Leveraging Wikipedia and a Large Language Model to Develop a Knowledge Resource Containing World Foods and Allergens

Presentation Time: 05:30 PM - 07:00 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Public Health
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

This study aimed to develop a knowledge resource for matching global foods to common ingredients and potential allergens. It leveraged Wikipedia's 'Infobox food' template and OpenAI's ChatGPT 3.5 API to create a catalog of ingredients and allergens for over 6,000 foods. Frquency counts against our electronic health record (EHR) system were obtained to understand how often these might appear in health records. The worked revealed wheat as the most prevalent allergen. Issues with data consistency were noted. Use of Wikipedia and large language models (LLMs) showed potential but highlighted limitations and inconsistencies.

Speaker(s):
Simon Shavit, BA candidate
University of Michigan