Deploying nEmesis: Preventing Foodborne Illness by Data Mining Social Media

Abstract

Foodborne illness afflicts 48 million people annually in the U.S. alone. Over 128,000 are hospitalized and 3,000 die from the infection. While preventable with proper food safety practices, the traditional restaurant inspection process has limited impact given the predictability and low frequency of inspections, and the dynamic nature of the kitchen environment. Despite this reality, the inspection process has remained largely unchanged for decades. We apply machine learning to Twitter data and develop a system that automatically detects venues likely to pose a public health hazard. Health professionals subsequently inspect individual flagged venues in a double blind experiment spanning the entire Las Vegas metropolitan area over three months. By contrast, previous research in this domain has been limited to indirect correlative validation using only aggregate statistics. We show that adaptive inspection process is 63% more effective at identifying problematic venues than the current state of the art. The live deployment shows that if every inspection in Las Vegas became adaptive, we can prevent over 9,000 cases of foodborne illness and 557 hospitalizations annually. Additionally, adaptive inspections result in unexpected benefits, including the identification of venues lacking permits, contagious kitchen staff, and fewer customer complaints filed with the Las Vegas health department.

Food Safety

“…Foodborne sickness symptoms can range from minor to severe and even fatal, and they include upset stomach, abdominal cramps, nausea, vomiting, diarrhea, fever, and dehydration. Anyone who consumes contaminated food can become ill from a foodborne illness, although some groups are more prone to getting sick and having a more severe sickness [28].…”

Emerging Applications of Machine Learning in Food Safety

“…Sources of NDS text data that have been applied together with machine learning techniques in food safety applications can be categorized into user-generated public post data and web-based data. Text data from user-generated posts include posts made on social media networking sites such as Twitter (Devinney et al 2018, Harris et al 2017, Harrison et al 2014, Kuehn et al 2014, Sadilek et al 2017 and Facebook, crowdsourced consumer review sites such as Yelp (Effland et al 2018, Nsoesie et al 2014, Schomberg et al 2016 and Amazon (Maharana et al 2019), and participatory systems such as IWasPoisoned.com (Quade & Nsoesie 2017). Post data may also include proprietary content such as company message and feedback boards, user forums and blogs (Kate et al 2014), and query data such as Google search history (Sadilek et al 2018).…”

“…Over the past decade, multiple studies have investigated the use of public social media data for foodborne illness surveillance. Some notable examples involve surveillance systems piloted in conjunction with local and federal health agencies, including the analysis of tweets in Chicago (Harris et al 2014), St. Louis (Harris et al 2017), Las Vegas (Sadilek et al 2017), and New York City (Harrison et al 2014) and Yelp reviews in New York City (Effland et al 2018) and San Francisco (Schomberg et al 2016). In an innovative application of machine learning and the joint analysis of text data with other types of NDS, a team involving Google researchers and the Chicago and Las Vegas health departments combined aggregated Google search queries with smartphone location data to identify restaurants violating health codes (Sadilek et al 2018).…”

“…Aggregated credit card transactions, which have been used to develop machine learning models of consumer shopping trajectories (Krumme et al 2013, Singh et al 2015, could be applied to identify locations (e.g., markets, restaurants) where contaminated products may have been purchased. Location data from cell phone call data records or GPS logged by smartphone applications have found many applications in studying the spread of human-to-human infectious diseases (Oliver et al 2020), but few examples exist so far involving foodborne disease (Sadilek et al 2017, Teyhouee et al 2017. Approaches operating on social media and search queries could be expanded from foodborne illness surveillance to other areas of food safety, including product recalls, allergens, or food safety regulations.…”

“…Environmental data, including weather reports or satellite imagery, a source of NDS not yet mentioned, can be combined with agricultural data to predict food safety events or hazards in agriculture (Pang et al 2020). An exciting emerging area involves integration across data sources, such as the algorithm developed by Google using location and web search text data and trained on restaurant inspection data (Sadilek et al 2017), or between genomic and supply data (Dallman et al 2016). In a long-term vision, it is possible to imagine the combination of many of these data sources, NDS and primary, together in large-scale, end-to-end predictive modeling systems.…”