Databases Reference
In-Depth Information
Their solution is to create a conceptual model, which “corresponds roughly to
the notion of integrated conceptual schema in the traditional approaches to schema
integration.”
They consider that both the data warehouse and the source schema are views
over the conceptual schema (i.e., local-as-view (LAV) [
Vijayaraman et al. 1996
]).
As mappings they use “adorned queries,” where the adornment is an annotation on
the variables in the query; these are referred to as Reconciliation Correspondences.
In particular, they consider three types of Reconciliation Correspondences: Con-
version, Matching, and Merging Correspondences. Conversion Correspondences
make data level transformations between the same real world objects. For example,
one might use a Conversion Correspondence to translate between income defined
monthly and income defined yearly. Matching Correspondences specify the match-
ing. The Merging Correspondences show how to merge the data based on the
existing Conversion and Matching Correspondences; they consist largely of calls
to the Conversion and Matching Correspondences.
4.2
Pottinger and Bernstein
The authors of
Pottinger and Bernstein
[
2008
] take as input a pair of relational
source schemas and a mapping between them, and then create a relational medi-
ated schema and the mappings from the mediated schema to the source. They also
show how this can be extended to a larger set of schemas. The mappings that they
expect between the sources is a set of conjunctive mappings similar to the ones
in
Madhavan and Halevy
[
2003
] - i.e., a set of select-project-join queries.
For example, assume that there are two travel websites: TravelOn and GoTravel.
Assume that TravelOn has the relations
TravelOnGuide(Name, ID)
and
Tr av e l O n -
Bio(ID, Bio)
for tour guides and their bios, respectively. GoTravel may, in contrast,
have the single relation
GoTravel-Guide(Name, Bio)
to represent both those
concepts. One possible mapping between these sources is the following:
Guide(Name, Bio) :- TravelOn-Guide(Name, ID), TravelOn-Bio(ID, Bio)
Guide(Name, Bio) :- GoTravel-Guide(Name, Bio)
.
This mapping holds the standard Datalog semantics: Guides can be found either
by taking the join of
TravelOn-Guide
and
Tr av e l O n - B i o
on
ID
, or by looking at
GoTravel-Guide
. Hence, it shows that the two concepts are mapped to each other
since instances of the same concept can be found by either conjunctive query. The
question is: what should be in the mediated schema?
Informally,
Pottinger and Bernstein
[
2008
] requires completeness, accessibility
to both all of the input relation (i.e., it preserves information capacity (see Sect.
2.1
)),
makes the concepts that are mentioned in the mappings accessible, does not combine
relations unless they are related by the mappings, and finally is minimal.
