Wyoming loves watercress, Texas loves brisket and, perhaps most surprising, Oklahoma is obsessed with the pickle.
Those were the results of a recent analysis of food-related language on Twitter, conducted by a cross-disciplinary team at the University of Arizona. Combing through 3.5 million food-hashtagged tweets pulled from Twitter's API between October 2013 and May 2014, the team identified distinctive food words for each state. Along the way, they also uncovered regional differences in the popularity of #breakfast, #brunch, #lunch, and #dinner. The midwest, apparently, is #breakfast country.
As it turns out, food Twitter is a remarkably rich data source. "People are less sensitive about food," says Mihai Surdenau, an information scientist with the group. "They're more likely to tweet that they ate a big burger than they did not exercise." Of course, even the most dedicated Twitter users only tweet a sliver of their food intake, so Twitter isn't a perfect window into the nation's diet, but the team was able to track some aspects quite well. By the time the work was complete, Fried's model was able to guess at a state's average BMI and diabetes rate based only on its tweets.
The final goal is to drill down to the personal level, identifying specific people at health risk based on their tweets. Tweet too much about bacon, and you might find the American Heart Association rolling into your mention with a warning. "The main idea is to introduce social media into public health," says Surdenau. "We're still thinking about what the best intervention is."