A machine learning algorithm for personalized healthy and sustainable grocery product recommendations

Abstract

Nowadays, retailers try to optimize the shopping experience for consumers by offering personalized services. Recommending food options, i.e. providing consumers suggestions on what products to buy, is one of such services. Food recommender systems for grocery shopping are typically preference-based, using consumers' shopping history to determine what products they would like. These systems can predict well what a consumer would potentially like to buy, however, they do not stimulate consumers to buy healthier or more sustainable food options. In response to increasing global concerns about public health and sustainability, this paper aims to integrate healthiness and sustainability levels of food options in recommender systems to encourage consumers to buy better food options. To assess the impact of integrating healthiness and sustainability information of food choices in predicting an item to buy, we employ three food recommendation models: a Baseline popularity-based model, Restricted Boltzmann Machine (RBM), and Variational Bayesian Context-Aware Representation (VBCAR) based on (1) preferences, (2) preferences and health, (3) preferences and sustainability, and (4) all combined attributes. Models were trained and tested using two different datasets: Instacart and a Dutch supermarket dataset. The experimental results indicate improved performance for VBCAR compared to Baseline and RBM. Models that emphasize healthiness and/or sustainability of food choices do not significantly alter model performance compared to preference-based models. The results of the health and sustainability-based recommender systems demonstrate the potential of recommender systems to assist people in finding healthier and more sustainable products that are also suited to their preferences.