Bot Recognition in Social Network

Social networking has now become a popular way of communication and sharing data between people. People spend a lot of time on the Internet browsing social networks: discussing news, sending messages and reading articles. Algorithms on social sites help people find interesting new content, meet new people based on the reaction of other users to events. If the reaction is incorrect, then the rating of the content or user will be calculated with an error. The accounts of some users try to break this algorithm or send spam messages, let's call these accounts bots and spammers. In addition, bots often forward ads and phishing emails to users. This harms the social network, which puts user security at risk and cannot use the platform and generate useless content that is recommended by recommendation systems. The way to deal with bots is manual markup of spam accounts. VK is the largest social network for Russian-speaking users and CIS countries. The number of users on VK is 587'000'000, so it takes a lot of time, resources and human effort to search for bots and spammers. The aim of this work is to develop a solution based on machine learning to solve the problem of classifying users into three classes: trusted, falsified and hacked users. To do this, it is necessary to collect data, prepare them for use, develop a model and test on real data. As a result of the work, a tool was obtained for streaming user processing, which includes collecting user data, processing and converting data, and also deciding on the classification problem based on the information collected.