Yesterday Twitter suspended 1800 sectarian bot accounts after me and a Bahrain Watch colleague contacted them with the results of an investigation conducted yesterday (You can read that here). A brief summary; the de-nationalisation of Isa Qasim prompted lots of activity from suspicious looking Twitter accounts that were all Tweeting the same thing, essentially justifying the move against Qassim. The accounts, which have been active since 2013, have tweeted thousands (probably millions) of Tweets containing anti-Shia and anti-Iranian content over the course of their existence. This, naturally, suggests a government, institution, individuals, or group of individuals, are using marketing techniques to promote sectarianism.
Today, I wanted to see if the 1800 that Twitter suspended was just the tip of the proverbial iceberg, and I have a feeling it is. Using similar techniques to yesterday, I looked on the Bahrain hashtag. As is usual, there were many replica tweets from different accounts. That is to say, tweets that are not RTs, but what look like copy and paste jobs.I then requested Tweets from the Twitter API under the #Bahrain hashtag. This returned 10887 Tweets from today (download the data here). The timestamp of the first tweet was 3,38,54 and the last 15,20,55. This means that these 10887 Tweets were from a 12 hour period (approx) on 22nd June 2016. That’s a lot of tweets.
Three repeating tweets among the 10887 bore the classic hallmarks of the suspicious accounts yesterday (i.e. they were created around the same time, they were all launched from Tweet Deck, they all recycled the same tweets a number of times, they had similar numbers of followers and followed, their bio was either absent or generic, and those created on certain periods (2014,2015) had no banner photo).
The results were quite staggering. If you look at the table below, you will see that 51% (5556) of Tweets on the Bahrain hashtag are most likely produced by bots with a sectarian, hate-inciting agenda. And this is the day after Twitter suspended 1800 accounts.
||Number of Tweets (Out of 10887)
||Persians and the Majus (Zoroastrians) hate the Arabs
||الفرس والمجوس ……حقد على العرب
||The Iranian regime are the leaders in war and the manufacturing of terrorism and crises #withdrawalofnationalityfromIsaQasim #Bahrain #Fitna
||النظام الإيراني رأس حربة في صناعة الإرهاب و الأزمات https://t.co/ZcOaOUvn1T … #سحب_جنسية_عيسى_قاسم #البحرين #Bahrain #الفقيه #القائد
||Iran’s Mullah’s politicise the Hajj (piligrimage) with slogans outside Islam and the Sunna of the Prophet #withdrawalofnationalityfromIsaQasim #Bahrain #Fitna
||دولة ملالي #ايران تسيس الحج بشعارات خارجة عن الإسلام و السنة النبوي https://t.co/0HCOKhuJf6 … #البحرين #Bahrain #آية_الله_الشيخ_عيسى_قاسم
||Sowsan Sha’ir*: Bahrain will not listen to the threats of Qasim Suleimani**, or others like him.
*Sowsan Sha’ir is a pro-status quo Bahriani Columnist
**QS is an Iranian general who recently said that the move against Isa Qasim would cause a fire in the region
|سوسن الشاعر:البحرين لا تعبأ بتهديدات قاسم سليمانى ولا غيره https://t.co/zkHjKY0u9O … #سحب_جنسية_عيسى_قاسم #البحرين #Bahrain #الفقيه #القائد
5556/10886 x 100 = 51% (2sf) of total Tweets are by bots.
A Closer Analysis
Although the bots cycle a number of different tweets, I wanted to examine some more accounts further. I settled on the Tweet الفرس والمجوس …حقد على العرب . Translated, this means ‘Persians and the Majus hate the Arabs’. Because of the term Majus, a slur used to mean ‘zoroastrians’ ((used to as derogatory term for Iranians), I thought it would be an appropriate example of a sectarian Tweet to explore. It was also the most commonly occurring Tweet in the above data.
Focusing on one Tweet, I searched the Twitter API, and it brought back 7500 tweets, from about 1387 different accounts (you can download the date here). That’s a lot of tweets,especially considering that they are just from today (22 June 2016). What’s more, they are all the same tweet, so considering each bot churns out a cycle of Tweets (at different rates as discussed yesterday), they are, when notably active, producing tens, probably hundreds of divisive and hateful tweets per day each. ( You may notice that the 7500 number is higher than the total number mentioned earlier from today (5556), which includes the same Tweet (Persians and Majus hate Arabs etc). This probably reflects the fact that the Twitter API search only brings back the latest tweets. Either way, it means that there are more sectarian Tweets out there per day than accounted for even in the large dataset used earlier).
It is important to note that once an account has been suspended, Tweets can no longer be extracted from the API. On this basis, (which I tested by seeing if I could access some of the Tweets/accounts of those I mentioned yesterday ((I couldn’t)), we can assume that accounts whose data shows up in this analysis are still active, and have not been suspended by Twitter. Indeed, I randomly tested at least 100 profiles in this dataset of 1387 accounts, and all those tested were still active accounts. Of these accounts, only 10 were probably legitimate accounts. However, while this a small number, some of them had retweeted the propaganda from the bot accounts, implying that there is some notable impact of this disinformation on real people.
Below are some more patterns that suggest the accounts are bots or automated in some way.
- Like Yesterday, all the accounts were frequently created on consecutive days within certain months. Some of the accounts were created in the same timeframe as those yesterday (i.e. June, July, August 2014 and February 2016). Today’s examination (which was much larger), threw up some new, and earlier, time frames. For example, approximately 101 accounts were created on either the 26th, 27th, 28th, or 29th August 2013. Like yesterday a substantial amount (around 200) were created on 6, 8th, 9th, 10, and 11th June 2014. (I checked to see if date creation matched any particular working week but it did not – so no lead there).
- All accounts tweet in a specific number of bursts. Most of those analysed tweeted the same tweet six or nine times. Naturally given the time frame, we may have cut off some accounts before they had a chance to complete their full Tweet cycle. See Fig 1 below for an example of the pattern of six tweets.
3) There are varying, but consistent patterns of Tweets. For example, if you look at the times below for the account @aliielsa3a75, you will notice how the account sends bursts of two tweets four seconds apart, three times.
A number of the other accounts follow a similar pattern, 6 identical tweets, spaced out in regular intervals. Below you will notice that the account @alielqany125 sends a burst of six tweets, 4 or 5 seconds apart. However, the account @aliesbashir sends six tweets in two sets of three. Within each set the tweets are also spaced out by four or five seconds.
This four or five second permutation is very common. Other accounts also engage in patterns, although they are not always the same as this. However, I am not going to go into all the patterns, safe to say, they exist within users, and vary to an extent across users.
4) Like yesterday, the only accounts with biographies are those created in 2016. All those created before 2016 do not have a bio. Similarly, only those created in 2016 have banner images. As with yesterday, all the bios are generic, mentioning religious idioms or generic Arabic pleasantries. The most common word was ‘Allah’ which appeared 816 times in the Bios. Presumably this is to convey a sense of pious zeal, or an image of a ‘good muslim’. As with yesterday, I’ve done a word cloud….
Word cloud of fake bios. So meta…
5) As with those accounts examined yesterday, different accounts created on the same day have a damningly similar number of Tweets. E.g. @, @ , , , all created on 19th February 2016, have 1342, 1350, 1366, 1335, and 1352 respectively. To take another random example, accounts , , and , all created on 4th March 2015, have 872,863, 839, and 848 Tweets respectively. I won’t go on, you get the picture.
6) Although I did not mention it yesterday, the profile pictures and images tend to be rather generic, usually consisting of slightly pixelated traditionally dressed Gulfi men, or children (not unusual in the Middle East for a DP). My favourite example is this elderly, yet obviously technologically savvy gentleman, and an orange. Why the orange?
Considering the above, it seems fairly clear that large numbers of bot accounts are still polluting the #Bahrain hashtag with sectarian content. They are also polluting other hashtags, so long as they relate to Iranian expansionism or Shia subversion. Unfortunately, this sectarian content, which amounts to hate speech, has amounted to approximately 51% of the Tweets on Bahrain hashtag today. For those familiar with the Bahrain hashtag, this appears to be a common occurrence, and it makes getting any actual, genuine news, more difficult. Not only does this damage the utility of Twitter, but it’s impact on regional tensions could also be hugely problematic. It’s hard to know the causal implications of this, safe to say it distorts the information sphere around the Arab Gulf in favour of what seems like Wahabi orthodoxy. So it’s not only Twitter who need to take note (I am grateful for their banning of 1800 bots), but also policy makers who worry about Daesh, or the general escalation of conflict in the region.
As a final note, these snapshots I have provided do not necessarily reveal the full scale of what’s going. It is still hard to determine how many of these accounts exist.