科学美国人60秒:Social Media Helps ID Belch Source

If you’ve ever checked out the restaurant reviews on Yelp, you know that these little missives can tell you whether a bistro is overpriced or understaffed or just nothing to write home about. But they may also be able to tell the local health department whether an establishment has been serving up Salmonella.

Researchers at Columbia University in New York City were looking into an outbreak of food poisoning at a local restaurant a few years ago when they got the idea of using social media to track gastrointestinal disturbances.

“During the investigation, the Department of Health noted that patrons had reported their illnesses on yelp in the reviews but hadn’t reported them via 311, the city’s official reporting service.”

Computer scientist Thomas Effland, who led the study.

Previous investigations had shown that monitoring social media for keywords associated with illness was a good way of rapidly identifying outbreaks of infectious diseases…such as the flu. So Effland and his team built a similar system for stomach symptoms. The Department of Health started using it in 2012.

“The tool works by sifting through the recent Yelp reviews for New York City restaurants each day to identify potential reports of foodborne illness.”

Yelp reviews get scanned for telltale terms such as “vomit,” “diarrhea,” “food poisoning,” and “sick.” Flagged entries then then get passed along to epidemiologists for a closer look.

The system has produced some false positives, for example, from reviews that stated things like, the food “had a weird chunky consistency…hopefully we won’t get sick tonight.” And it missed a few posts, like when the writer misspelled “diarrhea” (a challenging word to write down even when you don’t have it).

But overall, the results are nothing to sneeze at. Or in this case, barf at.

“We found that using Yelp data has helped the Health Department identify approximately 1,500 complaints of foodborne illness in New York City each year. In total the system has found 8,523 complaints since July 2012, resulting in the identification of 10 outbreaks.”

The study is in the Journal of the American Medical Informatics Association. [Thomas Effland et al., Discovering foodborne illness in online restaurant reviews]

The researchers plan to extend their analysis to Twitter. After all, many people use Twitter to let everyone know that they’re not well.

—Karen Hopkin