Ilgar V. Amiraslanov, Nikita O. Dorodnykh, Alexander Yu. Yurin

Irkutsk National Research Technical University, Matrosov Institute for System Dynamics and Control Theory SB RAS

Today, knowledge graphs are widely used in various domains, for example, in industry, commerce, finance, and social networks. A knowledge graph is a powerful means of information combination and representation using standardized knowledge modelling methods. However, the development of knowledge graphs and, in particular, their population with new specific entities (facts) remains a difficult task. The use of various information sources can facilitate this process. Such a source can be tables that potentially contain rich semantic information. In this paper proposes an approach and its software implementation for automated extraction of significant information from tabular data in the form of facts and population of a target knowledge graph with them. The main feature of the proposed approach is the combination of heuristic methods with deep machine learning models for semantic table annotation. The applicability of the proposed approach is demonstrated by two examples: labor market analyzing for the Irkutsk region and assessing the technical state of petrochemical equipment.

semantic web, knowledge acquisition, knowledge graph, semantic table interpretation, fact extraction, knowledge graph population, table