Abstract:
Modern information systems extensively use ontologies to model domain knowledge.
Nowadays, with the large amount of already available ontologies, there is a high demand
for sharing and reusing the knowledge in existing ontologies. Since ontologies are
complex structures, sharing of knowledge coming from various ontologies has become a
tedious task. This has resulted in the birth of research area called ontology alignment.
There are numerous techniques for the alignment of ontologies, and the field still faces
many challenges. For instance, these techniques are rather domain dependent and expect
considerable amount of human interaction.
Due to the inherent nature of multiple relationships among the ontologies, it postulates
that the Multi-Agent System technology is a better technology to automate the ontology
alignment with little human intervention. Multi-agent system technology has shown
promising results in modeling domains with interconnected and distributed entities.
This thesis presents multi-agent based approach for ontology alignment. The proposed
solution simulates how different processes interactively operate inside the human mind to
perform certain activities, intelligently. In fact, none of these individual entities are
supposed to be intelligent, nevertheless, through their interactions, intelligence is
emerged. Based on this idea, a novel solution for ontology alignment is proposed. Indeed,
the proposed solution uses agent communication, negotiation, and coordination as the
primary method of exploring the semantic relationships between the ontologies. The
system accepts ontologies maintained in any major form of ontology representation
languages as its inputs and generates ontology with new semantic relationships as its
output. The generated ontology could be used as a shared understanding between
information systems that are running on input ontologies. The system is designed based
on Request-Resource-Message Space-Ontology architecture. The solution is developed as
a plugin for the popular ontological modeling environment known as Protégé. The system
initiates an agent to represent each concept in input ontologies, and these agents execute
on behalf of their respective concept. Further, the system also uses string, linguistic, and
structural similarity matching agents together with upper ontology matching agent to determine the similarity between the concepts. The linguistic matching agent accesses the
WordNet database to fetch synonyms information whereas the upper ontology matching
agent uses the DOLCE upper ontology to fetch domain independent information. In
general, operational knowledge and the rules required for above agents to operate are
maintained in agent system’s ontology. The user could explicitly provide domain
knowledge at the beginning of the alignment process. In fact, this step is optional.
However, the accuracy of the alignment results are heavily depends on the amount of the
domain knowledge agents could access during the alignment process. Because of its
flexible design, user could easily expand the system’s ontology to suit any domain, and
thus, the solution could be used over ontologies of any domain. For example, if there is
an upper ontology that suits more for the current ontological domain, user could link that
ontology with the system. The success of the proposed approach was evaluated by using
ontologies of conference organizing and agricultural domains. It was evident that system
could discover over 70% accurate semantic relationships, and thus, the author claims that
the proposed approach could resolve the complexity in ontology alignment.