David Goldberg and his colleges at San Diego State University published an article on improving food safety by mining online customer posts (David M. Goldberg et al., Text Mining Approach for Post-Market Food Safety Monitoring Using Online Media, J. Risk analysis, December 2020). The research team proposes a new Food Safety Monitoring System that uses consumer comments posted on the website to identify products related to food illnesses and an AI technology called data mining. The team compiled an extensive data set of labeled consumer posts in two sites (Amazon.com and IWasPoisoned.com). The database consisted of 11,190 randomly selected Amazon reviews of “grocery and canned food” items purchased between 2000 and 2018, along with 8,596 reviews of food products posted on IWasPoisoned.com. Utilizing text mining and supervised machine learning, they identify unique words and phrases (such as “sick,” “vomiting,” “diarrhea,” “fever,” and “nausea”) related to food safety. Two of the products flagged by the computers had already been previously recalled. Utilizing a data set of 4.4 million online reviews, the data were 77–90% accurate in top‐ranking reviews, while sentiment analysis was just 11–26% accurate. The model was combined with knowledge of higher-risk products to increase accuracy. @ https://doi.org/10.1111/risa.13651
