Discriminating practical recipes based on content characteristics in popular social recipes

Abstract

Recipe websites sometimes contain vast collections of recipes, making it time-consuming for users to identify recipes that might suit them. In this study, we aim to support users in their recipe selection by discriminating "practical recipes" that are easy to understand, written concisely with sufficient description, and offer detailed tips and pointers. We performed a content analysis of popular recipes found on Cookpad, focusing on ten types of dishes, and decided to use seven content characteristics, such as "description of the heat level" and "description of the cooking time," as features to discriminate practical recipes. We have implemented a discriminator based on an SVM classifier that uses these features. The results of a discrimination experiment show that the mean value of the accuracy of the ten types of dishes is 0.813. This represents a significant difference from a baseline discriminator.