Computational Propaganda is Swaying North Macedonia’s Politics
North Macedonia’s 2020 parliamentary election unfolded under peculiar circumstances. Scheduled for April 2020, shortly after the country was denied a start to EU accession negotiations, and following entry into NATO, the election was postponed to July 2020 as a result of the COVID-19 pandemic.
During this prolonged pre-electoral period, citizens were offered one of two political futures: continuation on the path of Euro-Atlantic integration under a Social Democrat-led coalition, based on compromises over national identity; or preservation of socially conservative traditional values, and opposition to the change of the country’s name, under a VMRO-DPMNE-led coalition.
With these circumstances in the backdrop, and with the 2018 change to the country’s name still fresh in the minds of voters, the election represented a significant and divisive political event.
In a study supported by the Prague Security Studies Institute, we examined whether local or foreign actors had made use of this polarized environment to advance their own goals using computational propaganda methods on Twitter.
The relevance of the study is based on the events surrounding the 2018 referendum on the change to the country’s name.
Research conducted by the Atlantic Council’s Digital Forensics Research Lab showed that, during this time, automated accounts on Twitter significantly shaped discussions on the referendum through the “Bojkotiram” (“I am boycotting”) campaign, which attempted to discredit the referendum.
To examine whether a similar network shaped online discussion in the period surrounding the election, we collected all replies to posts from, or mentions of, the Twitter accounts of 26 political figures and media outlets for the period of February to August 2020. The dataset contains 51,969 unique posts and replies from 5,646 unique users.
The study combined a number of existing methods for botnet identification. These included looking for: accounts with activity rates that exceed normal human behaviour; a large number of accounts created in a short period; repetitive naming patterns.
Using these methods, the study identified a large network of users with similar characteristics, many of them created in a short period prior to the election.
Most of these accounts focused on vilifying the Social Democrats, SDSM, and Western officials, or amplifying posts from VMRO-DPMNE and Levica [“the Left party”], while also expressing opposition to the country’s name change and Euro-Atlantic integration.
Semi-automated accounts set the agenda
An illustration of an account belonging to this network was @burdush_gv. In the six-month period examined, this account had 801 interactions with the accounts of political and media figures. Created in 2013, it had posted 180,900 times by August 2020, an average of 55 tweets per day.
This high activity rate suggests that the account’s activity is at least partially automated. Most of the content it shared amplified messages from VMRO-DPMNE, criticized the Social Democrats and promoted anti-Western or conspiratorial views.
Others among the most active accounts identified in the study had also been key instigators of the “Bojkotiram” campaign.
These semi-automated accounts effectively set the agenda for political discussions on Twitter in this period. Almost none of the accounts investigated were supportive of the Social Democrats or in favour of the name change.
Examining the creation dates of all of the accounts that interacted with the 26 selected political figures and media outlets, we found that the largest increase in new accounts occurred in the three months leading up to the initial election date of April 2020, when nearly 500 new accounts were created.
By comparison, the number of new accounts created in every quarter over the last 10 years was substantially lower. As there is no reason to assume that the number of politically engaged Twitter users in North Macedonia surged organically in this period, we hypothesized that many of these accounts were artificially created for election-related goals.
To confirm that this increase was a result of an artificial injection of accounts, we relied on insights from past research. These show that when creating a large number of fake accounts, those responsible often rely on some degree of automation in the account naming process.
Many users – controlled from a single source
This approach then allows for the detection of large groups of users controlled by a single source through an examination of repetitive account naming patterns. A review of the usernames in the data set suggested two naming patterns worth investigating. The first is based on long arbitrary strings consisting of at least 1 number (such as o8zOS0lCcThHno0), while the second is based on “generic” names, followed by a set of eight arbitrary digits (such as Makedon27584769).
We found 808 accounts that matched one of the two naming patterns. Many were created in the three-month period leading up to the original election date; they made up just over one-third of the nearly 500 accounts created during this time. These accounts typically offer little information about the true identity of the users and their profile pictures and bios are based on political content.
These accounts are largely sympathetic to VMRO-DPMNE, Levica and the “Bojkotiram” network, while opposing the name change and North Macedonia’s Euro-Atlantic integration. They mostly share tweets originally sourced from some of the most active accounts identified in the study, such as @burdush_gv, as well as @aps_aana and the “Apsaana” hashtag, current manifestations of the “Bojkotiram” campaign.
Overall, the findings showed that a large number of the accounts interacting with the accounts of the selected politicians and media outlets were created prior to the original election date in April 2020. A substantial number of these accounts matched one of the two identified naming patterns, likely indicating an artificial injection of fake accounts. Their activity focused on anti-Western or anti-Social Democrat rhetoric, as well as amplification of VMRO-DPMNE and Levica officials’ views.
The findings suggest that many of these accounts belong to a single network with a clear ideological affiliation and with a reliance on computational propaganda approaches, such as numerous anonymous accounts managed by a single source.
While the network is likely managed by local actors, the study shows that there are ample conditions for future foreign computational propaganda campaigns, both in terms of technical know-how and in terms of existing networks with followers sympathetic to anti-Western discourses.
Given the uncertainty over North Macedonia’s Euro-Atlantic integration process, currently impeded by Bulgaria, computational propaganda methods will continue to represent a tangible threat in both the country and the region.