Global Positioning System Reference
In-Depth Information
Semantic similarity measures are also widely used to perform query
approximation in GISs. In fact, they provide approximate answers to the
user queries which are formulated in terms of geographic data that have
no match or are missing in the database. For instance, if the user asks the
list of waterways (i.e., navigable transportation canals used for carrying
ships and boats) in a certain region of USA, and the concept waterway is
not provided in the database, then canal, which is more abstract concept
than waterway in the hierarchy (see Fig. 1), is proposed. Similarly, suppose
only a database of river_system is provided and the user erroneously asks
the list of ditches. The approximate answer to this query can be obtained
by applying a similarity method, which is addressed in detecting the most
similar concept to ditch in the database that is river.
The proposal of Rodriguez and Egenhofer (2003; 2004) was one of the
fi rst similarity measures introduced for the geospatial domain, in which
the set of spatial entity classes and their relations are described as an
ontology. The authors defi ne a computational method for assessing semantic
similarity among geospatial entities by using the Tversky's feature-based
model (Tversky 1977). Tversky's model is a set-theoretic measure expressing
the similarity between concepts a and b defi ned by two description sets A
and B , respectively, as a function of their common and distinct features as
follows:
f ( A
B )
S Tversky =
f ( A
B ) + α f ( A - B ) + β f ( B - A )
where the similarity measure S Tversky in the above ratio model lies between
0 and 1 and α, β ≥ 0 are parameters to be defi ned appropriately. If α = β =
1, the ratio model reduces to f ( A
B ).
The method proposed by the authors (2003; 2004), called Matching-
Distance Similarity Measure (MDSM), is a linear combination of weighted
shared features and weighted distinct features (see the Similarity Methods
section for more details). Their method determines the similarity by using
a matching process over synonym sets, semantic neighborhoods, and
distinguishing features (such as parts, functions, and attributes) of a source
ontology. In order to determine features' relevance, two approaches, namely
the variability and commonality have been introduced in Rodriguez and
Egenhofer (2004). These proposals have inspired works on the context
dependency of similarity (Keβler et al. 2007; Janowicz 2008) and other
approaches like the proposals in (Pirrò and Seco 2008) and (Formica and
Pourabbas 2009) as well. In particular, the latter proposal is an ontology—
centered and information content-based method that is conceived to capture
both the concept similarity within the hierarchy, and the attribute similarity
(or tuple similarity) of geographic classes. It will be recalled in the Similarity
Methods section as a representative of combined approaches.
B )/ f ( A
Search WWH ::




Custom Search