Online Program Home
My Program

Abstract Details

Activity Number: 311 - SPEED: Environment and Health, Governmental Policies and Population Surveys, Part 2
Type: Contributed
Date/Time: Tuesday, July 30, 2019 : 9:25 AM to 10:10 AM
Sponsor: Government Statistics Section
Abstract #307713
Title: Using Supervised Machine Learning to Classify Customer Input
Author(s): Adrianna Steers-Smith*
Companies: USDA/FSIS
Keywords: machine learning; deep learning; text classification; customer service; neural network

The Food Safety and Inspection Service (FSIS) is the public health agency in the U.S. Department of Agriculture responsible for ensuring that the nation's commercial supply of meat, poultry, and egg products is safe, wholesome, and correctly labeled and packaged. FSIS receives many food safety related questions from consumers, the food production industry, and Agency’s inspection staff. These food safety related questions have resulted in a large volume of text data that can be used by the Agency to identify trends and help improve the Agency’s repsonses to these inquiries. We used a two-layer neural network, one hidden and one bag of words layer. The bag of words function transforms the input data (both training and testing sets) into a binary array. We used the Sigmoid function to normalize values and achieve an acceptable error rate. The accuracy rate of the model was 82%. Human-level performance for the same data was approximately 85%. This tool can be used to identify emerging trends in food safety related questions. FSIS can then use this information to create guidance that is intended to address questions and ultimately further help prevent food borne illnesses. Early guidance, or FAQ, development ensures customers have access to clarifying information when they need it. This reduces the number of supplemental policy questions FSIS receives and improves customer service by saving time for customers and freeing policy staff to assist in other areas.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program