|
Purpose: Ontologies are knowledge bases that contains
annotated terms used in a specific (usually scientific) area. An ontology captures a
shared understanding of a domain of interest. It embodies a set of concepts are organized
in a generalization hierarchy, which allows for reasoning and inheritance. Dr. Perl and
Dr. Geller have several years of experience in representing medical ontologies as well as
manufacturing ontologies as Object-Oriented Databases. The former activity was funded by a
major four-year grant of the National Insitute of Standard and Technology (NIST). As part
of this work, we have developed a number of powerful tools for building, analyzing,
querying, and browsing ontologies. These tools are based on our theory of schema
derivation for semantic network ontologies.
Approach:
We propose to build on ontology that represents
real-world knowledge about customers and their needs. For example, it is obvious to a
human that the mother of a new-born baby is a potential customer of baby formula and
diapers. However, this kind of knowledge needs to be collected and explicity represented
in a computer. This activity meshes well with our data mining and added to the ontology,
after review by a human. In addition, we wil be building a new graphical (diagram-based)
tool for viewing small extracts and abstracts of large and complex
ontologies.
Status:
Representation of Knowledge about
Populations of Customers
We have constructed different models
of how to represent knowledge about a population of customers. The underlying model of
knowledge representation is that of an ontology. The ontology contains a directed a cyclic
graph (DAG) of "IS-A links" as the backbone. Modeling a population with this
approach is an interesting and difficult scientific problem, because typically there is a
natural hierarchical organization in a domain of an ontology. However, for populations
this is not so, because there is no natural order for the different divisions of a
population. Thus, the division into different age groups has no priority over the division
into different genders or different values for the "parent status." We are
currently evaluating two main variants of such an ontology model:
-
A flat ontology, where every combination of features (e. g., age
group 20-30, parent, male) is expressed as one concept and connected to a maximum number
of parents.
-
A deep ontology where a concept is connected to parents that hve
one feature less than the child.
Related
Publications:
|
