Source-linked AI summary
Bots, #StrongerIn, and #Brexit: Computational Propaganda during the UK-EU Referendum
Philip N. Howard, Bence Kollanyi
TL;DR
Because average users may not distinguish bot from human activity, this study examines tweeting by people and bots during the UK-EU referendum. It finds that less than 1 percent of sampled accounts generated almost a third of messages, while leave-EU hashtags dominated and automation varied by perspective.
Problem
It was unclear whether average users could distinguish bot from human activity, even as political campaigns increasingly automated their messaging.
Source: CONCLUSIONS · p. 5Method
The study compares sampled Brexit-related Twitter activity across hashtag perspectives and levels of account automation.
Source: FINDINGS AND ANALYSIS · p. 4Results
Less than 1 percent of sampled accounts generated almost a third of messages; leave-EU hashtags dominated, and automation levels differed by perspective.
Source: FINDINGS AND ANALYSIS · p. 4, CONCLUSIONS · p. 5Takeaways & Limitations
Bots were used mostly to amplify referendum messages rather than for argumentative engagement or impression management.
Source: CONCLUSIONS · p. 5The hashtag-based dataset misses contributions without the tracked hashtags and excludes newer URL-based retweets with commentary.
Source: FINDINGS AND ANALYSIS · p. 3
Abstract
from arXiv · showhide
Bots are social media accounts that automate interaction with other users, and they are active on the StrongerIn-Brexit conversation happening over Twitter. These automated scripts generate content through these platforms and then interact with people. Political bots are automated accounts that are particularly active on public policy issues, elections, and political crises. In this preliminary study on the use of political bots during the UK referendum on EU membership, we analyze the tweeting patterns for both human users and bots. We find that political bots have a small but strategic role in the referendum conversations: (1) the family of hashtags associated with the argument for leaving the EU dominates, (2) different perspectives on the issue utilize different levels of automation, and (3) less than 1 percent of sampled accounts generate almost a third of all the messages.
FROM SOCIAL BOTS TO POLITICAL BOTS
Social bots automate posting across social media, and political bots are a subset assigned overt political tasks. Political actors use them in polarized moments to influence online political conversation.
FROM SOCIAL BOTS TO POLITICAL BOTS
- Bots range from benign news or feed updates to malicious spam, but all can deploy messages and replicate themselves.Source: FROM SOCIAL BOTS TO POLITICAL BOTS · p. 1
- Social bots are computer-generated programs that post, tweet, or message on their own, and are particularly prevalent on Twitter.Their profiles often lack basic account information, and they access sites directly through APIs for real-time posting and information parsing.Source: FROM SOCIAL BOTS TO POLITICAL BOTS · p. 1
- Political bots are deployed during sensitive, polarized moments to manipulate public opinion, choke off debate, and muddy political issues.Their use varies across regime types.Source: FROM SOCIAL BOTS TO POLITICAL BOTS · p. 1
TWITTER BOTS AND POLITICAL COMMUNICATION
In the StrongerIn–Brexit debate, bots were among the most active accounts and amplified existing messages by mechanically repeating content rather than creating it.
TWITTER BOTS AND POLITICAL COMMUNICATION
- The two most active accounts on each side of the StrongerIn–Brexit debate were bots that mechanically retweeted messages from their respective sides.The accounts, @ivoteLeave and @ivotestay, followed a similar algorithm and generated no new content.Source: TWITTER BOTS AND POLITICAL COMMUNICATION · p. 2
- A repurposed pro-Palestine bot, @Col_Connaughton, and @Rotenyahu amplified #Brexit content by retweeting messages from that account.It is unclear whether any human curation was involved; both accounts used automation to aggregate and repeat content.Source: TWITTER BOTS AND POLITICAL COMMUNICATION · p. 2
SAMPLING AND METHOD
The study analyzes more than 1.5 million tweets collected during June 5–12, 2016, through pro-leave, pro-remain, and neutral referendum hashtags. Because Twitter’s Streaming API provides only a limited and not fully characterized sample, the data cannot establish the total number of bots in the debate.
SAMPLING AND METHOD
- The dataset contains more than 1.5 million tweets and 313,832 distinct accounts, collected June 5–12, 2016, using pro-leave, pro-remain, and neutral hashtags.Source: SAMPLING AND METHOD · p. 2
- The study cannot report the total number of bots in the StrongerIn–Brexit debate; Figure 1 instead shows the sample’s daily tweet rhythm.Source: SAMPLING AND METHOD · p. 2
- Researchers built the hashtag list from smaller datasets to capture terms used across both sides and neutral referendum discussion, then collected matching tweets and analyzed them in R.Hashtag-based sampling was intended to focus on referendum-related content and avoid irrelevant Eurovision tweets.Source: SAMPLING AND METHOD · p. 2
- Twitter’s Streaming API sampling method is unknown, and its available data represent at most 1% of global public communication at a given time [9].Source: SAMPLING AND METHOD · p. 2
- The API returns hashtag- or keyword-matching tweets, linked web sources containing those terms, and retweets with or without the original text included.Source: SAMPLING AND METHOD · pp. 2–3
FINDINGS AND ANALYSIS
Leaving-oriented hashtags were more prevalent in the sampled Twitter conversation, and fewer than 1% of accounts generated 32% of Brexit-related traffic. Automation levels varied across hashtag groupings, while the sampling does not support inference about voter intent.
FINDINGS AND ANALYSIS
- Hashtag counts omit tweets without tracked hashtags, can count multi-tag tweets in multiple groups, and exclude newer URL-form retweets with added comments.The authors caution that sampling parameters are not sufficient to infer how social-media opinion translates to voter intent.Source: FINDINGS AND ANALYSIS · p. 3
- #voteleave appeared 341,839 times versus 110,653 uses of #StrongerIn, and Brexit-perspective users generated more content and tagged it better.Most instances of #brexit accompanied messages about leaving the EU, although the hashtag was also used by both sides; these counts do not establish voter intent.Source: FINDINGS AND ANALYSIS · pp. 3–4
- 5.7% of accounts using exclusively neutral hashtags used heavy automation, versus one-third of tweets using mixed hashtags generated by heavily automated accounts.Heavy automation was defined as at least 50 posts a day, or 350 or more tweets during data collection; the threshold can also include extremely active humans.Source: Table 2 · p. 4, FINDINGS AND ANALYSIS · p. 4
- 32% of Brexit-related Twitter traffic came from fewer than 2,000 users (less than 1% of accounts) posting 100 or more related hashtags that week.The top 100 users generated more than 121,000 tweets, about 8% of all Twitter traffic related to StrongerIn-Brexit.Source: FINDINGS AND ANALYSIS · p. 4
- All top 10 accounts by volume appeared to use some automation; researchers judged 7 almost certainly bots, and these top users mainly retweeted others' content.One account was UKIP-curated and probably automated, while two appeared to be bots with some human curation.Source: FINDINGS AND ANALYSIS · p. 4
CONCLUSIONS
Political bots played a small but strategic role in StrongerIn-Brexit conversations, mostly amplifying messages rather than engaging in argument. The study identifies their communication patterns but cannot determine their influence on public sentiment.
- Bots mostly retweeted referendum content, and the sampled bots amplified messages rather than engaging in argument or impression management; their traffic was largely non-original.Source: CONCLUSIONS · p. 5
- Repeating the study over a longer period and closer to voting day could reveal additional features of bot activity on the referendum.Source: CONCLUSIONS · p. 5
- Less than 1 percent of sampled accounts generated almost a third of all messages, while pro-leave hashtags dominated and perspectives used different levels of automation.Source: CONCLUSIONS · p. 5
- The study identifies bots’ role in political communication but cannot say how much social-media discourse or bots shape public opinion on the referendum.Source: CONCLUSIONS · p. 5
ABOUT THE PROJECT
The Project on Computational Propaganda is an international, interdisciplinary team investigating how automated scripts affect public life. It tracks social bots and draws on multiple disciplines to interpret its evidence.
ABOUT THE PROJECT
- The project tracks social bots and analyzes evidence through organizational sociology, human-computer interaction, communication, and political science.It investigates the impact of automated scripts, described as computational propaganda, on public life.Source: ABOUT THE PROJECT · p. 6