Database Reference
In-Depth Information
partially ordered sets of levels, COM unites all levels into one model. The assignment of such roles as
dimension, cube, fact table and level is done later during each concrete analysis which is defined in
terms of set-based projection and de-projection operations. This extension of partial order on the whole
model (rather than considering it within the scope of each individual dimension) allows us to treat it as
a model of data rather than a model of analysis (OLAP model).
One of the main characteristics of any semantic data model is its ability to represent complex rela-
tionships among elements and then using them for automating complex tasks like reasoning about data.
There has been a large body of research on semantic data models (Hull et al, 1987; Peckham et al, 1988)
but most of them propose a conceptual model which needs to be mapped to some logical model. COM
proposes a new approach to representing and manipulating data semantics where different abstractions
of conventional semantic models such as aggregation, generalization and classification are expressed in
terms of a partially ordered set. From the point of view of aggregation, greater elements are constituents
of this aggregate. From the point of view of generalization, greater elements are more general elements.
One important feature of COM is that references change their role from navigational tool to an elementary
semantic construct. Another unique property of COM is that it uses two orthogonal structures: inclusion
and partial order. A similar approach is used in (Smith et al, 1977) where data belongs to two structures
simultaneously: aggregation and generalization.
The functional data model (FDM) is based upon sets and functions (Sibley et al, 1977; Shipman,
1981). COM is similar to this model because dimensions can be expressed as functions which return a
super-element. However, COM restricts them by only single-valued functions while set-valued func-
tions are represented by inverted dimensions which are expressed via de-projection operator. In addition,
COM imposes a strong constraint on the structure of functions by its order principle which means that
a sequence of functions cannot return a previous element.
CONCLUSION
In this paper we described the concept-oriented model and query language which propose to treat ele-
ments as identity-entity couples structured using two relations: inclusion and partial order. The main
distinguishing features of this novel approach are as follows:
•
Concepts instead of classes.
COQL introduces a novel construct, called concept, which general-
izes classes. If classes have only one constituent then concepts are made up of two constituents:
identity class and entity class. Data modeling is then broken into two orthogonal branches: iden-
tity modeling and entity modeling. This creates a nice ying-yang style of symmetry between two
sides of one model. Informally speaking, it can be compared to manipulating complex numbers
in mathematics which also have two constituents: real and imaginary parts. In practice, this gen-
eralization allows us to model domain-specific identities instead of having only platform-specific
ones.
•
Inclusion instead of inheritance.
Classical inheritance is not very effective in data modeling
because class instances exist in flat space although classes exist in hierarchy. Inclusion relation
introduced in COM permits objects to exist in a hierarchy where they are identified by hierarchi-
cal addresses. Data modeling is then reduced to describing such hierarchical address space where
data elements are supposed to exist. Importantly, inclusion retains all properties of classical in-
