How many hyperactive Twitter accounts do you need to predict the diffusion of political information on Twitter?

The study investigates the impact of hyperactive accounts in the diffusion of political information on Twitter during the 2019 European parliamentary election campaign in Poland. We examined interactions between 114,036 randomly selected Twitter users in Poland and 936 most-followed accounts (referred as political opinion leaders) that were systematically publishing political content during the 2019 European parliamentary election campaign (1st February 2019 - 26th May 2019). The collected data included the most trending hashtags on Polish Twitter (for each day), all statuses (tweets, retweets, quotations, replies; N=1,793,220) published by accounts in the sample, all likes given to leaders’ tweets by the accounts in our sample, and all statuses published by opinion leaders (N=826,050). We identified 626 social-bot-like (semi)-automated or human-controlled hyperactive Twitter handles using the isolation forest method on their behavioural features. The purpose was to examine 1) the differences between hyperactive and regular accounts in their interactions with political opinion leaders, and 2) the relationship between hyperactive accounts’ reactions to opinion leaders’ tweets and the number of retweets the latter obtain. This study shows that less than 0.5% of the randomly selected accounts substantially outperformed the remaining ones in terms of the number of interactions with Twitter opinion leaders. They were responsible for almost all likes, retweets, quotations, and replies. The results of random forest regression (R-squared = 0.78) reveal that the likes and retweets from analyzed hyperactive accounts were the most important predictors of the dissemination of political elites’ tweets in general, even if such factors as the number of elite accounts’ followers, the type of account (politician, journalist, celebrity etc.), political orientation or the inclusion of a trending hashtag in a tweet were controlled. It may suggest that a small number of accounts are enough to impact information flow and change what Polish Twitter users are exposed to. We used the results to discuss the interactional asymmetry on Twitter, its consequence for disseminating political information, and predictive power of information about hyperactive accounts’ behaviours.