Development of a Neural Network Model for Recognizing Russian-Language Generated Texts

With the rapid development and growing popularity of text generation models, the task of automatically collecting data from open sources for further research (e.g., for application in psychometric models) faces a new challenge: is the resulting text human or generated using neural network models and large language models? In order to solve the problem of collecting human texts within the framework of the project aimed at forming an extended model of cognitive-behavioral concept revealing the main predictors of human behavior, the existing models of text classification were investigated - Logistic Regression, Stochastic gradient descent, Extreme Gradient Boosting, convolutional neural network and RuBERT model fine-tuning. The models were trained on a composite Russian-language dataset and validated on a dataset including our sample of human texts obtained from verified sources. The best performing models will be embedded in an information and analytical system to collect and analyze empirical data from users' textual publications in a digital environment.