Davydovskyi M. A Model and a Method for Ontology Instance Migration in Decentralized Systems

The thesis is devoted to the improvement of mathematical models and computational methods for harmonization and alignment of knowledge models, represented in the form of ontologies, in decentralized systems. The dissertation solves the actual scientific and practical problem of developing new mathematical models and computational methods, as well as improving existing methods and algorithms in order to increase their efficiency, contribute to the development of effective and efficient software tools for ontology instance migration in decentralized and distributed environments. The author examines the state of the art in ontology alignment and analyzes the ontology instance migration problem as a specific task in the context of ontology alignment. The problem of ontology instance migration in decentralized systems is formalized and an iterative incremental approach to its solution is proposed in the thesis. The developed approach consists of 4 steps: (i) automatic computation of ontology mapping, matching and computation of structural difference between schemas of ontologies; (ii) automatic generation of transformation rules for individual assertions; (iii) automatic transformation of assertions; and (iv) manual control of the completeness of those transformations and problem cases resolution. Theoretical contributions include: the development of a new method for determining graph matching for ontology schemes based on the concept of isometric cycle and spectrum of graph edge cuts, a proof of the graph isomorphism theorem; the analysis of expressibility and decidability issues which may arise during A-Box transformations; proof of the convergence of the schema matching process following the strategy of automated meaning negotiation; proof of the rationality and efficiency of ontology schema and instance transformations by means of the ontology evolution framework based on the belief revision; elaboration of a fragmentary model of the ontology matching task and special evolutionary-fragmentary algorithms aimed at overcoming complexity issues and increasing the overall efficiency of ontology alignment software tools. The solution is implemented by a decentralized system of intelligent software agents and has been evaluated by using two different ontology test sets. The system can work in a fully automatic and autonomous mode and completely fits to fully decentralized and distributed settings. The practical applicability of the results is confirmed by the use of the system for aligning industrial knowledge bases in the field of construction and building materials and services, as well as online product and service ontologies.