Bot detection by friends graph in social
networks St. Petersburg Federal Research Center
of the Russian Academy of Sciences, St. Petersburg, Russia {kolomeec,
chechulin, ivkote}@comsec.spb.ru Abstract In this paper, we propose a machine learning approach to detecting malicious bots on the VKontakte social network. The advantage of this approach is that only the friends graph is used as the source data to detect bots. Thus, there is no need to download a large amount of text and media data, which are highly language-dependent. Besides, the approach allows one to detect bots that are hidden by privacy settings or blocked, since the graph data can be set indirectly. To do this, we calculate graph metrics using a number of graph algorithms. These metrics are used as features in classifier training. The paper evaluates the effectiveness of the proposed approach for detecting bots of various quality - from low quality to paid users. The resistance of the approach to changing the bot management strategy is also evaluated. Estimates are given for two classifiers - a random forest and a neural network. The study showed that using only information about the graph it is possible to recognize bots with high efficiency (AUC-ROC greater than 0.9). At the same time, the proposed solution is quite resistant to the emergence of bots having a new management strategy. Keywords: +: Corresponding author: Igor Kotenko Journal of Wireless Mobile Networks, Ubiquitous
Computing, and Dependable Applications (JoWUA), Vol. 12, No.
2, pp. 141-159, June 2021 [pdf] DOI: 10.22667/JOWUA.2021.06.30.141 |