Examining the Effects of Personalized Explanations in a Multi-list Food Recommender System

Abstract

In the past decade, food recipe websites have become a popular approach to find a recipe. Due to the vast amount of options, food recommender systems have been devel- oped and used to suggest appetizing recipes. However, recommending appealing meals does not necessarily imply that they are healthy. Recent studies on recommender sys- tems have demonstrated a growing interest in altering the interface, where the usage of multi-list interfaces with explanations has been explored earlier in an unsuccessful at- tempt to encourage healthier food choices. Building upon other research that highlights the ability of personalized explanations to provide a better understanding of presented recommendations, this thesis explores whether a multi-list interface with personalized explanations, which takes into account user preferences, health, and nutritional aspects, can affect users’ evaluation and perception of a food recommender system, as well as steer them towards healthier choices. A food recommender system was develop, with which single- and multi-lists, as well as non-personalized and personalized explana- tions, were compared in an online experiment (N = 163) in which participants were requested to choose recipes they liked and to answer questionnaires. The analysis re- vealed that personalized explanations in a multi-list interface were not able to increase choice satisfaction, choice difficulty, understanding or support healthier choices. Sur- prisingly, users selected healthier recipes if non-personalized rather than personalized explanations were presented alongside them. In addition, users perceived multi-lists to be more diverse and found single-list to be more satisfying.