NRNU MEPhI has developed a comprehensive system for intelligent analysis of heterogeneous data, the university's press service told "Pervy Tekhnichesky". It works with unstructured arrays – from scientific articles and patents to social media profiles. The system was created under the guidance of Doctor of Technical Sciences Alexey Artamonov.
Classical approaches are ineffective for such data, the scientist explained. An article is not just text, but a network of connections between authors, organizations, topics, and citations. To extract knowledge, a different methodology is needed.
The system is based on a unified digital object model. Any data is described identically: static properties, dynamic indicators, and a graph of connections. This allows for flexible processing of information from various sources.
The system automatically extracts physical quantities, organization coordinates, key terms, and international cooperation data from texts. For visualization, 3D maps of scientific directions – scientific and technological landscapes – were created. They help compare the dynamics of scientific development across countries, identify growth points, and evaluate cooperation. Such tools are in demand at the state level.
Currently, the system uses classical machine learning methods. Plans include integration with neural networks for deeper context analysis. The system has already been tested in nuclear energy, medicine, and financial security. Development has been ongoing since 2008.
Комментарии