USING GENERATIVE ALGORITHMS TO GENERATE DOCUMENTS

Kasymov Alexey Alekseevich, Maximov Yuri Maksimovich

This article provides a brief overview of the latest text classification models with an emphasis on data flow, from raw text to output labels. The differences between earlier methods and later methods based on deep learning are emphasized, both in their functioning and in how they transform input data. To give a better idea of text classification, an overview of the data sets for the language is provided, as well as instructions for synthesizing two new data sets with multiple labels. At the end, we describe an overview of new experimental results and discuss the problems of open research related to language models based on deep learning.

Keywords:text classification; tokenization; topic labeling; news classification; transformer; surface learning; deep learning; multicomponent corpora