Attractiveness Analysis for Health Claims on Food Packages

Abstract

Health Claims (Health Claims) on food packages are statements used to describe the relationship between the nutritional content and the health benefits of food products. They are popularly used by food manufacturers to attract consumers and promote their products. How to design and develop NLP tools to better support the food industry to predict the attractiveness of health claims has not yet been investigated. To bridge this gap, we propose a novel NLP task: attractiveness analysis. We collected two datasets: 1) a health claim dataset that contains both EU approved Health Claims and publicly available Health claims from food products sold in supermarkets in EU countries; 2) a consumer preference dataset that contains a large set of health claim pairs with preference labels. Using these data, we propose a novel model focusing on the syntactic and pragmatic features of health claims for consumer preference prediction. The experimental results show the proposed model achieves high prediction accuracy. Beyond the prediction model, as case studies, we proposed and validated three important attractiveness factors: specialised terminology, sentiment, and metaphor. The results suggest that the proposed model can be effectively used for attractiveness analysis. This research contributes to developing an AI-powered decision making support tool for food manufacturers in designing attractive health claims for consumers.