Letâ€™s say you fire off a tweet about the bomb grilled cheese you just made. Maybe, you wonder into the Twitterverse where you should go for good sushi. Or perhaps you get into an argument with a follower about the merits of the pumpkin spice latte. You barely think before you release your random food musing into the virtual public square, within minutes it gets buried in the timeline and soon itâ€™s forgotten.
Well, not exactly.
Researchers at the University of Michigan recently spent some time resurfacing local dispatches just like those, identifying more than 822,000 food-related tweets sent from the southeast area of the state between 2007 and 2015.
Obviously, no one sat and read millions of tweets, two of the researchers,Â V.G.Vinod VydiswaranÂ andÂ Daniel Romero, patiently explained to me. (They are both assistant professors in U of Mâ€™s School of Information; Vydiswaran also teaches in the Department of Learning Health Sciences.) They designed classification models for machine learning and used a slew of automated processes to collect, filter and analyze the relevant set of tweets; a paper on the methodologyÂ was published earlier this year.
Tweets were given healthiness scores based on what foods they mentioned (with classification help from a U of M nutrition expert), and judged as having a positive or negative sentiment based on 1,200-plus common feelings terms used with food. So, â€œThat salad hit the spotâ€ (incredibly, this hasÂ been tweeted many times) would show a positive sentiment about a healthy food. It would also have a healthy score of +2.
And a tweet so pure and simple and good they named the study after it, â€œBacon bacon bacon,â€ has a score of -6.
Altogether, there were more than 1 million food words mentioned in the tweets, and almost 700,000 were unhealthy foods. People were more likely to share positive than negative sentiments about food, whether it was healthy or unhealthy. Here’s the foods us southeast Michiganders love to eat, or at least love to tweet about:
In their study, Vydiswaran and Romero tried to figure out if this way of using social media to research social science questions around food would work. They concluded yes, their methodology â€œcan provide a reliable signal for dietary patterns and food attitudes.â€
Vydiswaran and Romero clarified that the information left after all the data mining doesn’t exactly tell us what people eat. Instead, it lets them look at how attitudes towards food and the health â€œinformation environmentâ€ vary by location. They found they could get significant results even drilled down to the census tract level, meaning they can use tweets to examine how food and health attitudes differ between neighborhoods with different demographics, income levels, grocery store access, etc.
Though the published paper doesnâ€™t go this deep, Vydiswaran said you could see patterns to the differences between affluent and lower-income neighborhoods: recurring words in the former included sushi, Starbucks and vegan, while frequent foods in the latter included fries, McDonalds and pizza. (I thought pizza transcended class, but who knows.)
â€œThat it comes up so obviously and so clearly was surprising,â€ Vydiswaran said.
Iâ€™m curious to see if social media posts can really tell us things we donâ€™t already know about poverty and health and food access. Fortunately, the researchers are already on it, currently in the review process with a paper that takes this methodology and uses it to analyze food tweets by Metro Detroit neighborhood.
In the meantime, keep tweeting whenever you make a kale smoothie or get a pizza craving or eat something gross or order aÂ full-meal Bloody Mary. Do it for science.Â
Photo credit: Helena Lopes/Pexels