You are here

Different Bots in the 2016 U.S. Presidential Election

The next speaker at AoIR 2017 is Olga Boichàk, who begins by highlighting the role of social media platforms in structuring specific forms of human sociality. But this also means that automated accounts – specifically, bots – can imitate and affect genuine human interactions in these spaces. What does this mean for online discussions in the context of the 2016 U.S. election campaign, then?

This project draws on the Illuminating 2016 research project that gathered some one billion social media messages, and focussed especially on major retweet events (where a candidate's message is widely retweeted by a substantial number of social media users). The retweeting accounts were then tested through the Botometer service, which uses three families of markers to assess the 'bot-ness' of these accounts; this is not entirely reliable, of course, but provides a useful first approximation of participation patterns.

Donald Trump's retweet events turned out to be substantially larger than Hillary Clinton's; the minor candidates in the election received considerably less engagement, unsurprisingly. Of the retweeters, some three quarters were considered by Botometer to be human-operated; bots were a relatively minor group, but interestingly there were also notable groups of protected or deleted accounts.

Bot accounts had far more followers on average than human accounts, as well as a much higher tweet rate; many of these bots were news bots, aggregator accounts (that retweet all Trump-related messages from Miami, for instance). Another category of accounts, however, had virtually no followers, but posted a very substantial number of tweets.

Retweet events, in turn, can be classified according to the steepness of the initial spike in retweeting, and the speed of the subsequent decay in interest. Trump's events tended to be more spiky and rapid, while Clinton's were less drastic; events for Greens candidate Jill Stein (which were far less pronounced, of course) did not follow this pattern at all, and did appear to be driven largely by bot networks.

Ultimately, this means that not all bots are equal; there are many different forms of bots and they plat distinct roles in information diffusion.