METHOD FOR DETECTING PSYCHOLOGICAL DIGITAL OVERLOAD BY ANALYZING TEXT DATA WITH DEEP LEARNING NEURAL NETWORK MODELS

Abstract

The paper presents a method for detecting psychological digital overload, based on the use of deep learning neural network models, in particular the BERT architecture adapted for binary classification of text content. The research is aimed at solving the scientific and applied problem of automated analysis of the psycho-emotional state of users in the digital environment, which arises under the influence of prolonged interaction with digital technologies.The relevance of the proposed approach is due to the growing need for objective, scalable and interpretable means of identifying cognitive overload, which are able to function without expert intervention. The method is implemented as a seven-stage procedure that includes text preprocessing, neural network classification, thematic modeling, isolation of target objects of the subject area and the formation of a multi-level assessment of psychological digital overload with subsequent visualization of the results. The basis of the computational core is the BERT model, which forms contextualized representations of text sequences and provides high accuracy in detecting latent states of users based on the analysis of their digital speech. Semantic detailing is implemented by integrating thematic classification with key concept detection methods based on NER, TF-IDF and variance analysis, which allows determining information centers of cognitive load within text profiles. The model was tested on open corpora of social messages, where the values of Accuracy 0.83, Precision 0.87, Recall 0.88 and F1-score 0.87 were achieved for the main metrics. The obtained results demonstrate the suitability of the neural network approach for solving the problem of classifying psychological digital overload as a state manifested through language indicators. The method allows not only to carry out an integral assessment of the level of psychological digital overload, but also to localize its manifestations in thematic segments of the text, which forms the basis for interpreting the classification decision and making management decisions in the field of digital interaction.The results of the study can be applied to the development of information systems to support mental well-being, adapting digital services to the individual user load and monitoring psychological risks in communication-saturated environments. The proposed approach expands the tools of natural language analysis in the context of psycho-emotional diagnostics and contributes to the formation of a scientifically sound basis for the implementation of artificial intelligence methods in the field of digital hygiene.