Databases Reference
In-Depth Information
Fig. 1.4
Coma
CC
reuse scenario after the evolution of the target schema
There are several other attempts to provide repositories of schemas and mappings
for matching or information integration, in general. For example, the OpenII project
is developing an infrastructure for information integration that includes a repository
of schemas and mappings to permit their reuse (
Seligman et al. 2010
). While the
OpenII schema matcher, Harmony, does not yet exploit this reuse potential, there
are several other tools to explore and visualize the schemas and mappings. In par-
ticular, the Schemr search tool determines a ranked list of schemas in the repository
that are similar to a given schema fragment or list or keywords (
Chen et al. 2009
).
For this purpose, Schemr uses an index on schema element names to first find repos-
itory schemas that are linguistically similar to the search input. In a second step, the
candidate schemas are matched with the input schema to obtain refined schema sim-
ilarities used for ranking. The search tool could thus be useful to determine relevant
schemas for reuse.
A new project at the IBM Almaden research center investigates the repository-
based reuse of schema fragments and mappings, particularly for enhancing schema
matching (
Alexe et al. 2009
). The repository stores conceptual schema fragments
called unified famous objects (UFOs) such as address or employee structures that
are in use or relevant for different applications and schemas. By maintaining map-
pings between similar UFOs in the repository, these mappings may be reused when
matching schemas that contain the respective UFOs. Successfully implementing
such an idea is promising but also highly complex and apparently not yet finished.
First, the repository has to be built and populated; a first design is sketched in
Gubanov et al.
(
2009
). For schema matching, the schemas to be matched have to
be analyzed whether they include schema fragments from the repository for which
mappings exist. Finally, the fragment mappings need to be properly assembled (and
