Nudging Healthy Choices Leveraging LLM-Generated Hashtags and Explanations in Personalized Food Recommendations

Making healthy recipe choices can be challenging for users, requiring time and knowledge to differentiate among various options. These choices are often generated by personalized recommender systems that account for individual preferences. One effective approach to encouraging healthier food choices is to intervene in how these choices are presented to users. In this paper, we explore the impact of nutritional food labels and evaluate the effectiveness of a Large Language Model (LLM) in generating high-quality explanations and hashtags to support users in making healthier food decisions. In an online experiment (N = 240), we designed a knowledge-based recommender system to generate personalized recipes for each user. Recipes were annotated with one of four intervention, a Multiple Traffic Light (MTL) nutrition label, LLM-generated explanations, LLM-generated hashtags, or no label (baseline). Our findings indicate that the interventions significantly enhanced users’ ability to select healthier recipes. Additionally, we examined how different system components affected the overall user experience and how these components interacted with one another.