NEURAL NETWORK METHOD FOR DIAGNOSING PSYCHOLOGICAL DISEASES BY ANALYZING MESSAGES BASED ON A SEPARATE APPROACH TO CLASSIFICATION

Abstract

The present work considers the use of neural network methods for diagnosing psychological diseases through the analysis of messages. The study shows that the use of NLP technologies and deep neural networks for automated assessment of the psycho-emotional state of individuals can significantly increase the accuracy of diagnostics, as well as ensure scalability and innovation in the field of healthcare. The use of such technologies also corresponds to the global goals of sustainable development, in particular in improving mental health, developing digital infrastructure and reducing inequalities in access to medical services.

The proposed method is based on a separate approach to the classification of various psychological diseases, which increases the accuracy and reliability of diagnostics. Each psychological diseases is analyzed separately, which avoids mutual influence during classification. The diagnostic process includes three main stages: tokenization of texts using appropriate tokenizers, processing of tokens using neural network models, which were trained from scratch on specialized text sets, and drawing conclusions about the probability of the presence of each of the diseases.

Two open datasets were used to train the models. Experimental results showed high efficiency of the proposed method: accuracy values vary within 0.81–0.90, and Precision, Recall and F1-score indicators reach 0.91, which indicates high classification accuracy and ability to differentiate psychological states. The proposed method demonstrates better results compared to existing analogues and has great potential for use in automated detection of mental diseases. Further research can be aimed at improving the robustness of models to language variations and expanding the diagnosed states.