Semantic Similarity based on Weighted Ontology - Geographical Information Systems: Trends and Technologies - page 289

The chapter is structured as follows. In the next section the state of the
art of semantic similarity methods is given. In the third section, the basic
notions of geographic data and knowledge base are presented. In the fourth
section, we discuss the problem of weight assignment to ontology and
introduce the probability-based approaches. In the fi fth section, the selected
similarity methods are described. In the sixth section, the experimental
analysis is provided. Finally, the seventh section contains conclusions and
future works.
State of the Art
Studies on semantic similarity have been stemmed in the cognitive sciences
and particularly in psychology (Lakoff 1988; Medin et al. 1990; 1993;
Schwering 2008). They mainly refer to similarity between individuals,
classes, complex (pictorial) scenes, and processes. However, semantic
similarity as reasoning support in information retrieval and approximate
query answering has been extensively discussed in information sciences.
These studies have been mainly focused on similarity of concepts and
relationships. Concerning the relationships, different types have been
considered as follows: associative (e.g., cause-effect), equivalence (or
synonymy), hierarchical (e.g., Is-A or hyperonym-hyponym, Part-of or
meronym/holonym, etc.). The hierarchical relationship, particularly,
has attracted a lot of interest in the research community since it has
been considered very suitable for mapping the human cognitive view of
classifi cation (i.e., taxonomy).
In the literature, the traditional approach to evaluate semantic similarity
in taxonomy is the so-called edge/node-counting approach (Lee et al. 1993;
Rada et al. 1989; Wu and Palmer 1994), which corresponds to the semantic
distance approach. In such a method, the taxonomy is treated as an
undirected graph and the semantic distance is equal to the minimal path
length between concepts. The greater the distance between two concepts,
the less similar they are. As mentioned above, the semantic distance between
concepts can be measured by using edge counting or node counting
approaches. In the former, the distance is given by the number of edges
connecting the concepts, whereas in the latter it is the number of nodes
along the shortest path that connects concepts, including the end nodes
representing the concepts. In Fig. 1, which represents an ontology of water
system, the semantic distance between River and Aqueduct is 4 using edge
counting, while it is 5 using node counting. Given an ontology formed by
a set of nodes and a root node, E 1 and E 2 represent two ontology elements
of which we will calculate the similarity.