The document proposes a novel method for detecting malicious social bots based on analyzing clickstream sequences. It presents a method that uses transition probabilities of clickstream sequences and semi-supervised clustering. The method was tested on 450 thousand data points collected from a social media platform. It was able to detect malicious social bots with 12.8% higher accuracy than detection methods based only on quantitative analysis of user behavior. Diagrams of the proposed system are included showing modules like the OSN server, user modules, and examples of interfaces.