INFORMATION EXTRACTION FROM UNSTRUCTURED TEXTS FOR DEEP LEARNING OF LANGUAGE MODELS

The concept of obtaining factual data from natural language texts, which are materials of political articles in electronic mass media (mass media), is presented. The article considers the original architecture of an intellectual system that uses instrumental means of semantic patterns, hints, and procedures for expanding data elements to extract information. This toolkit allows you to create large sequences of data for deep learning of neural network language models. The accumulation of facts makes it possible to logically substantiate the results of the automatic classification of political articles into unreliable, toxic, or vice versa, based on truthful data.