Identification of a Depressive State Among Users of the Vkontakte Social Network

Abstract

The article examines the analysis of the depressive state of social network users. It is noted that the VKontakte social network will be used as a social platform for collecting information in the study. It is noted that a combination of vocabulary-based and machine learning methods is used to achieve the highest accuracy. Two methods based on vocabulary are considered: the dictionary-based method and the corpus method. The stages of analysis are considered: collecting data obtained using the VK_API script creation module, preprocessing data through the natural language processing pipeline (deleting raw data that does not carry a semantic role), creating a model and evaluating it. It is noted that the implementation of this task uses a high-level Python programming language with dynamic strict typing and automatic memory management, the syntax of which contains a natural language processing module (NLTK). The paper presents 4 machine learning classifiers: support vector machine (SVM), k—nearest neighbor method (KNN), random forest, logistic regression, LSTM. It is revealed that machine learning algorithms such as decision tree, support vector machine, logistic regression and LSTM demonstrate good accuracy in detecting the depressive mood of a social network user. The LSTM network showed the greatest accuracy during this experiment. In conclusion, the main conclusions on the work done are formulated.