Exploring the Feasibility of Identifying Nutrition Misinformation on Social Media

Abstract

Social media platforms are major sources of nutrition information, but they also spread misinformation, challenging nutrition and dietetics professionals. This project investigates using large language models to identify misleading nutrition content on social media, focusing on misinformation on seed oil and no-/low-carbohydrate diets as proof of concept. The study involves developing a pipeline to identify, collect, and process Instagram posts, identifying those conflicting with U.S. Dietary Guidelines, and fine-tuning large language models like BERT, RoBERTa, and BioBERT to detect misinformation. Our results show F1 scores above 80% for two of the classifiers on both topics, indicating feasibility. Future plans include creating tools to identify and warn about misinformation across various topics, with implications for nutritionist professionals, dietitians, health administrators, policymakers, the text-mining community, and social media users.