Deploying nEmesis: Preventing Foodborne Illness by Data Mining Social Media

• DOI: 10.1609/aaai.v30i2.19072
• Publication Date: 2022-06-23T00:17:38Z
• Authors: Adam Sadilek; Henry Kautz; Lauren DiPrete; Brian Labus; Eric Portman; Jack Teitel; Vincent Silenzio
• Keywords: Las vegas

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