Detection of depression among social network users using machine learning methods

Statistical data provided by the FSBI “NMITSPN named after V.P. Serbsky” of the Ministry of Health of Russia indicate that depression, as a psychoemotional state, is the main cause of concern around the world, which in most cases leads to suicide, if not detected, and to a threat to others. Studies show that depression tends to have an impact on writing style and appropriate language use. The main purpose of the proposed study is to study user messages on the VKontakte social network and identify attributes that may indicate depressive symptoms of users. The article uses machine learning approaches (logistic regression, random forest, support vector machine, XGBoost) and natural language processing methods (removal of stop words, character deletion, tokenization, lemmatization) to prepare data and evaluate their effectiveness. The work demonstrated that the ability to search for depressed users with an accuracy of 77% using the XGBoost classifier. This method is combined with other linguistic functions (N-gram + TF-IDF) and LDA to achieve higher accuracy. In conclusion, the main conclusions of the study are formulated.