On reduction of the results of ontological analysis of object-attribute data about a subject area

This paper explores the development of ontological data analysis, an author’s methodology for generalizing empirical object-attribute data on the studied subject area. The practical aspect of this methodology is to support the building (logical inference) of a formal ontology of the probed reality. Ontological data analysis relies heavily on methods of formal concept analysis of object-attribute data, which are based on mathematical lattice theory and, despite having objective advantages, are characterized by high computational costs, as well as the large dimensionality and structural complexity of the hierarchy of formal concepts about the subject area extracted from the source data. The subject of the study is the reduction of the ontological description extracted in this way while preserving in it information about the subject area that is particularly important for the research subject. It is known that it is possible to evaluate such subjective interest of the researcher using various indices of the interestingness of formal concepts. However, a separate subset of the most interesting formal concepts does not always represent a correct ontological construction. The challenge was to develop methods for building a formally correct ontological description based on such a subset. The article proposes two effective methods for solving it, ensuring the reduction of the ontological description initially extracted from the data. In developing a pragmatically oriented method, a hypothesis was put forward about the prevailing need of the subject to preserve in the reduced ontology a description of interesting formal concepts with non-empty own extent, the analogue of which in object-oriented programming are data-classes. The semantically oriented method is free from such a hypothetical assumption. The effectiveness of the developed methods is compared. The presentation of the material is accompanied by an example of processing simplified object-attribute data on 4th generation fighters.