Analysis of User Groups in Social Networks to Detect Socially Dangerous People

摘要

The article proposes a method for identifying socially dangerous people on the basis of metadata left by a user during registration in groups of social networks. This method will allow law enforcement agencies to identify socially dangerous elements of the society that actively use social networks in an unattended mode. Additionally, the method functions without violating the constitutional rights of users to privacy, with respect to personal correspondence. To increase the accuracy while identifying socially dangerous people, it was proposed to use a neural network that allows solving problems with reinforcement learning techniques. The article focuses on structuring the analytical system of social groups, on the scheme of an adaptive neural network, on the stages of the dictionary compilation for an adaptive neural network. Further, the article describes a software package with the help of which the idea of identifying socially dangerous people was realized. The authors provide the code for processing the request for text classification in the server of the neural network. The program complex is written implementing two programming languages: JavaScript and Java. To confirm the program functioning, the screen shot of the program-testing interface is given. It illustrates the process of identifying socially dangerous people on the basis of three selected criteria: normal, aggressive and suicidal. The results were verified by a psychologist on the basis of a special projective methods of personality assessment.