On the Usage of Semantic Text-Similarity Metrics for Natural Language Processing in Russian

This work is devoted to the analysis of the applicability of text processing methods in natural languages to the problem of assessing the similarity of text sequences of arbitrary length in Russian. This task is a particular problem that finds its application in many areas of computational linguistics, primarily in the analysis of the effectiveness of the computer-generated text. The analysis revealed the advantage of using the BERTScore method, based on the semantic similarity of text attachments, especially in terms of the reliability and sequence of detecting semantic similarity of sentences compared to traditional text metrics based on character-by-character analysis of texts.