Food poisoning is something almost everyone experiences at some point, but what if we could identify unsafe foods more quickly and reduce instances of illness?
According to the Food and Drug Administration, there are 48 million cases of food-borne illnesses every year. The FDA can take months to identify an issue with a food product, verify the problem and issue a recall.
But BERT can do it faster.
Also known as the Bidirectional Encoder Representation from Transformations, BERT is the nickname of an artificial intelligence computer.
“Broadly, machine learning is training a computer to do specific tasks,” said Elaine Nsoesie, an assistant professor in global health at Boston University and one of the authors of this new research, in an interview with the Idaho Statesman. “We used machine learning to identify Amazon reviews to find products that got consumers sick.”
The research group used algorithms and statistics to teach BERT how to identify unsafe foods. BERT predicted food product recalls using Amazon reviews with a 74% accuracy rate.
“We looked at past recalls to the FDA to identify key phrases that show up in reviews of unsafe foods,” Nsoesie said. “Words like ‘sick’, ‘rotten’ and ‘’foul’. We then used that information to go through Amazon reviews.”
BERT looked at 20,000 Amazon reviews for standard food products such as baby food, cheese, packaged foods and bread.
“The idea of this research is to make this process faster,” Nsoesie said. “It can take a long time after people first report a food-related illness. But if we can identify these unsafe foods faster, then the FDA can start an investigation sooner and recall foods sooner, so less people are affected.”
Finding food poisoning
Food poisoning can be more than just an inconvenient few days off work.
“Globally, people get sick from consuming unsafe foods,” Nsoesie said. “In 2010, the World Health Organization said every year 600 million people get sick from contaminated foods. And death can happen.”
The Center for Disease Control reports that 3,000 people die every year in the U.S. from food-borne diseases such as E. coli. Last year 16 states, including Idaho, experienced an E. coli outbreak from contaminated romaine lettuce.
According to Nsoesie, there are two main reasons why people get sick from food: diseases such as salmonella or E. coli, or mislabeled food containing allergens that are not declared.
Earlier this year the FDA announced the New Era of Smarter Food Safety. In a statement, the FDA said as part of the program it will begin a pilot study to use artificial intelligence at ports to help determine which imported food products should be inspected to see whether they meet U.S. standards.
“This pilot will build upon FDA initiatives already underway, which are also looking at how use of these new technologies may be able to help us continue meeting our public health mission,” said an FDA spokesperson.
Nsoesie said she has not reached out to the FDA about using BERT but has plans to soon.
Rachel Hager is writing for the Idaho Statesman this summer on a fellowship through the American Association for the Advancement of Science. She is a master’s student in ecology at Utah State University and earned a bachelor’s degree in biology at Bryn Mawr College.