A Culinary Computational Creativity System - Computational Creativity Research: Towards Creative Machines - page 323

Fig. 16.3 Classifying constituents in artifacts; To p General model with items; Bottom Left Ingre-
dients; Bottom Right Recipes
The knowledge categorizer serves two purposes. First, it classifies both the
work products and the constituents, possibly using more than one classification, as
illustrated in Fig. 16.3 . Recipes are categorized into dishes (pizza, quiche, pasta…)
and cuisines (American, Italian…), and ingredients into ingredient types (meat, veg-
etables, fruits…) and also by the cuisines in which they are prevalent. The online
repositories used to populate the knowledge database provide a starting point for this
categorization effort. Wikia recipes are tagged with ethnic cuisine categories, and
the dish represented by each recipe is inferred from the recipe name. By parsing the
Wikia and Wikipedia [ 20 ] pages on the various cuisines and ingredients, one can
further associate ingredients and dishes with cuisines. The pairwise compatibility of
ingredients is studied by analyzing how often any given pair of ingredients is found
together in the corpus of recipes. Some classifications, such as the ingredient types,
may be difficult to derive solely from the information contained in existing repos-
itories. One approach consists in clustering the ingredients based on their nutrient
or flavor compound composition. Domain experts can also refine the classification
with a crowdsourcing tool.