Modern Possibilities of Joint Application of Ontologies and Machine Learning for Working with Data

This paper analyzes the problems of the joint application of machine learning methods and ontologies. It explores the potential of the neurosymbolic approach to address the well-known issue of limited interpretability and explainability in machine learning outcomes. An analysis of recent literature has revealed that the combination of symbolic knowledge representation in the form of ontologies with neural networks has given rise to a new field of research: ontology-driven neuro-symbolic intelligence. This field is actively developing graph database platforms, vector storage techniques, and large language model technologies. The paper presents the first results of testing neuro-symbolic technologies using the AllegroGraph database and thermophysical ontology as an example.