Ontologies is a key technology for information engineers to shape information by formalizing agreed conceptualizations in specific domains. The aim is to enhance the proper manipulation of the existing information. Still, although ontologies provide formal and unambiguous representations of domain conceptualizations, it would be a surprise if two independent parties had constructed the same ontology even for the same domain. Such a situation in an open and anonymous environment such as the World Wide Web is very common. Interoperability can be admittedly achieved through reaching an agreement, by producing a single and well-agreed ontology, or by aligning ontologies. This thesis focuses on the ontology alignment area of research. Specifically, the presented thesis makes the following three contributions:
Current state of the art methods exploit “surface features” of the ontologies, such as labels of elements, instances of concepts, and the structure of the ontologies, for the computation of equivalence mapping relations. This is achieved through various techniques that these methods employ. In this thesis we take a step further and try to generate “latent features” (based on the features of the source and target ontologies), which are not directly present in the ontologies, but can be utilized for a more precise representation of the ontological elements. Towards this end, the method utilizes Probabilistic Topic Models, for the generation of these representative features.
We propose the CSR method for the location of subsumption mapping relations between elements of different ontologies, by utilizing supervised machine learning techniques. Currently, although the usefulness of subsumption mappings is known to the ontology alignment community the majority of the state of the art methods focus on the location of equivalence mappings, and only few of them aim to ordered mappings, such as subsumption mappings.
It is a common practice of the majority of mapping systems, to be composed of numerous individual mapping methods. Each method locates its own mapping pairs, by utilizing its own logic and then the final result is produced by the synthesis of the results of all individual methods. The problem of synthesizing different ontology alignment methods is considered an open and vital issue in the ontology alignment community. Towards this end, we propose a “model-based synthesis” method, which addresses this problem by maximizing the social welfare within a group of interacting agents (i.e. maximizing the sum of utilities of individual agents).