Scientists Can Now Judge Your Diet by Your Twitter Feed

gettyimages-598079051-8dd2e129-96ea-4ee4-91bd-024759c02c03
Jonathan Knowles / Getty Images


If America could be defined by what it likes on social media, you might as well call us the United States of Pizza-Eating and TV-Watching. At least, that’s according to research the University of Vermont recently published in the journal PLOS One.

Co-authors Dr. Peter Dodds and Dr. Chris Danforth didn’t go around surveying Americans about their leisure-time activities or compiling data based on the total dollar amount spent on different types of foods. They turned to Twitter, and with a tool called the “Lexicocalorimeter,” they scoured roughly 50 million tweets sent between 2011 and 2012, seeking out terms associated with foods or activities. It turns out America’s tweets read like a middle-school sleepover. “The dominant words that feed into our scores are pizza and watching TV,” Dodds says. “Those are really the dominant terms across the board.”

Fun? Yes. Healthy? Debatable.

Here’s how the Lexicocalorimeter works: Every time one of the thousands of identified foods or activities popped up in a tweet, it’d be assigned an associated caloric value. For instance, if you tweeted, “Capped off a training run with a steak dinner,” the Lexicocalorimeter would tag run with the average number of calories burned for an hour-long workout, and steak with the number of calories associated with a single serving.

Clearly, it’s not an exact science, but getting exact, individual numbers wasn’t the point. Identifying state-based trends and providing a broader, health-related correlation, was.

“We’re interested in large-scale problems where there’s something really important about humanity that’s hard to measure,” Danforth says. “Quantifying how well we’re functioning as a society is a multi-dimensional process. A lot of the data traditionally is based on economics, diseases, emergency-room visits, and healthcare costs. It takes time and money to compile these stats.” But using the Lexicocalorimeter, Danforth and Dodds were able to show a strong correlation between geographical location, how people tweet, and common diseases or risk-factors, such as diabetes and obesity.

“This is a proof of concept to quantify behavior in shorter time periods,” Danforth says. “People near you use the same words you use. People near you exercise the same way you do. We’ve shown that the behaviors of those who tweet reflect those of the people who don’t tweet. There’s a strong correlation between the calories we consume and the foods we talk about, the exercise we engage in, and the balance between the two.”

While fascinating and entertaining at the individual level — Texas, why is eating one of your biggest activities? Michigan, why do you talk about lying down so much? — the end-game of this new technology is broad in scale. “It’s big picture, public health, public policy stuff,” Dodds says. “The tool is complementary to other data, such as surveys. It’s a real-time way of looking at what people are talking about and how their health might be changing — improving or declining — over time.”

While real-time stats won’t be available until the project is fully funded, for now you can check out the data from 2011 to 2012 to see how your state’s doing, and whether you fit in or stand out from the crowd. “You can say, “Hey look, there’s a culture of activity in my state that I’m engaging in or not engaging in,” Dodds says. “It’s helpful to get a bigger picture, another lens into what your culture’s like.”