EVALUATION OF THE EFFICIENCY OF ARTIFICIAL NEURAL NETWORK MODELS IN THE PROCESS OF THEIR INTEGRATION INTO SYSTEMS FOR MONITORING POLITICAL INTERNET DISCOURSE

The article highlights the problem of evaluating the effectiveness of artificial neural network models in their integration into systems for monitoring political Internet discourse with a complex integrated natural language. The author emphasizes the prospects, relevance, etc. of using innovative tools in modern linguistic research. The article is about the actualization in such studies of neural network models (in particular, with the Transformer architecture), which show high efficiency in automating the processing of textual data (including political Internet discourse). The analysis showed that evaluating the effectiveness of such models solely based on conventional computational metrics designed to work with structured data is not productive. That is why it is important to develop clear, transparent, and logical linguistic criteria, a variant of which is presented in this study. The reliability of the results of processing and analyzing textual data by neural network models correlates with integrating linguistic verification methods into their work at all stages of training and testing of the latter. This, in turn, will significantly reduce the risk of misinterpretations of politically labeled statements, which is especially productive when this tool is used in political discourse monitoring systems. Developing interpreted neural network architectures is also important, as they will ensure transparency, logic, and consistency in the text data processing process. In addition, such architectures will improve the quality of linguistic analysis and contribute to a deeper understanding of how models form semantic representations. This is also important in actualizing such systems in the political context, as a basis for making certain political decisions. The prospect of further research on the analyzed problem is to improve approaches to training neural network models that will consider the dynamic changes in political online discourse and the variability of its contexts. This improvement will significantly increase the efficiency of using neural network models in a rapidly changing information environment and changes in digitalization. Key words: neural network modeling, neural network modeling research, applied linguistics, computer linguistics, computational linguistics, artificial neural networks, neural network models, evaluation of neural network models.