Domain-aware ontology matching
View/ Open
Date
31/07/2021Author
Quesada Real, Francisco José
Metadata
Abstract
During the last years, technological advances have created new ways of
communication, which have motivated governments, companies and institutions
to digitalise the data they have in order to make it accessible and transferable to
other people. Despite the millions of digital resources that are currently available,
their diversity and heterogeneous knowledge representation make complex the
process of exchanging information automatically. Nowadays, the way of tackling
this heterogeneity is by applying ontology matching techniques with the aim of
finding correspondences between the elements represented in different resources.
These approaches work well in some cases, but in scenarios when there are
resources from many different areas of expertise (e.g. emergency response) or
when the knowledge represented is very specialised (e.g. medical domain), their
performance drops because matchers cannot find correspondences or find incorrect
ones.
In our research, we have focused on tackling these problems by allowing
matchers to take advantage of domain-knowledge. Firstly, we present an
innovative perspective for dealing with domain-knowledge by considering three
different dimensions (specificity - degree of specialisation -, linguistic structure -
the role of lexicon and grammar -, and type of knowledge resource - regarding
generation methodologies). Secondly, domain-resources are classified according
to the combination of these three dimensions. Finally, there are proposed several
approaches that exploit each dimension of domain-knowledge for enhancing
matchers’ performance. The proposals have been evaluated by matching two
of the most used classifications of diseases (ICD-10 and DSM-5), and the results
show that matchers considerably improve their performance in terms of f-measure.
The research detailed in this thesis can be used as a starting point to delve into
the area of domain-knowledge matching. For this reason, we have also included
several research lines that can be followed in the future to enhance the proposed
approaches.