METHOD FOR DETERMINING DEPRESSIVE STATES ASSOCIATED WITH LEARNING IN EDUCATIONAL INSTITUTIONS USING DUAL ARCHITECTURE NEURAL NETWORK

Abstract

Article identifies the problem relevance of automated detection of depressive states associated with learning in educational institutions, relevance in the modern social and academic environment, when pressure, stress and anxiety have become widespread phenomena that can contribute to the development of depression. This is especially relevant in conditions of intensive educational process, high demands and limited time for rest and self-regulation. Detection of depressive states at early stages can significantly affect the timely support and prevention of more serious mental disorders, therefore this issue in the mental health field has become widespread in scientific research and correlates with the sustainable development goals of UNDP. Paper proposes a method for detecting depressive states associated with learning in educational institutions using a dual architecture neural network, which is designed to convert input data in the form of text and a trained neural network model into output data in the form of a numerical assessment of the presence of a depressive state. The proposed method differs from analogues in that it combines a dual-stream architecture, which is based on the use of two parallel neural networks of the transformer architecture, each of which specializes in the analysis of different aspects of the text – syntactic and semantic. The syntactic analysis stream is aimed at identifying the syntactic structure of the text, and the semantic analysis stream is aimed at understanding the content and context of text. Research of developed method effectiveness in the form of revealed software implementation, in comparison with the known analogues given in previous studies, the ROC curve area value of 0.98 was achieved, which is 0.1 higher than the analogue implementation of the BERT neural network and 0.12 higher than the analogue implementation of the RedditBERT neural network. The implementation of the method for detecting a depressive state associated with learning using dual-architecture neural network contributes to the implementation of Sustainable Development Goals SDG3 and SDG4. This allows for a healthy lifestyle and well-being of participants in the educational process SDG3 through timely intervention in their mental state, and also contributes to ensuring high-quality, inclusive and equitable education SDG4 by creating supportive learning environment.