Database Reference
In-Depth Information
Figure 4. Example of a XML document conformed to the DTD tree of the Figure 3
Note that we may have conditional mappings
when the link with an attribute
att
depends on its
value
val
(C
1
↔ //e/[@att = 'val']/@att).
of the XML source in order to identify central
concepts.A depth-first search is performed on the
DTD tree
T
until an element
e
D
of
T
belonging to
a mapping is found. This element will necessarily
be associated to a class
e
O
in
O
representing a
central concept. The search of additional central
concepts will be pursued later starting from the
brother node of
e
D
. Indeed, all the elements be-
longing to the sub-tree rooted in
e
D
and mapped
with entities in
O
should be linked to
e
O
in
O
.
Second the construction process of a view is
guided by the ontology in order to complete the
description of the central concepts. We introduce
the properties of the classes corresponding in
O
to the central concepts (relations and attributes)
if they are properties with mappings, the classes
linked by the introduced relations (called subordi-
nated concepts), their properties with mappings,
and so on. Indeed, the same completion process
is performed recursively on each subordinated
concept. For example,
name
,
located
and
has-
Contact
are three properties of the class
Cultural-
Place
with mappings.
located
and
hasContact
are two relations establishing respectively a link
with the classes
Address
and
Contact
. The view
under construction corresponding to
S
1
will be:
S
1
(x,y,z,t) →
CulturalPlace
(x) ∧
name
(x,y) ∧
Data Extraction and Transformation
Data extraction and transformation are completely
automatic tasks usually performed by wrappers.
It is a two-step process. First, an abstract descrip-
tion of the content of the external source is built.
Second, data is extracted and presented in the
format of the data warehouse.
Abstract Description of a Source
The content of an external source is described in
terms of views in the language accepted by PIC-
SEL (Rousset & Reynaud, 2003) by a set of rules.
Each rule links a view
v
i
(x)
with a local name to
domain relations
p(x)
in the ontology. It indicates
which kind of data can be found in the source. Our
proposal is to build a limited number of views, one
view per central concept in a source. A concept is
said central if it is mapped to an element in
O
and
if none of its predecessors is mapped.
The construction process of a view is incre-
mental. At first, it is guided by the DTD tree
T
Search WWH ::
Custom Search