UDC 007:681.512.2
NEURAL NETWORKS WITH KNOWLEDGE FOR BIG DATA ANALYSIS
I. Yu. Kashirin, Dr. in technical sciences, full professor, RSREU, Ryazan, Russia;
orcid.org/0000-0003-1694-7410, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The most important task to design learning algorithms for mining big data, including neural network, is considered. To improve their efficiency, taxonomies for artificial intelligence knowledge models are used. It is proposed to use ICF contiguous inheritance relation as basic taxonomy relationship. This approach makes it possible to effectively design general and applied ontologies to solve artificial intelligence problems. The article uses mathematical apparatus of hierarchical numbers allowing to analyze applied ICF ontologies by calculating the main characteristics of taxonomic hierarchies. Such characteristics include, for example, a measure of concepts semantic similarity or the complexity of structural path in taxonomic hierarchy between two concepts. As an example that confirms the possibility of effective use of ICF taxonomies in solving real problems of data analysis, we propose the solution of clustering and classification problems in «Reducing churn and attracting customers of telecommunication company» subject area. For this problem the possibility of interdependent design of neural networks and applied ontologies with a priori libraries of causal taxonomies, for example, in Semantic Web technology, is shown. In the experimental part, to test the capabilities of information retrieval in global networks based on Semantic Web technology using ICF taxonomy, an experimental bench program for data mining methods (ESMIAD v.12.02.2021) is considered, which allows by placing fragments of program text in comments or vice versa perform experiments for analytical algorithms with training samples. The experimental stand is implemented in Python v.3.7 in Anaconda v.3.0 development environment. The purpose of this stand is to analyze the effectiveness of data mining methods using knowledge representation models.
The aim of the work is to analyze design features of learning algorithms for data analysis using general and applied ICF ontologies with the apparatus of hierarchical numbers to efficiently solve urgent problems of clustering and classification.
Key words: knowledge models, big data, artificial intelligence, neural networks, ICF ontology, ontological taxonomies, hierarchical numbers, software tools, measure of concept semantic similarity, Semantic Web technology, Python, Anaconda, generic and causal relationships.