Database Reference
In-Depth Information
Although we have identified some concepts from each Linked Data set that
might be related to each other, as we can tell from the descriptions of the classes,
we cannot immediately say that Merea Maps' Parish is the same as Merean Mail's
Locality—they may not be spatially identical, and they do not have the same purpose.
It is clear that many of the links will be relationship links on the instance level; so, the
next step in the link design process is to decide what these predicates will be.
8.5.3 S
tep
3: S
pecify
y
our
rdfS of
ntology
The step of specifying the RDFS ontology is primarily a task of choosing relation-
ships to use as link predicates. It is likely to be an iterative process with the Link
Data generation step (see more about link discovery and creation in Section 8.6). For
the first iteration, Merea Maps identifies the classes in the two Linked Data sets that
are candidates for linking together (either at the class level or, more probably, at the
instance level.) It then considers some options for potential link predicates, such as
owl:sameAs
,
owl:EquivalentClass
, or spatial relationships from the OGC
GeoSPARQL standard, where
geo:rcc8-po
(partially overlapping) will be par-
ticularly useful, and it coins some new relationships, such as
isAdministeredBy
,
hasPostCodeArea
, and
hasAddressLocality
. The namespace of these new
relationships should belong to Merea Maps; since Merea Maps is doing the link
generation, it should retain control over the new predicates and their namespace.
However, it is better if the new predicates are awarded their own namespace as they
belong to a different ontology than that used by the stand-alone Linked Data sets.
This helps to maintain the modularity of the Linked Data sets, so Merea Maps'
administrative regions data can be reused without including these additional link
predicates, if necessary.
Step 4 of the stand-alone Linked Data creation process, to mint new URIs, is not
needed unless intermediate resources are required to bridge the gap between the
two datasets. Step 5, to generate the data, is more complicated than the stand-alone
process, as this is now a question of link discovery and creation, and hence the entire
Section 8.6 is devoted to this issue.
8.6
LINK DISCOVERY AND CREATION
8.6.1 M
anual
l
link
c
reation
Manual link creation is really only suitable for small datasets or high-quality data-
sets for which editorial oversight is paramount. It can also be useful for evaluating
the success of automated methods. A Semantic Web search engine such as Sindice,
8
SWSE (Semantic Web Search Engine),
9
Falcons,
10
or SameAs.org
11
can be used to
search for URIs of resources that are candidates for linking. Often, semiautomatic
methods are used; for example, if there is a single or limited number of third-party
datasets to link to, queries can be crafted to identify the most likely candidates within
those datasets (for example, where the
rdfs:labels
are the same). Note that the
link should go to the resource itself, not the document describing the resource (recall
that resources typically have URIs containing the word “id”).
