Ingredient-based Group Recommender for Recipes (IGR2)

Abstract

The number of food recipe options in modern society is vast and growing. While often being considered positive, the abundant options also lead to the so-called paradox of choice, i.e. that more options can lead to less happiness. The paradox of choice creates the need for recommender systems, while data and technological development act as enablers. Recommender systems have been adopted extensively in some domains. However, prior research on group recommender systems for recipes is sparse. The need for group recommender systems in the recipe domain is greater than in many other domains as people often consume food in groups. This study aims to use previous research on recommender systems and apply it to the recipe domain, targeted to groups. A publicly available dataset with user-generated content has been used to train and evaluate two models. The first one, Ingredient-based Group Recommender for Recipes (IGR2), is a content-based model using ingredients as features. The second one (IGR2+) is an extension of the first introducing two modifications: (1) the group profile is separated into a positive and negative part, and (2) bias variables are added and take into consideration the average popularity of a certain recipe and the specific user’s general tendency to rate recipes low or high. As part of the preprocessing of the data, Singular Value Decomposition (SVD) is used to reduce the dimensionality of item descriptions and group profiles. The models are valuable as inspiration for industrial applications. Furthermore, this study contributes to future academic research with findings associated with group recommender systems in general and for recipes and user-generated content in particular. Moreover, potential solutions to some of the challenges with group recommender systems are suggested both as part of the IGR2 models and the following discussion.