HYBRID BIOINSPIRED ALGORITHM FOR ONTOLOGIES MAPPING IN THE TASKS OF EXTRACTION AND KNOWLEDGE MANAGEMENT
Daniil Kravchenko, Yu. A. Kravchenko, V.V. Markov · Известия Южного федерального университета. Технические науки · 2020
The article is devoted to solving the problem of mapping ontological models in the processesof extracting and knowledge management. The relevance and significance of this task are due tothe need to maintain reliability and eliminate redundancy of knowledge during the integration(unification) of various origins structured information sources. The proximity and consistency ofthe conceptual semantics of the combined resource during the mapping is the main criterion forthe effectiveness of the proposed solutions. The article considers the problems of choosing appropriatesolution approaches that preserve semantics when displaying concepts. The strategy ofchoosing bio-inspired modeling is substantiated. The aspects of the effectiveness of various decentralizedbio-inspired methods are analyzed. The reasons for the need for hybridization are identified.The paper proposes to solve the problem of mapping ontological models using a bio-inspiredalgorithm based on hybridization of bacterial and cuckoo search algorithms optimization mechanisms.The hybridization of these algorithms allowed us to combine their main advantages: a consistentbacterial search that provides a detailed study of local areas, and a significant number ofthe cuckoo agent during the implementation global movements of Levy flights. To evaluate theeffectiveness of the proposed hybrid bio-inspired algorithm, a software product was developed andexperiments were performed on the mapping of different sizes ontologies. Each concept of anyontology has a certain set of attributes, which is a semantic vector of attributes. The degree of thesemantic vectors similarity for the compared concepts of displayed ontologies is a criterion fortheir integration. To improve the quality of the display process, a new encoding of solutions hasbeen introduced. The quantitative estimates obtained demonstrate time savings in solving problemsof relatively large dimension (from 500,000 ontograph vertices) of at least 13 %. The timecomplexity of the developed hybrid algorithm is O (n 2). The described studies have a high level oftheoretical and practical significance and are directly related to the solution of classical problemsof artificial intelligence aimed at finding hidden dependencies and patterns on a multitude ofknowledge elements.