Volume 16, No 1, 2019
Gaming Method of Ontology Clusterization
Petro Kravets, Yevhen Burov, Vasyl Lytvyn and Victoria Vysotska
Abstract
In the real world an intelligent system often consists of intelligent agents, each having its own perspective and goal and executing the common task interacting with others. Those agents are often created by different developers, are based on different conceptualizations of subject area and are working with incomplete and error-riddled data. In conditions of uncertainty of ontologies, the application of the traditional methods of cluster analysis, which assume a clear breakdown of the initial set into subsets, in which each element gets into only one cluster, is not always correct. In this work is proposed the stochastic game method of ontology clustering for optimization of operations under conditions of uncertainty. The essence of game clustering is that intelligent ontology agents randomly select one of the clusters. For agents who chose the same cluster, the current measure of similarity of ontologies is calculated. This measure is used to adapt the recalculation of mixed player strategies. In the game process the probability of choosing clusters, the current composition of which led to an increase in the degree of similarity of ontologies is increased. During the repetitive game, agents will form vectors of mixed strategies that will maximize the averaged measures of similarity for clusters of ontologies. The results of the computer simulation confirm the convergence of the game method during the clustering of ontologies with the conditions and constraints of stochastic optimization.
Pages: 55-76
Keywords: Stochastic game; Clustering, Ontology; Knowledge base; Intelligent agent