TEXT CLASSIFICATION METHOD AND SYSTEM Russian patent published in 2024 - IPC G06F40/279 G06F16/35 G06N20/00 

FIELD: physics.

SUBSTANCE: group of inventions relates to computer engineering and can be used for additional training of a language model for solving the problem of text classification. Method of classifying text by a language model is carried out by at least one computing device and comprises the following steps: obtaining an input data set corresponding to the required classification task, in the format on the basis of which the language model is additionally trained; formatting it, supplementing it with symbols, each of which corresponds to an abstract pseudoword; performing tokenization and vectorization of the input data set, wherein symbols corresponding to abstract pseudowords are replaced with trained vector representations of symbols; processing the obtained data, obtaining a vector of logits, which reflects the probability distribution of classes corresponding to words of the dictionary of the language model; selecting target components of logits, corresponding to tokens of target classes of solved problem of classification; determining the logit component reflecting the highest probability of belonging to the target class; response is generated in text form corresponding to the selected component.

EFFECT: enabling automatic generation of hints for additional training of the language model.

7 cl, 5 dwg