Abstract:
Hierarchical classification ontologies are light weight ontologies, represented in the form
of directed acyclic graph where a node models a concept and its label codifies the meaning of the
concept. Relationships among nodes are usually represented by “narrow-than” or “broader-than”
relations in the graph. Such ontologies classify concepts at each level and proceed from
generalized to specialized concepts. In the same subject domain, different hierarchical ontologies
can have different classification of concepts. Similar concepts in different ontologies may
classify in different ways and are placed at different hierarchical levels in their respective
ontologies.
We need to know the implicit context of concepts in the ontologies to map the concepts. Data
types properties, object types properties, relationship among concepts and their respective axioms are
required to identify the context. But these multi-facet features are mostly unavailable in hierarchical
classification ontologies that emanate a great need of exploring and identifying the context. This
thesis proposed a structural matching methodology to identify the hidden patterns in hierarchical
relationship of concepts in ontologies. Such patterns help in describing the implicit context of the two
subject ontologies. The proposed methodology can be embedded as a component to an existing
mapping system to resolve the mapping complexities in hierarchical classification ontologies.
We have implemented our proposed methodology to validate the rules. The methodology has
been evaluated on two pairs of hierarchical classification ontologies: (i) Dmoz and Yahoo web
directories and (ii) ACM computing classification and Mathematics Subject Classification. The
methodology was compared with existing ontology matching systems in terms of precision, recall
and interpolated precision. The evaluation results show the significant improvement over the existing
ontology matching systems in case of identified patterns for aligning hierarchical ontologies.
