In the following paper the authors present a GAN-type model and the most important stages of its development for the task of emotion recognition in text. In particular, we propose an approach for generating a synthetic dataset of all possible emotions combinations based on manually labelled incomplete data. Dataset vectorizationInitially, it is necessary to form a dataset for training the model. The dataset should contain data about seven basic emotions for each certain piece of text -a paragraph or a sentence. The authors couldn't find a publicly available dataset with this or similar data. Therefore, it was