Monitoring Safety of Food Supply by Analyzing Consumer Complaints

• Activity Number: 98
• Type: Topic Contributed
• Date/Time: Monday, August 7, 2006 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statisticians in Defense and National Security
• Author(s): Artur Dubrawski*+ and Maheshkumar Sabhnani and Andrew Moore
• Companies: Carnegie Mellon University and Carnegie Mellon University and Carnegie Mellon University
• Keywords: food safety ; biosurveillance ; Bayesian methods ; causal models ; machine learning ; biosurveillance ; Bayesian methods ; causal models ; machine learning

Abstract

We present a new approach, called TipMon, to monitoring streams of multivariate, event-based data such as customer complaints, public safety hotline tips, or individual patient health events. Its objective is to identify groups of independently collected records that may originate from common, underlying causes. It has been applied successfully to screen consumer food complaints for emerging patterns of adverse events of natural and intentional (criminal or terrorist) origins. We use that example to explain theoretical background of our approach and its implementation details. We also discuss its performance observed so far. We believe early success in food safety domain indicates a wider applicability of TipMon. It possesses a unique ability to remain sensitive to small signals in multivariate, heterogeneous data---even when it is spotty, noisy, and in short supply.