A Study of Analytical Model for Diversity of Recipes Based on Embeddings

Abstract

Recently, a large number of cooking recipes have been shared on the Internet. Several machine learning approaches to analyze those recipes have been proposed such as a method to discover alternative ingredients by obtaining distributed representations from cooking procedures and ingredient names, or a method to extract basic procedures from common features in cooking procedures. Such methods utilize the constructed semantic space to calculate the distances between cooking steps and ingredients for each recipe, and show their usefulness by considering the similarity between recipes defined by these distances. In a similar space, we believe that it is possible to analyze not only the similarity between recipes but also their diversity. Even for the same dish, there are a variety of recipes depending on the contributor, and the diversity varies from dish to dish. By taking this diversity into account, it is possible to perform various analyses such as extracting recipes that are suitable for users. In this study, we propose a method to analyze the diversity of recipes using distributed representation. In addition, we apply the proposed method to the data of an actual recipe site and show its usefulness.