Clustering for Similar Recipes in User-Generated Recipe Sites Based on Main Ingredients and Main Seasoning

Today, many people are constantly using user-generated recipe sites when they prepare their meals. Users search for recipes for their meals from the recipe sites. However, when they search for a recipe using an ingredient name, numerous similar pages which are coincidentally similar or which have been plagiarized are found for them. Moreover, a user who searches for a recipe usually does not select a high-ranking recipe from the search results, reacting better than one might with usual web searches. Given many similar recipes included in the search results, it appears to be difficult for them to compare multiple recipes. Actually, when users compare similar recipes, they must better understand the different points of similar recipes. That need for comparison imposes a great burden on users. Therefore, a system classifying the results of user searches according to similar pages in real time would be beneficial for users. In this paper, we propose a clustering method for user-generated recipe sites based on page structure and main ingredient and main seasoning of the food. It provides a means of classifying the user search results according to similar pages. We conducted an experiment to measure the benefits of our proposed method. The experiment results presents the benefits of our proposed method, which classifies similar recipes based on the main ingredients and main seasonings.

I. Introduction

Today, many people are constantly using user-generated and commercially generated recipe sites, such as Food.com¹

Food.com http://www.food.com/

(U.S.), Mis Recetas²

Mis Recetas http://www.misrecetas.com/

(Hispanic), Beitai Chufang³

Beitaichufang http://www.beitaichufang.com/

(China) and Cookpad⁴

Cookpad http://cookpad.com/

(Japan), when they prepare their meals. Users search for recipes for their meals from the recipe sites. When a user searches for a recipe, the user poses queries of two types, typically incorporating food names such as beef stew or lasagna, and ingredient names such as chicken, cabbage, or onions. Maruha-Nichiro Holdings investigated which query is more used when users search recipes posted on recipe sites[1]. Results show that ingredient keywords are used more often than food name keywords. In fact, ingredient names account for 75% of all keywords. Therefore, when users use recipe sites, they input an ingredient name as a query. However, when they search for a recipe using an ingredient name, they are often deluged by similar recipes which use the same ingredients. Consequently, they become confused. Such similar recipes impede a user's recipe searches. For instance, when a user inputs “Chicken and Onion” in Cookpad, a famous Japanese user-generated recipe site, the search results extend to more than 46,000 pages. Numerous similar pages are found. The similar pages which have become coincidentally similar or which are plagiarized. The resultant “information overload” confuses users.