Politics-oriented online communities on social networking sites are becoming more influential in the conditions of lack of the independent media agenda. But how can the fragmented landscape of topics and ideologies represented by them be described using (para)linguistic features? In this research, we consider the sample of politics-related communities from the most popular Russian social networking site VKontakte. Additive regularization algorithm allowed us to classify the communities with similar topics and describe 9 salient clusters. It was shown how the thematic focus of the particular clusters is framed by linguistic and paralinguistic features, such as morphological specificities, tone, punctuation and so on. We propose a framework and a pipeline for the analysis of communicative strategies of communities discussing politics which could be used as a preliminary stage for the qualitative content analysis.