Recipe Recommendations for Toddlers Using Integrated Nutritional and Ingredient Similarity Measures

Abstract

Nutritious and healthy diets in early childhood are critical determinants of children's health, growth and development. An extensive selection of cookery books and recipe sharing websites, containing toddler's recipes, have been provided. The overload of recipe data available makes the recommendation system become indispensable in assisting individuals when making food decisions for their young children. This paper presents a framework that utilizes a combination of nutrient-based and weighted ingredient-based similarity measures to make recipe recommendations. The method is divided into three processes: nutrient analysis, ingredient extraction and integration of similarity measures. The dataset from a reliable source, containing the total of 35 recipes and 87 ingredients, is used in the study. The experimental results show that the proposed technique can effectively generate similarity-based recipe recommendations. The recipe results share both high nutritional and ingredient similarity scores. The comparison results further suggest that the presented approach offers a promising balance between nutrients and ingredients. It is shown that the existing nutrient-based similarity measure tends to overproduce the outputs, while the weighted ingredient-based similarity plays a role to mitigate the shortcomings by removing insignificant recipe pairs with tf-idf weights.