Pro-Trump highly automated accounts ‘colonised’ pro-Clinton Twitter campaign

A growing number of political movements are employing both people and ‘bots’ to shape political conversations and influence election results. Bots can deliver news and information but also undertake malicious activities, while passing as human users. Researchers have revealed the scale of automated account activity during the US Election. Using a sample of 19.4m tweets taken between 1-9 November containing political hashtags like #CrookedHillary or #DraintheSwamp, they show pro-Trump activity increased over the period and was used far more than the pro-Clinton camp, from a ratio of 4:1 during the first debate to 5:1 by the time of election day. Pro-Trump traffic effectively ‘colonised’ pro-Clinton hashtags in tweets that had combinations of neutral and pro-Clinton hashtags, adds the working paper, so by the time of the election, 81% of the highly automated content involved some form of Trump messaging.

The research by Professor Philip Howard from the University of Oxford, with Bence Kollanyi from the University of Corvinus and Samuel Woolley from the University of Washington reveals that Trump supporters’ use of highly automated accounts was ‘deliberate and strategic’. As early as the evening of 8 November, as Donald Trump’s victory became clear, traffic from automated pro-Trump activities suddenly stopped. By contrast, during the debates, the researchers estimate that highly automated accounts generated between 23-27% of the Twitter traffic about the politics relating to the election, dropping to 18% during the lead up to the election.

From their sample in the lead up to voting day, the researchers show the overall level of pro-Trump Twitter traffic at 55% was much higher than the volume of tweets containing only hashtags associated with Clinton at 19%. A ‘fairly consistent proportion’ of the traffic on associated hashtags (such as #ImWithHer, #strongertogether, #CrookedHillary and #SendHerToJail) was generated by highly automated accounts. The working paper explains that these automated accounts are often bots that have occasional human curation such as by using scheduling algorithms and other applications for automating social media.

Researchers identified automated accounts as those producing at least 50 tweets a day with political hashtags associated with Clinton or Trump or the election. They found that the pro-Trump campaigners and programmers ‘carefully adjusted the timing of their tweets during the debates’. While activity from pro-Clinton and pro-Trump automated accounts occurred around the same time during the first debate, by the third debate the tweets with pro-Trump hashtags were going out much earlier than the pro-Clinton tweets. The pro-Clinton camp increased their use of highly automated accounts over the course of the campaign period but never reached the level of automation behind the pro-Trump traffic, says the research.

Twitter activity was also found to reflect the shift towards Trump that was largely undetected by traditional polling, says the research. In the last debate, 30% of the traffic about the debates used relatively neutral hashtags, says the paper. In the lead up to voting, however, only 15% of traffic used neutral hashtags, with pro-Trump traffic rising from 46% to 55%, while pro-Clinton traffic rose from a lower level of 10% to 19%.

Philip Howard, Professor of Internet Studies at the Oxford Internet Institute, comments: ‘We can’t say who was behind the highly automated pro-Trump accounts but they were purposeful, thoughtful and deliberate about when to release messages, what those messages should be, and what their targets were. We found similarities in the rhythm of the pro-Trump Twitter campaign and that successfully employed by the Leave campaign in the run up to the EU referendum in the UK.

‘The political success of movements using a high level of campaigning on Twitter suggests that social media are a very powerful tool in politics. The social polarisation we have witnessed could be set to snowball if algorithms continue to shape the news we receive and people start to “unfollow” friends and neighbours who voted differently to themselves. Another question raised is whether social media firms who reap the advertising revenue should be made more responsible for the “fake news” that spreads during the course of highly charged political Twitter campaigns.’

For more information, please contact the University of Oxford News Office on +44 (0) 1865 280534 or email: [email protected]
Or contact Professor Howard at [email protected]

The working paper, Bots and Automation over Twitter during the US Election, can be viewed at: www.politicalbots.org