Insulting speech acts have become the subject of public discussion in the media, social media, the basis for speculation in political communication, and a working concept in the legal environment. The present research article explores insulting speech acts on the social network site “VKontakte” aiming to develop an algorithm for automatic classification of text data. We conducted semantic analysis of the text of “Article 5.61” of the Code of Administrative Offenses of the Russian Federation, which made it possible to formulate inclusion criteria for formal classification. We used three common word embeddings models (BERT, ELMo, and fastText) on the original Russian language dataset consisting of 4596 annotated messages perceived as insulting speech acts. General findings argue that even in a specialized dataset the share of messages that meet criteria of inclusion is negligible. This indicates a low probability of going to court on the fact of an administrative offense under Article 5.61 based on speech communication on social network sites, even though such communication is public in nature and is automatically recorded in writing. Machine learning text classifier based on BERT model showed best performance.