Information Technology Reference
In-Depth Information
16.2 Assembling a Culinary Knowledge Database
We propose a representational model for culinary computational creativity that cap-
tures all the aspects of domain knowledge necessary to design, assess, and plan
recipes. In this model, shown in Fig. 16.2 , the basic unit is the recipe, which is then
decomposed into all the elements one usually finds in a cookbook: an ingredient
list, and a sequence of steps with their inputs, outputs, and properties. Another set
of entities (dish, cuisine, ingredient type, and ingredient pairing) are used by the
knowledge categorizer to define the ontology of ingredients and recipes.
However, additional elements are needed to enable the data-driven assessment
of new ideas. Through chemical analysis, ingredients are broken down into flavor
compounds. The compounds are in turn characterized by odor descriptors and pleas-
antness evaluations, all of which are used by the assessor to compute quality metrics.
Similarly, ingredient proportion generation makes critical use of the ingredient nutri-
tion facts, such as carbohydrate content, as we will see later.
Peer-produced online repositories make knowledge in several creative domains
accessible to be learned. By extracting data from the Wikia recipe repository [ 17 ],
we can populate the greater portion of the representation model gravitating around
the recipe entity. The human-readable text, though less structured than recipes in
published cookbooks, can be parsed using natural language processing. Statistical
parsing with domain-specific tokens is able to identify the correct task, tools, ingre-
dients, and tips from a recipe instruction with sufficient accuracy. Other databases,
whether public like the USDA National Nutrient Database [ 18 ], or commercial like
the Volatile Compounds in Food database [ 19 ], provide additional ingredient prop-
erties.
Fig. 16.2 Knowledge representation for culinary recipes and ingredients
Search WWH ::




Custom Search