Abstract
Knowledge graphs become an important form of representing data and information. Their intrinsic ability to express semantics via relations enables development of novel methods of processing data and building data models.
In the paper, we propose a methodology for generating definitions of concepts and constructing their hierarchy. It is a fully data-driven process that uses information about multiple entities represented in a form of a knowledge graph. In this work, we state that a concept is defined via relations between the concept and other concepts. We perform a thorough analysis of relations and determine their levels of importance and degrees of their contributions to the definitions. This allows us to include impression reflecting the dependence of the construction process on the context in which it is performed - a limited amount of available data in our case.
We provide details of the proposed approach, and illustrate its performance presenting a case study using a set of facts from dbpedia.org. In the study, we construct a structure of concepts and investigate how importance of relations between them changes when levels of concept abstractions change.