Early detection of internet trolls: Introducing an algorithm based on word pairs / single words multiple repetition ratio

Troll internet messages, especially those posted on Twitter, have recently been recognised as a very powerful weapon in hybrid warfare. Hence, an important task for the academic community is to provide a tool for identifying internet troll accounts as quickly as possible. At the same time, this tool must be highly accurate so that its employment will not violate people’s rights and affect the freedom of speech. Though such a task can be effectively fulfilled on purely linguistic grounds, as of yet, very little work has been done that could help to explain the discourse-specific features of this type of writing. In this paper, we suggest a quantitative measure for identifying troll messages which is based on taking into account certain sociolinguistic limitations of troll speech, and discuss two algorithms that both require as few as 50 tweets to establish the true nature of the tweets, whether ‘genuine’ or ‘troll-like’.