TrollTracker: Following the footsteps of online Trolls
Polarising deleted tweets from previously identified Twitter trolls that continued to participate in political campaigns in Kenya.
By Code for Africa — iLAB
Executive summary: What’s the story
This investigation leveraged information from TrollTracker to identify continued engagement in Twitter campaigns by previously identified Twitter trolls (accounts that post inflammatory, digressive or abusive messages in online forums). These accounts were monitored over a course of four months.
The TrollTracker platform, monitoring a total of 1,731 trolls (also referred to as actors of interest) from previous research done by disinformation researchers across the world, accumulated a total of 20,564 tweets, of which 6,384 were from actors monitored and affiliated with Kenyan political conversations.
An analysis of the most used hashtags from the deleted posts showed that the accounts continued to engage in potentially paid advertisement and political campaigns. The majority of the hashtags were politically motivated and included:
- #ConNickMwendwa (a hashtag used to targeted Nick Mwendwa, the former president of the Football Kenya Federation) ,
- #AzimioLaUmoja [Declaration for Unity] (a hashtag that mirrored former prime minister and presidential aspirant, Raila Odinga on his national convention dubbed ‘the Azimio la Umoja national convention’),
- #INAWEZEKANA [It’s Possible] (a hashtag that accompanied the #AzimioLaUmoja hashtag)
- #Tapeli [Conman](a hashtag targeted the deputy president of Kenya, William Ruto, calling him a fraudster or a con),
- #WakikuyuMtafinywa [Kikuyus will be crushed] (a controversial tribal hashtag that targeted the United Democratic Party’s supporter, and a member of parliament, Oscar Sudi)
The investigation further identified that the monitored actors deleted polarising content which was still available as evidence on TrollTracker .
This shows that TrollTracker could be a source of evidence needed by investigators to identify tweets that have been deleted. and would otherwise have been used to share controversial information or hate speech, especially in political settings or conversations.
The background
Kenya’s media landscape is diverse, vibrant and sophisticated, with the freedom of media guaranteed in the Constitution of Kenya, 2010 (Articles 33, 34 and 35). This has often resulted in political discussions being held over social networking sites.
In a study on Kenya 2017 general election, digital advertising techniques amplified the spread of hate speech and disinformation targeting political opponents with one study finding that 9 in ten Kenyans had seen false information about the election online, and 87% of respondents believing that information to be deliberately false.
A previous CfA investigation in Kenya revealed a case of coordination of rival political factions creating sophisticated digital operations that conscripted amplification of two distinct narratives that were identified: one against President Uhuru Kenyatta and another against Deputy President William Ruto.
A separate investigation also shows how a coordinated network of paid Twitter influencers ran a Twitter campaign using the hashtag #RutoPlanningViolence. We compiled the list of identified trolls from the investigations and continued to monitor them for other polarising pieces of content, including deleted content.
TrollTracker
This is a Twitter-based disinformation tracking tool built on a web-based dashboard that collects deleted tweet content from previously identified trolls and disinformation actors. The tool seeks to help monitor the social posts of known disinfo actors. The primary tangible output of the tool is to expose trolls behind toxic disinformation campaigns who routinely cover their tracks by deleting original inflammatory social media posts that sparked hate speech, disinformation campaigns or conspiracy theories.
Read more here: https://medium.com/@AfricanCIR/troll-tracker-tracking-volatile-sources-of-mis-disinformation-keeping-disinformation-agents-90f45e604242
The tool was used to conduct continuous monitoring of trolls and disinformation actors identified in previous investigations. It also helped us keep track of persons-of-interest, including indicators of foreign influence operations and disinformation campaigns.
Currently, the tool monitors a total of 1,731 actors of interest from previous research done by disinformation researchers across the world. As at 18 January, 2021, the tool had collected a total of 20,564 tweets. A total of 6,384 tweets had been deleted by monitored actors from Kenya.
Notably, @_krisdee_, a twitter account that had been identified in a previous investigation showing a coordinated network of paid Twitter influencers, had the highest number of deleted tweets — a total 1,367 representing ~21% of the total number of deleted tweets from Kenyan-linked accounts.
Key findings
Engagement in political hashtag campaigns
CfA further mapped the hashtags that had been used within the deleted tweets to identify the most prominent campaigns the accounts have been part of over the period 01 October 2021 to 18 January, 2022. There were a total of 48 hashtags that had been used more than five times.
We categorised the hashtags into the focus area and noted that the majority of them were politically motivated.
This indicated that the identified trolls continue to participate in political hashtag campaigns on Twitter targeting different political divides, and worse, deleted some of these posts to ensure that polarising content is no longer available on Twitter, which would affect their next campaign focus.
Advertisement hashtags
Our key investigative focus is on political hashtags, but advertising hashtags indicate that the accounts are also engaged by corporate institutions who want to get visibility of their products on Twitter.
Below are the top 10 advertising hashtags used by the accounts.
The hashtag that garnered the highest number of posts was #realmec25y, a campaign by Realme,a smartphone company, which was running a campaign from 10 December, 2021 to 06 January, 2022
We noted that the accounts promote each other’s tweets, which indicates a coordinated network of users that work together on advertising campaigns for different clients.
Political hashtags
Many politicians have realised the power that social media holds for campaigning. As observed in many parts of the world, Twitter is on the frontline as it engages users in political debates and, ultimately, mobilises them for grassroot movements. On Twitter, hashtags work as easily searched labels for users to participate or link to ongoing debates. Hashtags are used consciously by key influencers to frame a political debate as well as to define the vocabulary used.
Political hashtags accounted for more than 47% of the total number of hashtags, indicating the prominence of the involvement of monitored accounts in political campaigns on Twitter. CfA’s recent conversation with Twitter’s integrity team confirmed that Kenya is a key hotspot for development of techniques that game trending algorithms, such as use of advertisements and automated tools to amplify conversations on the platform.
Below are the top 10 advertising hashtags used by the monitored accounts.
As observed above, a number of hashtags targeted presidential aspirants in the upcoming presidential elections. The following is a breakdown of the target for the top five hashtags the network participated in:
#ConNickMwendwa
This hashtag started trending on 1 November, 2021, garnering a total of ~1,600 tweets. The hashtag targeted Nick Mwendwa, the then president of the Football Kenya Federation who had been summoned to respond to questions on claims of spending Ksh 244 million for Harambee Stars’ participation at the Africa Cup of Nations (AFCON) in 2019 and several other corruption allegations.
#AzimioLaUmoja
This hashtag had been used a couple of times before on Twitter but ended up as a trending topic between December 8, and December 11, 2021, garnering a total of ~125,000 tweets. The hashtag mirrored the ongoing national convention by ODM leader Raila Odinga that happened on December 10, 2021 and was dubbed ‘the Azimio la Umoja national convention’ held at the Kasarani Stadium where he declared his interest as a candidate for the upcoming Kenyan presidential elections.
#Tapeli
This hashtag has been used previously on Twitter but ended up as a trending topic between December 8 and December 11, 2021, garnering a total of ~2,050 tweets. The hashtag targeted Ruto — calling him a fraudster or a con.
#INAWEZEKANA
This hashtag has been used previously on Twitter but ended up as a trending topic between December 8 and December 11, 2021, garnering a total of ~125,000 tweets. The hashtag that accompanied the #AzimioLaUmoja hashtag, which was in support of presidential candidate Raila Odina.
#WakikuyuMtafinywa
This hashtag started trending December 16, 2021, garnering a total of ~5,680 tweets. The hashtag targeted the United Democratic Party’s supporter, and a member of parliament, Oscar Sudi.
The hashtag was started by the account @_Lastbron claiming that Oscar Sudi had threatened the Kikuyu tribe in Kenya at one of the political campaign rallies. The claim had already been fact-checked by The Star, a digital media outlet in Kenya.
The hashtag did not garner high levels of interactions on the platform but was identified to be polarising and several Twitter users reported the trend to Twitter support.
Deletion of polarising content
Monitoring disinformation on social media relies on detecting crucial evidence in the form of tweets or posts on platforms like Twitter and Facebook and referencing such content on evidence-based reports. This enables researchers to track and attribute false content and hate speech or propaganda to different online personas and ultimately track the real account owners.
However,some usually engage in disinformation campaigns, then try to cover their tracks by deleting their original inflammatory social media posts or video content once the content/ campaign begins to gain organic momentum. In some cases, social media platforms often delete these posts to slow down the spread of harmful content; as such, crucial evidence that investigators need to understand who is behind the content or campaign disappears. It has become quite evident that digital evidence is volatile and fragile and can easily be altered in cases of improper handling. New challenges are emerging for investigative journalists and researchers to provide evidence-backed analysis when the actors delete posts that may be required to combat or counter false information.
CfA noted that there were several polarising deleted posts on TrollTracker using the hashtags identified in the previous section of the report.
A deleted tweet from user @leerah_ using the hashtag #Rutoplanningviolence was deleted just four minutes after the user posted it. The tweet claimed that William Ruto, Kenya’s current deputy president and a presidential aspirant in the upcoming election in August 2022, was a terrorist and thief. Such content could potentially polarise citizens into conflict.
Another tweet from user @ItsPOS using the hashtag #WakikuyuMtafinywa was deleted three hours after being posted. The tweet claimed that Oscar Sudi, an MP aligned to William Ruto’s political party United Democratic Alliance (UDA), had made claims that the only reason there would be conflict in Kenya would be when there is an issue between Kenya’s two largest tribes, Kikuyu and Kalenjin. The tweet adds that Sudi said the rest of the tribes can be controlled in case of conflict. This tweet has the potential of polarising citizens from different tribes, especially ahead of elections.
A tweet from user @MtotoSumbua said that Kenya’s third president, Uhuru Kenyatta, and former prime minister and a presidential aspirant, Raila Odinga were criminals who haven’t been jailed for crimes committed in the past. This tweet also has the potential of polarising citizens who support the two political figures in the country.
Conclusion
This investigation leveraged information from TrollTracker to identify continued engagement in Twitter campaigns by previously identified Twitter trolls. The investigation further identified that these actors delete polarised content in an effort to mask their involvement in both advertising and political narratives on the platform. This shows that Trolltracker can be a source of evidence needed by investigators to identify tweets that have been deleted and would otherwise have been used to share controversial information or hate speech, especially in political settings or conversations.
By ANCIR iLAB investigative team.
Code for Africa (CfA) is the continent’s largest network of non-profit independent civic technology and open data laboratories, with a multidisciplinary team of more than 80 full-time technologists and analysts in 18 African countries. CfA’s laboratories build digital democracy solutions that give citizens unfettered access to actionable information to improve their ability to make informed decisions, and to strengthen civic engagement for improved public governance and accountability.
The African Network of Centres for Investigative Reporting (ANCIR) is a CfA initiative that brings together the continent’s best investigative newsrooms, ranging from large traditional mainstream media to smaller specialist units. ANCIR member newsrooms investigate crooked politicians, organised crime and big business. The iLAB is ANCIR’s in-house digital forensic team of data scientists and investigative specialists who spearhead investigations that individual newsrooms cannot tackle alone. This includes forensic analysis of suspected digital disinformation campaigns aimed at misleading citizens or triggering social discord; polarisation using hate speech; radicalisation, or other techniques.
The iLAB subscribes to CfA’s guiding principles:
- We show what’s possible. Digital democracy can be expensive. We seek to be a catalyst by lowering the political risk of experimentation by creating successful proofs-of-concept for liberating civic data, for building enabling technologies and for pioneering sustainable revenue models. We also seek to lower the financial costs for technology experimentation by creating and managing ‘shared’ backbone civic technology and by providing resources for rapid innovation.
- We empower citizens. This is central to our theory of change. Strong democracies rely on engaged citizens who have actionable information and easy-to-use channels for making their will known. We therefore work primarily with citizen organisations and civic watchdogs, including the media. We also support government and social enterprises to develop their capacity to meaningfully respond to and effectively collaborate with citizens.
- We are action oriented. African societies are asymmetric. The balance of power rests with governments and corporate institutions, at the expense of citizens. Citizens are treated as passive recipients of consultation or services. We seek to change this by focusing on actionable data and action-orientated tools that give ‘agency’ to citizens.
- We operate in public. We promote openness in our work and in our partners’ work . All of our digital tools are open source and all our information is open data. We actively encourage documentation, sharing, collaboration, and reuse of both our own tools, programmes, and processes, as well as those of our partners.
- We help build ecosystems. We actively marshal resources to support the growth of a pan-African ecosystem of civic technologists. Whenever possible, we reuse existing tools, standards and platforms, encouraging integration and extension. We operate as a pan-African federation of organisations who are active members of a global community, leveraging each other’s knowledge and resources. All of our work is better if we are all connected.
This report was authored by the iLAB’s team, consisting of investigative manager Allan Cheboi, senior data analyst Robin Kiplangat and data analyst Anita Igbine. The report was copy-edited by Amanda Strydom and approved for publication by CEO Justin Arenstein.