We know Donald Trump likes to tweet. Given his waning popularity, he probably needs those social media dopamine hits more than most of us. However, while Donald Trump’s actual popularity among living humans remains the stuff of intrigue, especially with accusations of various forms of electoral manipulation by Cambridge Analytica, the Russians, and bots, his Twitter numbers are easier to gauge. As of 13 November 2017, Trump has about 43 million followers. Many of these are fake accounts, but that in itself is not unusual.
What is perhaps more interesting is how bots can and do boost the popularity of Trump’s posts. This post focuses on Trump’s recent Tweet in support of King Salman of Saudi Arabia. Specifically, Trump said: “I have great confidence in King Salman and the Crown Prince of Saudi Arabia, they know exactly what they are doing”
I have great confidence in King Salman and the Crown Prince of Saudi Arabia, they know exactly what they are doing….
— Donald J. Trump (@realDonaldTrump) November 6, 2017
See, he really said it.
As with most of his tweets, Donald’s pro-Salman Tweet made the news. Trump’s support for Salman is, of course, not unusual, but it is problematic. However, this post is not about US-Saudi relations, however grim they are. Instead, this post demonstrates how the tweet was retweeted by thousands of bots. This practise, can be termed megaphoning – the practise of automated amplification of tweets/online speech.
Graphs. graphs, and more graphs
Look at the graph below. Each individual bar represents an individual tweet sent by Donald Trump between the 1st and 10th November 2017. The height of the bar shows the amount of retweets given to each of the Real D’s tweets, while the horizontal axis shows the time the tweet was sent.
As you can see, there is a clear spike in the number of ‘retweets’ on 7th November 2017. This spike is the tweet praising King Salman of Saudi Arabia.
When this sample was taken (between 7th and 9th November), the tweet had been retweeted 82,946 times. As shown by the dashed line in the above graph, the average number of retweets per Donald Trump tweet (in this sample) is around 17k. A study in May showed that the average (mean) number retweets for Trump was 13,100 retweets per tweet (Bear in mind this has probably increased since May). So as you can see, the tweet praising King Salman was noticeably more popular than Trump’s usual tweets.
So what, you might add? Well, given that there is no real reason to see that such a tweet should be more popular than any other of Trump’s tweets on Saudi Arabia. (Trump’s last tweet on Saudi Arabia had a modest number of tweets in comparison). I thought I’d investigate those accounts that retweeted it.
I downloaded a sample of around 24,704 accounts that retweeted the popular tweet. It is useful as a starting point when locating crude Twitter bots to see if a number of those accounts were created around the same time. Unusually large spikes in the creation of Twitter accounts on the same day/month indicate suspicious activity, especially when those bots happen to all be tweeting about the same thing.
The below graph shows retweets across time (by minute) for Donald Trump’s pro-King Salman Tweet. Colours represent the month of account creation for Twitter accounts retweeting the aforementioned tweet. The vertical axis shows the total number of tweets (I removed some in order to make the graph smaller). Generally, the larger the areas of a specific colour, the more bots. What you can see initially (to the left of the graph), is a large amount of Twitter accounts (all created across a wide variety of months – as you’d expect in a normal sample of real accounts), all retweeting the tweet. Yet at around 05.44 you suddenly see a large chunk of orange and pink appearing (apologies if the colours aren’t clear).. This means that unique Twitter accounts, all created around August and September 2017, account for most of the retweeting. In other words, shortly after Trump tweeted, thousands of bots began to retweet the Pro-Saudi tweet.
If we zoom in a little on the the period following 14.55 we can see more clearly how bots account for many of the retweets. In the graph below, broadly speaking, the purple orange and brown regions show a disproportionate amount of accounts that were created in the same month. Most of these accounts are bots.
If we simplify the graph further, and look at the retweeting patterns over hours instead of minutes, we get an even more stark impression of the activity. In the graph below, the first column, showing the initial retweets, highlights a more varied range of accounts created across tens of different months. However, we can see from columns A, B and C labeled below that a large amount of retweets were from accounts created in March, August, July and September 2017. 
If we take August 2017 as an example of month with a lot of account creation, we can have a closer look at the data. As you can see from the screenshot below, many of the accounts were created on the same dates, in this case August 2017. Each row represents a unique account. As well as having the same creation date, all have few followers, and in this particular batch, are all from Riyadh. Most tend to Tweet from Twitter Web Client.
When you click on profiles you begin to see that most tend to not interact with others, retweet the same thing as each other, and have generic posts – and do not interact with other people. Below are two screenshots from different accounts mentioned in the data above, created on 11th August 2017. They are both Arabic accounts. Note that their last retweet was the Trump tweet, and the one before that was an identical IAAF one about voting for Elijah Manangoi.
Twitter’s web browser client is often used by those people creating bots. The below graph showing retweets across time (by hour) for Donald Trump’s pro-King Salman Tweet. However, it only shows those users using Twitter Web Client. Colours represent the month of account creation for each unique Twitter account. As you can see for example, the three largest columns (A, B and C) contain around 6055 unique accounts, most of which are bots.
The below graph shows the account creation date by month of all those tweets in the sample using Twitter Web Client (The time on the horizontal axis here represents account date creation, not time of tweet). Notice how there are massive spikes in account creation. Each column shows a month, but the colours represent individual days within that month. As an example, on one day alone in September 2017, approximately 833 bots accounts were created. What is striking about the graph too is that most of the accounts seem new. This may be due to the fact Twitter will sooner or later cull some bot accounts, and so they are constantly being created and destroyed.
That’s not to say all bots use Twitter Web Client. Increasingly, many are able to tweet or retweet from iphone or android. This is not unusual per se, but it has been said that it has been harder to do this. You can see, for example, in the below graph, that 127 android accounts were created on November 7th 2017.
In the below screenshot, many of the accounts are bots. A number on this particular network have names in cyrillic and English. If you click on their profiles, they’re mostly retweeting various adverts.
Although it’s perhaps useful to estimate how many of the retweets are from bot accounts, it’s hard to know for sure. We do know that thousands of them are though. What the time analysis shows is that the majority of bots not programmed to automatically retweet Trump perhaps kick in after a certain point, thus it would be better to have the entire sample of retweets to determine the total number of bots.
Yet from the below graph, which shows account creation date over time, with colours representing day of the months and columns representing month, look at how many of the accounts in the entire sample were created recently, and how many of them were created on the same day! Indeed, this pattern of graph seen on trending topics is quite suggestive of manipulation by bots. While a spike the right of the graph can indicate a general increase in the popularity of Twitter over time, such a surge in accounts created on similar dates is obviously suspicious.
As a point of comparison, I created a similar graph for people tweeting about Nigella Lawson (hello cooking fans) the other week in the UK. Obviously context is important, as Twitter takeup varies across region, but i thought as an example for future examinations about trend curves and how they may indicate bot activity.
Conclusion
Nigella aside, the purpose of this post has been to highlight how pro-Saudi bots have megaphoned Trump’s tweet to make it seem more popular than it actually is. I am not quite sure what the value of doing this is, apart from for perhaps gaming trend, or maybe a thank you from Saudi bot overlords. Either way, Trump’s message has undoubtedly been amplified by an entity paying (presumably) for this service.
Either way, I like to think that Trump cares about how many people retweet him. If by chance he brings it up, ‘oh hai, yeah my last tweet about King Salman got 82000 Tweets’, then simply retort by saying , ‘yeah but, like your electorate, most of them do not exist’.
Ok that last bit isn’t true, but Twitter definitely has a bot problem. Case in point, they suspended my Twitter account two weeks ago because they thought I was a bot. Go figure….
Marc Owen Jones 
Leave a comment