Google Can Tell Which Restaurant Gave You Food Poisoning

Eating out at a restaurant takes a certain amount of trust. You need to trust that the kitchen is clean, the people handling the food have good hygiene, and that the food is both fresh and has been stored correctly. If any of those aren't happening, you could end up with a nasty bout of food poisoning. Right now, we rely on reputation and regular health inspections to keep us safe, but Google and Harvard University have a better idea.

As Harvard reports, Google and Harvard T.H. Chan School of Public Health have been working together to develop a very quick and automatic way of identifying where outbreaks of food poisoning occur. It requires no visits to restaurants and no health inspections, all it needs is data.

What Google developed is a computer model that relies on de-identified search and location data. It looks for search terms commonly associated with food poisoning, such as "stomach cramps," "vomiting," or "diarrhoea," and then checks that against the location history of the user. For one person that doesn't tell Google very much, but if tens or even hundreds of people start searching for similar food poisoning terms and they all frequented the same restaurant in a short space of time, you've got good evidence there's something amiss and worth acting upon.

The system underwent testing in both Chicago and Las Vegas during 2016-17 and the results were positive. The data was used to trigger health inspectors visitin a restaurant if a potential source of foodborne illness was detected. The health departments didn't know the detections were being generated by Google's system during the test.

Overall, the model detection rate achieved 52.3 percent. That may not seem great, but the rate based on routine inspections alone is just 22.7 percent across the two test cities. Google's system also proved to be more intelligent in identifying the correct restaurant for the outbreak. 38 percent of the detections were for locations that weren't the last place a person visited for food, this is because the system takes into account how long illness remains dormant and factors it in to the results. Complaints by ill people usually focus on the last place they consumed food and therefore can easily be wrong.

Because Google and Harvard's system uses machine learning, there is a lot of room for improvement. The more it is used, the better it should become at tracking back food poisoning to its source and push that percentage hit rate even higher.

Recommended by Our Editors

What Chemicals on Your Smartphone Reveal About You

How to Rethink Online Food Delivery

John Hancock Life Insurance Now Rewards Fitness Tracker Use

If you'd like to know more, the full study was published online this week in the npj Digital Medicine journal.