities, fad diets—the StoryWrangler could provide signals of financial, social, or political turmoil, the early stirrings of unrest that spark crashes, protests, movements, or even revolutions. It also has a lot to tell us about population health and the spread of infectious diseases.
At the beginning of the COVID-19 pandemic, UVM researchers found that worldwide happiness dipped for much longer than the typical 24 hours as quarantines and social distancing became the norm. In March 2020, there was an unprecedented, sustained downturn in global mood. Danforth says every day that month was sadder than the day of the Boston Marathon bombing in 2013.
“We had grown accustomed to seeing downward spikes brought on by bad news days, but what we saw early on in the pandemic looked more like a trench,” says Danforth.
UVM researchers also gained a better understanding of how the world was responding to COVID-19, using Twitter to discern when our collective attention was and was not focused on the virus. Usage of words related to the virus peaked in January and February 2020, and then attention fell to other topics.
“Basically, everyone forgot about it,” says Danforth. “In February, the world had stopped paying attention. It was alarming, because it’s during that time that it’s surging.”
Dodds is hopeful that an understanding of how major disasters affect the well-being of populations will help mitigate and manage future pandemic outbreaks.
“Real-time measurement will enable local authorities and policymakers in health systems and governments to respond to the immediate situation and evaluate possible scenarios,” he says.
What Comes Next
“This is a revolution time in science, and the big change was the internet,” says Dodds. “It’s allowed these data sets to be available to scientists and has allowed us to record and model. It’s also a time to be more responsible, to ensure that the data is used to the benefit of society.”
Building on their last decade of work, the duo is now looking to other sources of data, beyond social media, to study how incentivizing healthy behaviors like exercise, talk therapy, and experiences in nature might influence well-being in individuals. “Computational social science has so much potential to address pressing issues,” says Danforth. “Understanding how to help people live longer, live better…what more could we hope for?”
