Signal 3: City used Yelp data to predict health code violations in San Francisco restaurants

Foodborne illness is one of major concerns of health and state departments, which enforce certain policies on restaurants to prevent it. Current efforts and surveillance coverage, however, has not been yet fully successful in targeting the constantly increasing number of outbreaks. Recent study leveraged social media data from Yelp to identify restaurants in San Francisco which are at high risk of health code violation. The project used keywords in customer reviews, number of reviews, Yelp stars, and price range of a restaurant. Results along with predictive accuracy turned out to be extremely useful — among all restaurants studied, many previously undetected violations were discovered.

In my opinion the project is an exceptional example of leveraging the social data for public and city’s well-being. Using actual customers’ feedback on how restaurants are doing provided a great deal of information that is usually difficult for government and state departments to obtain. Given the very high amount of restaurants in cities and the limited capacities to do inspections, the approach is a break-through, especially on the cost side.

However, few important points should be raised. First of all, data coming from customers is subjective. There are many factors that influence customers’ review, such as interior design, service, and atmosphere of a place. Also given that people eat out often and in several places, it might be difficult to point a true reason for the illness. Another concern should be restaurants themselves, who can skew the data by adding reviews on their own. Owners are aware of the Yelp significance in food choices and can easily improve ratings by asking friends or even hiring a second party to do that. To deal with the above issues, I would suggest using also other, less prone to bias data in the study. In any case, the model is already a great step ahead and gives hope to future preventing methods and policies in public health.

Source: https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0152117