Multi-list Food Recommender Systems for Healthier Choices

Abstract

Recipe websites are a popular destination for home cooks to discover new recipes and find what to cook. However, the most popular way of recommending recipes to users is trough similarity and popularity-based recommendations, which previous research has shown tend to be unhealthy. Building upon knowledge on how diverse sets of options increases satisfaction, this thesis investigates whether a multi-list recommender interface can support healthier food choices compared to traditional single-list interfaces, as well as increase choice satisfaction. As diverse set of options may introduce choice overload to users, explanations were investigated in terms of how they affect user evaluation with regards to choice difficulty, perceived diversity and understandability. A developed recommender system was used in a online study (N = 366), where users could select recipes from recommendations, as well as answering short questionnaires regarding their choices. The analysis showed that a multi-list recommender system was not able to support healthier food choices. However, users who interacted with the multi-list interface found it more satisfactory compared to single-list users. No significant evidence was found that explanations could mitigate choice difficulty. This thesis provides novel work on the utilization of multi-list recommender systems with explanations in the food recommender domain, which can further be expanded with considering other factors such including personalized recommendations in the multi-list interface.