The investigation revealed the influence of foreign, white nationalist groups, pro-RET (Radical Economic Transformation) trolls and bots, and bots for hire from the USA. The findings were based on millions of tweets collected from as far back as the 2014 general elections.

By comparing how many of the authors of the tweets in each data set were later suspended by Twitter for bot-like behaviour, the relative influence of these accounts could be gauged within their respective communities.

Earlier this year, Twitter suspended millions of users as part of a massive clean-up operation.

The micro-blogging platform came under pressure to clean up its user base after details surfaced of an orchestrated campaign by the Russia-based Internet Research Agency during the US election in 2016.

Home-grown harvesting

The analysis spans 28 data sets, collected over four years. These include tweets related to a diverse range of topics, such as Fees Must Fall protests (during 2015, 2016 and 2017), Western Cape Premier Helen Zille’s “legacy of colonialism” tweets, Black Monday protests and the ANC’s 54th elective conference in December 2017.

READ MORE: Twitter responds to ‘maid’ and gardener’ labelled mugs

In each case, a list of keywords would harvest tweets on a specific topic. The tweets were collected either directly from Twitter using its REST (Representational State Transfer) or streaming application programming interfaces (API), or from third-party social listening services.

The largest single data set analysed was the Fees Must Fall protest during 2016, consisting of more than 2,3 million tweets. In some of the bigger data sets, Superlinear made use of randomised samples of the collected tweets for practical purposes.

The data presented several peculiar findings. For example, one specific user brought an entire botnet to bear on Telkom after his service complaints went unanswered.

In another example, the analysis showed that about 70 of every 1 000 users tweeting about #BlackMonday protests were later suspended, although they had very minor impact on their communities.

Most of these suspended users were retweeting content created by foreign white nationalists, while the local suspended users were what was described as a “pro-EFF” grouping.

Groups, mobs and cabals

The analysis provides some insight into which topics are particularly susceptible to interference, or manipulation by Twitter accounts.

READ MORE: Twitter celebrates first academic to complete PhD in isiXhosa

These topics gravitate towards racially polarised issues (xenophobia, Vicki Momberg, Black Monday protests), elections (the ANC elective conference and the 2014 general elections) and protests (Fees Must Fall and North West riots).

Superlinear’s analysis also identified three main groupings with a high prevalence of suspended users, divided primarily along racial and political ideologies.

These consist of:

  • pro-RET: a group of pro-black, pro-radical economic transformation accounts focusing on ANC factionalism, anti-DA topics and racial divisions;
  • pro-EFF: a group of pro-black, pro-EFF and anti-ANC actors focusing on white racism; and
  • pro-Far Right: a group of mostly international, pro-white accounts focusing on white fears to create content for their own foreign communities.

These communities would surface and engage as topics that affected them were introduced.

Gearing up for 2019

The analysis concludes with both a consolation and a warning.

READ MORE: Twitter schools Kanye following White House rant

“Finally, this investigation shows that there is definitely interference in South African politics on Twitter, both by local and international actors. However, the interference does not yet appear to be on the scale experienced in some countries.”

“However, if you take into account South Africa’s smaller Twitter population compared to the US, where 2 848 known sockpuppet accounts had some impact on their politics, the number of suspended accounts (which includes trolls, sockpuppets and bots) present in the ANC54, Helen Zille and Black Monday data sets start to loom larger.

– News24