[Thread]1/ As mentioned before, but delayed slightly due to food poisoning 🤢, what follows is an analysis of the hashtag "Strong Turkiye" which began trending as a response to the hashtag 'help turkey' on 2nd August. As expected, it shows signs of manipulation and co-ordinated 
2/ activity. For some context, the hashtag help Turkey, which I analysed previously, sprung up as a response to wildfires in Turkey. The hashtag prompted an angry response from the Turkish authorities, & the hashtags "we don't need help" & "strong Turkey" emerged. #disinformation
3/ This sample includes around 20,000 interactions involving 9100 unique accounts extracted on 2nd August. First up, signs of deliberate signal boosting of a hashtag by generating high volumes of retweet and mentions. If you look below, you will see three clusters/communities
4/ that are densely connected. The nodes are also large due the fact most are Retweeting the hashtag. The blue (as opposed to purple edges) indicated mentions and replies as opposed to retweets. In case you wonder what that looks like on Twitter, see the tweet on the right>
5/ The purpose of such intense tagging is usually to promote a hashtag at speed and to generate engagements. We see this in low quality campaigns and more often than not, influence campaigns. At the very least, it could be called spam... The fact these are retweeted also
6/ Can also increase the salience of tweets (and thus propaganda) in Twitter's 'top tweets' section.
7/ Another interesting aspect of this is that the fourth most active tweeter on the hashtag within the sample has also changed his account name, Edip Ekmen has become Ahmet Baran Ekmen. This was done soon after RT'ing loads of tweets with 'Help Turkey'
8/ There is also a large number of account churn, - accounts changing their names - likely to avoid detection. Some of those communities seem to exhibit patterns of name changing that suggest co-ordination. For example see below > The following names either add an underscore
9/ or an additional letter. A number of these are well followed accounts, in the tens of thousands. This one for example with Erdogan in the banner and links to a barely used isnta page, the Twitter profile is an actual photograph -although who knows if it corresponds with
10/ whoever operates the account. Another member of this sub community asil545asil545 has also been suspended. Given the speed of the suspension (and others), it possible it triggered one of Twitter's automated suspension algorithms.
11/ Of course then you get your usual entourage of marketing bots that either automatically retweet big trends or are part of sockpuppet/marketing firms that deploy to boost trends for clients. See for example - marketingbabaji - which was then suspended
12/ Around 3% of the account in the sample analysed have either changed their name, been suspended, or been deleted. Some have also chanted their handle at least twice in a few days. A number of people in this category also fall into the same sub group as those doing minorchanges
13/ to their account names. Anyway. That's it for now. A few points and caveats. Firstly, as with all threads on manipulation, many/most/majority/some of those involved are real people. As with any trend of this scale, manipulation is likely to be high. Generalizing the
14/ the trend we see here the number of suspect accounts here probably runs into the thousands. This brings me to another point. This should not be surprising! As I mentioned before, one study puts almost half of Turkey's trends as 'fake'. Furthermore actu.epfl.ch/news/mass-scal…
15/ this manipulation includes multiple types of trends, from political to commercial, including pro AKP and anti AKP (though the swiss study found more pro AKP accounts) Influence campaigns do not discriminate in this regard.
16/ Also remember that while 'fake' or 'sock' puppet accounts are easier to define, some things are more tricky. An influence operation, for example, is a concerted attempt to influence, and may involve people highly motivated an ideologically aligned (nationalism) for example
17/ Such campaigns may or may not include inauthentic activity (such as tweeting from multiple accounts or fake accounts). Such behavior would constitute deception. Propaganda campaigns are even more difficult to define, but most heavily nationalist campaigns would almost
18/ certainly fall into that category. Anyway, if we acknowledge that manipulation is the norm on Twitter, and significantly so on Turkish Twitter, the question should be how much is a specific hashtag manipulated, rather than is it manipulated at all?
19/ Another problem we have in terms of attribution is the problem of adversarial politics. We know disinformation outfits now play both side. For example, Russia's Internet Agency was found to run both pro BLM accounts and pro -MAGA accounts. Why? Because the point is to
20/ drive polarisation in society. (I am not saying this is what is happening but it is one of the many things that is probably happening to some degree). So it's no good being triumphant finding out your political opponent is probably using disinformation, because you
22/ Apologies for the extended conceptual discussion, but I'm aware some readers really could have benefited from more discussion. Anyway, I hope this thread will not be taken in bad faith or manipulated, as the last one was by some parties.
23/ I should add doing *timely* disinformation research is tricky. It is important to acknowledge that just the Turkish government has had Twitter accounts suspended for pro-government propaganda, it also a prime target of regime-change propaganda, especially from the Gulf atm
24/ It is important to document potential misinformation and disinformation without getting caught up in the partisan nature of it. Soc media disinfo respects no borders, and is endemic. Investigating a hashtag as Turkey are doing is an absurd and dangerous move, but I
25/ imagine we will see more of it if unless social media companies implement betters means of tackling
manipulation/gaming/influence ops so easily.
manipulation/gaming/influence ops so easily.
26/ Ok that's it for now. Let me know if you have any questions, I have many for myself as detecting influence operations and manipulation is an emerging and ever-changing field. I hope at least there is food for thought here
• • •
Missing some Tweet in this thread? You can try to
force a refresh
