Information Technology Reference
In-Depth Information
Fig. 16.4 Example of an ingredient bill for a quiche recipe
Categories can overlap or be disjoint. A given recipe usually represents a single,
well-defined dish, but may be shared by the cuisines of multiple neighboring coun-
tries. In this case, it might make sense to establish a hierarchy of categories, in order
to calculate a similarity distance between them.
The second purpose of the knowledge categorizer is to provide a bill of constituents
for the work product categories. An example of an ingredient bill is provided in
Fig. 16.4 . Using an association rule algorithm and performing a statistical analysis
of existing recipes, the categorizer determines what combination of ingredient types
is required to execute a given dish, as well as the frequency of each ingredient type
and the minimum and maximum numbers of ingredients of that type commonly used
in the recipe.
The ingredient bill is not immutable. A single dish can lead to several possible
bills. A pie can represent a savory dish or a dessert, both of which command very
different lists of ingredient types. At a more fine-grained level, a pie can be made with
store-bought dough, or the dough can be made from scratch, which results in more
variations that should be captured by the associative rules. Furthermore, the bills are
refined using online learning methods. As users create recipes with the system, their
inputs constitute new data points that are fed back to the knowledge categorizer. Only
successful ingredient bills are used for feedback.
Creation, in our view, is the process of decomposing work products into their
constituents as depicted in the data model, categorizing the entities, and then recom-
posing and reconstituting new work product ideas. This is different from simple
modification of existing work products that are often modularly designed, because
no modular structure is maintained from existing work products. Since data-driven
idea generation can only use features that are derivable from the underlying data,
the knowledge database is the cornerstone of our computational creativity approach.
Although this limits the creative universe to combinations of previously seen com-
ponents, completely new pairs of components certainly can and do arise.
Search WWH ::




Custom Search