Extraction of Characteristic Sets of Ingredients and Cooking Actions on Cuisine Type
Tomonobu Ozaki, Xia Gao, Mako Mizutani · 2017
Because of the recent increase in health consciousness, social networking services on user generated recipes pay a wide attention and grow rapidly. In recipe sites, each title of recipe sometimes contains the cuisine type (Japanese, Chinese, Indian, etc.) and certain words representing the arrangement, and they play an important factor in finding favorite recipes in databases. However, the cuisine type is not always provided explicitly. In addition, it is not always easy to imagine the concrete arrangement even if some words are provided. In this paper, we attempt to characterize the cuisine type and words for arrangement from the aspect of ingredients and cooking actions. For that purpose, we conduct a series of analyses by using a dataset in a Japanese recipe site. First, statistical methods are applied to explore the relationships among cuisine types. Then, an extended TF-IDF is employed to extract important pairs of cooking actions and ingredients. Furthermore, to identify characteristic sets of ingredients for each cuisine type, association rules on ingredients and cuisine types are extracted and evaluated by various evaluation measures. Through the analysis, we discover certain sets of ingredients and cooking actions they relate cuisine type deeply.