Who Are They?

The 200 accounts shown above are a sample of a network on Twitter talking about Voter Fraud and amplifying false and/or misleading narratives about election integrity and the democratic process. We discovered that this group of 200 accounts either generated or were mentioned in over 140 million tweets over the last year. As you will see below, this network is not only growing at an accelerating rate but also coordinating with effective tactics that appear to bypass many of the detection methods of existing disinformation research.

As you read through the rest of this story and the subsequent report, you'll probably be left with more questions than answers. We certainly are. You might even be in awe of these networks. We can relate to that too. Some days the size, scale, and effectiveness of these modern tactics to influence conversation have fueled our curiosity. On other days, however, we're left angry, sad, and frustrated at the content these accounts push, and how we’ve all helped create an environment that allows people to weaponize participation and wield influence over civic dialogue so effectively.

We are a volunteer team of researchers, technologists, and artists that started this project to explore the conversation about Voter Fraud in US politics on Twitter. We became interested in this topic because it sits at the intersection of the VoterID and Voter Suppression conversation, and while instances of Voter Fraud are statistically infrequent it is the subject of considerable debate online. We wanted to know if there was a consistent conversation happening, was it happening on Twitter, and was there something behind the charged nature of the dialogue that we should be concerned about. Here is what we're not gonna say:

We are also not claiming that there have been no documented cases of Voter Fraud. We are wondering if the reality warrants the intensity and urgency of stories that we see, or if the narratives about Voter Fraud are in fact undermining the Democratic ideals they claim to be protecting. In a brief titled Debunking the Voter Fraud Myth the Brennan center used phrases like "vanishingly rare" and "nearly non-existent" to describe the results of research looking at documented cases of Voter Fraud on US elections. If that research is thorough and accurate, then along with other research we've seen on this issue it was clear that many of the narratives related to Voter Fraud seem to at the very least be overreactions, and at worst some kind of propaganda, demagogic messaging, and/or a strategy to distract people from real issues related to election integrity.

Our hope is that by presenting our work in this format, we can discuss what influence looks like, and investigate the roles we all play and the way coordination is being used against all of us online, right now. While we don't know who these people are or why they're doing this, we do know that they're effective, influential, and coordinated in some way.

We want to know more.

Ok, What Happened?

1. Tracking #VoterFraud

This investigation began when we found something strange. When we looked for use of the hashtag #VoterFraud on Twitter over the last twelve months, we saw this:

Usage of #VoterFraud hashtag in 2018

2018-01 2018-02 2018-03 2018-04 2018-05 2018-06 2018-07 2018-08 2018-09 2018-10 2018-11 0,000 0,500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 5,500 6,000 6,500 7,000 7,500 Posts containing #VoterFraud hashtag

This didn’t seem normal. In our experience, spikes in conversation around a topic sometimes have a fairly consistent pattern, but rarely at a frequency this consistent. As we went further back in time in the data, the pattern of spiking just kept going, almost like a heartbeat in an EKG machine. Then we looked at the last three years:

Usage of #VoterFraud hashtag since 2016

2016-01 2016-04 2016-07 2016-10 2017-01 2017-04 2017-07 2017-10 2018-01 2018-04 2018-07 2018-10 00,000 10,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000 90,000 Posts containing #VoterFraud hashtag

Total mentions of #VoterFraud: 1429718

This heartbeat had been going for a while. We asked ourselves, who’s involved in these spikes? Could these spikes just happen around real news items related to Voter Fraud or is something else going on? From here we expanded our investigation beyond the ‘What’ to the ‘Who.’

2. Who Are They?

We dug into activity around the days where consistent spikes of #VoterFraud were happening. We hoped that looking into the spikes would give us insight into who was behind these mentions and why they were happening. So, we made a list of users that mentioned #VoterFraud on the same day when there was a spike of 5,000-7,500k mentions. The first thing we noticed when comparing the spikes is that a lot of the same faces kept appearing over and over again, talking about or retweeting Voter Fraud stories on the same day, often in a similar way.

Usage of #VoterFraud hashtag around 2018-08-10

07-27 08-03 08-10 08-17 08-24 0,000 0,500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 5,500 6,000 Posts containing #VoterFraud hashtag

Accounts from the sample 200 participating in the 2018-08-10 spike

In this field of research, when a group of accounts tweet similar messages at the same time, on multiple occasions, it's called co-spiking. Co-spiking can happen organically when groups of people share interests and sources that they follow.

Many of these accounts were connected to each in a number of ways, like:

  • following each other
  • mention/retweeting each other often
  • profile bios with similar aesthetic patterns (many of the same emojis and hashtags)
  • talking about a lot of the same things at the same time besides #VoterFraud (Qanon conspiracies, immigration and deportation, and ‘false flag operations’)

There was something uncanny about these accounts. Were they all bots? We kept digging for answers. We clicked on all of the accounts, went to personal or business sites that they linked to in their Twitter bios, did reverse image searches on profile pictures, looked at linked Instagram accounts, anything we could find. We were able to verify that some accounts might, in fact, be human. But that didn't really explain what was going on here. Were they next generation bot accounts cleverly disguised? Maybe the ‘bot or not’ question didn't even matter? We started to feel like confusion and distraction were the point. So, instead of trying to figure out who they were, we tried analyzing their activity from another angle.

3. Seeing The Surge Pattern

Beyond Voter Fraud, we wanted to understand the full history of these accounts, when did they sign up? How often do they tweet? How often do they get mentioned? What other things do they talk about? Something stood out when we combined the data for "Tweets from" these accounts and @mentions (other accounts mentioning or replying to the sample of 200) over a three-year timeframe.

Posts by and mentions of @girl4_trump

2016-07 2016-10 2017-01 2017-04 2017-07 2017-10 2018-01 2018-04 2018-07 2018-10 00,000 01,000 02,000 03,000 04,000 05,000 06,000 07,000 08,000 09,000 10,000 11,000 12,000 13,000 Posts by @girl4_trump Mentions of @girl4_trump

Accounts from the sample 200 mentioning @girl4_trump

We call this the surge pattern, and we saw it over and over again among the 200 sample accounts tweeting about Voter Fraud.

What’s happening here? The gray line is tweets from @girl4_trump and the black line is other accounts mentioning or replying to @girl4_trump. The account @girl4_trump tweeted consistently over the last two years and received very few mentions until January 2018. Between January and April 2018 @girl4_trump went from being mentioned around 60 times a day to being mentioned over 8000 times a day. 8000 mentions a day is something you might expect if a famous or semi-famous person just recently joined Twitter, or a regular person gets caught up in a major news story, but that wasn’t the case here.

Here are more examples from the sample list of 200 accounts.

Posts by and mentions of @battleofever

2017-10 2017-12 2018-02 2018-04 2018-06 2018-08 2018-10 0,000 0,500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 5,500 6,000 6,500 7,000 Posts by @battleofever Mentions of @battleofever

Accounts from the sample 200 mentioning @battleofever

Posts by and mentions of @bbusa617

2016-10 2017-01 2017-04 2017-07 2017-10 2018-01 2018-04 2018-07 2018-10 00,000 02,000 04,000 06,000 08,000 10,000 12,000 14,000 16,000 18,000 20,000 22,000 24,000 26,000 Posts by @bbusa617 Mentions of @bbusa617

Accounts from the sample 200 mentioning @bbusa617

Not all of these accounts surge at the same time, but most had a sudden, and dramatic surge from little to no activity to getting thousands or tens of thousands of mentions in a day. Some of them had a surge moment 6 months ago, many others began surging 15 days ago.

The surge was especially troublesome to us for a couple of reasons. It represented a dramatic increase in conversations around toxic and divisive narratives, and it’s also much, much more difficult to detect. Most researchers in this space look for botnets and coordination by looking at follower and following relationships or accounts retweeting each other. Mentions and replies are harder to detect and analyze, they're also more effective at hiding coordination and getting people to engage. Were others seeing this evolution in tactics?

4. Replicating the Surge & Growing the Network

We wanted to know how these accounts were coming onto Twitter and gaining mentions at such a high velocity — what was leading accounts to gain influence, so quickly? So we took a sample set of accounts from a group of suspicious Voter Fraud accounts and started looking at their activity day-by-day, starting at day one. What we began to notice is a pattern for how the influence machine might be working, and how coordination could be happening.

Here's the consistent network pattern we saw:

  • User signs up for an account.
  • User starts replying to multiple accounts—some known verified Twitter users and many other accounts that are also on our list of actors, or that fit a similar profile.
  • The replies tend to contain: text, memes, hashtags, and @mentions of other accounts, building on common themes.
  • At some point the pattern shifts from being all replies to original tweets. Those original tweets contain the same types of content as their replies do.
  • It appears that this pattern cycles and repeats when the next batch of new accounts come online. The next batch starts replying to the existing, newly influential accounts, and carry on with the same sequence of events for gaining influence.

Posts by and mentions of @bbusa617 and @pinkk9lover

2015-10 2016-01 2016-04 2016-07 2016-10 2017-01 2017-04 2017-07 2017-10 2018-01 2018-04 2018-07 2018-10 00,000 02,000 04,000 06,000 08,000 10,000 12,000 14,000 16,000 18,000 20,000 22,000 24,000 26,000 Posts by @bbusa617 Mentions of @bbusa617 Posts by @pinkk9lover Mentions of @pinkk9lover

Accounts from the sample 200 mentioning @bbusa617 and @pinkk9lover

The above image tracks @bbusa617 and @pinkk9lover. These two accounts both display a surge in activity, and also mention and retweet each other frequently. This is an example of what we call stacked surging.

5. Three Coordination Patterns

When we put this analysis together, three distinct patterns emerged, which in conjunction with the past and present emphasis on Voter Fraud, indicated that a possible coordinated influence operation is happening right now. Moreover, given the scope and content of the narrative being amplified, it appeared that this network was broadly attempting to undermine the credibility of the electoral process and sabotage the broader civic dialogue around the 2018 mid-term election. The three coordination patterns are:

  • Co-Spiking: We have seen co-spiking of accounts amplifying narratives about Voter Fraud on days when these groups of accounts tweet similar messages about Voter Fraud at the same time, on multiple occasions. This is reflected in the heartbeat pattern. Over the last three years, there have been numerous accounts consistently amplifying these narratives every day. This pattern is ongoing.
  • Surge: Within this list of accounts we've seen numerous examples of "surge" pattern where accounts come online, start replying to other accounts, shift to posting their own content, and quickly gain @mentions in order of magnitude week over week, and month over month.
  • Stacked Surging: Looking at the origin patterns of these accounts and their ongoing behavior, we see an evolutionary network pattern where clusters of accounts mention each other and collaborate around specific hashtags. By tracking this networks use of the Voter Fraud hashtag, who they mention, who mentions them, we've begun to show examples of coordination amongst the accounts amplifying the Voter Fraud narrative.

We don’t know why this activity is occurring, or who is behind it. However, the best we can do is look at the data around what’s actually happening. What we've discovered along the way is that there are overlapping patterns of behavior, demonstrating some form of coordination.

We think it's possible that some of these accounts don't realize that they're coordinating or part of a larger influence network. For example, one of these sample accounts might genuinely care about Voter Fraud. A bad actor, coordinating large numbers of accounts could find this person’s tweets useful, then amplify those tweets through thousands of @mentions and replies.

By focusing on the hard data around coordination, we can better understand how public conservations are being distorted and how it affects society. Whatever your views are on Voter Fraud, these accounts and the accounts that amplify them are rapidly accelerating their activity in the lead-up to Election Day.

6. What else are they talking about?











With this project, we set out to provide a new way for the public to understand how influence works. Inspired by the idea to combine research on the fundamental mechanics of these networks, we wanted to create an experience that can be shared to anyone, no matter the starting point, so that you can explore these accounts on your own. Today a small group of people can wield increasingly more powerful AI, big data, and psychological targeting to influence society, and we feel that it’s a fundamental human right to know who’s influencing you, how it’s happening, and why. The Founders of America understood that Democracy only worked with a well-informed public. How can we be informed if we can’t see the invisible influence shaping our society, public conversations, and the thoughts of our friends and family?

What you can do

SHARE and USE this report if you care about Voting Rights, Election Integrity, and believe that America deserves to have better conversations and laws rooted in data. America has seen a dramatic increase in voterID laws, voter suppression tactics, and tens and sometimes hundreds of thousands of voters kicked off the voting rolls. The justifications for these policies were built on the kind of Voter Fraud narratives amplified by the accounts that we have just shown you. There is a tragically ironic relationship between the perception that large groups of people are voting illegally, and a small group of suspicious accounts on Twitter wielding massive influence to spread disinformation, affecting the public’s understanding of Voter Fraud.

The next batch of election laws and policies amplified by these type of accounts could disenfranchise large groups of Americans, undermining the core pillars of our democratic society. This group of accounts will likely be amplifying Voter Fraud related disinformation up until Election Day and beyond. Americans can’t protect each other’s rights without understanding the influence driving the conversations and laws that affect us.

If you find that you or someone you know is having their voting rights violated in any way, contact the Election Protection Coalition by calling 866-OUR-VOTE or visit their website at 866ourvote.org.

Reach out us to if you’re a researcher working to mitigate disinformation’s effect on Democracy. We hope you can use these new patterns of influence and incorporate them into your own work. We’re open to collaborating on how to help the public, journalists, and pro-democracy groups catch up to the evolving disinformation tactics used by bad actors.

Who We Are

🙏🇺🇸🦅 The project could not have happened without the hard work and brilliant insights of these individuals and organizations — we are so grateful for their contributions to this research that we all felt was important to share.

George King, Michele Graphieti, IV.AI, San Diego Supercomputer Center’s Data Science Hub at the University of California San Diego, and Dave Krafstow.

The primary authors of this report are Zach Verdin, Brett Horvath, and Alicia Serrani.

To go deeper, below you'll find further analysis and multi-dimensional visualizations of VoterFraud networks in "Evolving Influence", and a detailed exploration of our methodology in "The Report."

Sincerely,

Guardians.ai

Evolving Influence

Examples of Evolutionary Clustering

Further analysis of the sample network of 200 accounts and the overall network of all accounts mentioning VoterFraud, and their tweets over the last three years reinforced the characterization of this group of accounts as a network. Statistically, when a group of accounts mention similar hashtags and consistently mention each in fairly consistent time intervals, they can be said to have a strong relationship.

The following Evolutionary Clustering graph shows the network relationships among accounts talking about #VoterFraud. The yellow sphere is @battleofever, an account that is part of our sample list of 200. The level of dots below that are accounts that mention @battleofever, the level below that are accounts that mention the prior level, and so forth.


Powered By Zignal Labs Technology

These are the steps we took to establish these relationships:

1. First, we identified an account from the original list of 200, which was @mentioned by any accounts that have tweeted about #VoterFraud over the last three years. We call this Level-0.

2. Once an account was identified, we looked at the relationship between that account and the broader list of over any accounts tweeting about #VoterFraud in the last three years. Specifically, we grouped users @mentioning the specific account identified in Level-0. We define these accounts as a cluster and call them Level-1 because they have a direct relationship with the account mentioned in Level-0.

3. We then tracked which of the broader pool (any accounts mentioning Voter Fraud) were being @mentioned by Level-1. The list produced some targets who are separated by one intermediary degree, which are blue. Any grey nodes in Level-2 connecting level-1 and level-3 are considered intermediary amplifier nodes. We similarly clustered these accounts based on who they mentioned and identified these as Level-2 accounts, as they were two degrees of separation from the original account.

4. Finally, in Level-3 we show any target account clustered by similarity of mentions. One product of this group is to get similar influencers to our target (yellow) node. The other product is getting a pool of targets mentioned by the same pool separated by either 1 or 2 degrees. (Similarity is defined by running a series of machine learning clustering algorithms, mainly tSNE with the output relationships/weights between nodes being calculated with some GPU matrix math similarity algorithms.)

Here are further examples of this analysis:

@bfraser747


Powered By Zignal Labs Technology

@crimsonfaith88


Powered By Zignal Labs Technology

@lindasuhler


Powered By Zignal Labs Technology

@richardtburnett


Powered By Zignal Labs Technology

The identification of clustering among not only the 200 sample accounts but also the broader group of any account mentioning Voter Fraud in the last three years, shows the scope of the dialogue around VoterFraud. It also suggests that since the 200 accounts are strongly linked to the larger set of all VoterFraud related accounts it is likely that the patterns we've identified as related to the 200 accounts are representative of patterns that could be found in the full dataset.

Full Report

1. Tracking #VoterFraud

Usage of #VoterFraud hashtag in 2018

2018-01 2018-02 2018-03 2018-04 2018-05 2018-06 2018-07 2018-08 2018-09 2018-10 2018-11 0,000 0,500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 5,500 6,000 6,500 7,000 7,500 Posts containing #VoterFraud hashtag

Usage of #VoterFraud hashtag since 2016

2016-01 2016-04 2016-07 2016-10 2017-01 2017-04 2017-07 2017-10 2018-01 2018-04 2018-07 2018-10 00,000 10,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000 90,000 Posts containing #VoterFraud hashtag

Usage of all five hashtags in 2018

2018-01 2018-02 2018-03 2018-04 2018-05 2018-06 2018-07 2018-08 2018-09 2018-10 2018-11 00,000 02,000 04,000 06,000 08,000 10,000 12,000 14,000 16,000 18,000 20,000 22,000 24,000 26,000 28,000 Posts containing #VoterFraud hashtag Posts containing any of five hashtags

Usage of all five hashtags since 2016

2016-01 2016-04 2016-07 2016-10 2017-01 2017-04 2017-07 2017-10 2018-01 2018-04 2018-07 2018-10 000,000 010,000 020,000 030,000 040,000 050,000 060,000 070,000 080,000 090,000 100,000 Posts containing #VoterFraud hashtag Posts containing any of five hashtags

Total mentions of #VoterFraud: 1429718
Total mentions of all five hashtags: 2667554

Over the last 12 months, we observed a large network of accounts on Twitter consistently mentioning #VoterFraud. Mentions of #VoterFraud did not remain at a consistently high level, but rather spiked and fell at certain intervals, like a heartbeat. This heartbeat pattern is an example of co-spiking where a count of tweets over a specific time unit (e.g., day) results in a surge of spikes from multiple users within a short time-span (e.g., 3 days). Consistent and repetitive patterns of mentions on Twitter for a sustained period can imply coordination. When looking at mentions of #VoterFraud over the last twelve months, we identified five spikes between 5 - 7.5 thousand mentions over the last year. We expanded the research to determine if we could observe that this co-spiking pattern of #VoterFraud mentions extended back beyond 12 months. Not only did we find consistent spiking around #VoterFraud and the continuation of this pattern over the past three years, but we also found that the most significant spike in mentions of #VoterFraud occurred around the 2016 election. The persistence of the co-spiking pattern over the last three years and consistent mentions of #VoterFraud, coupled with the massive spike around the 2016 election, seemed to further indicate possible coordination.

Expanding our search to include #VoterFraud and adjacent hashtags we observed further examples of the co-spiking pattern, magnified in a larger dataset. We expanded the list of terms for three reasons 1) we wanted to determine if the co-spiking pattern was linked to #VoterFraud, 2) we wanted to see if coordination was focused on amplifying #Voterfraud specifically, or on amplifying a broader narrative and 3) if #VoterFraud was part of a broader narrative, was it amplified with the same level and cadence of coordination. When we added data to include mentions of #VoterFraud, #VoterID, #DemandVoterIDNow, #VoterIDNow, and #ElectionFraud over the last 12 months, we found that the co-spiking pattern with consistent spikes was replicated almost exactly. We found the same results when we expanded the dataset to a three-year timeframe. This finding further indicated the possibility that a coordinated network on Twitter was not only attempting to influence discussion around #VoterFraud but also discussions on election integrity and the democratic process more broadly.

Given the clear overlap in accounts mentioning #VoterFraud versus those involved in the broader conversation that includes mentions of #VoterFraud, #VoterID, #DemandVoterIDNow, #VoterIDNow, and #ElectionFraud, we determined that all ongoing research should track the broader conversation. The rationale being that the patterns that resulted from #VoterFraud versus the broader list were nearly identical and that analysis of the broader list would allow us to track the interplay between these different hashtags as well as provide a broader sample population of accounts promoting this narrative. Thus, for this investigation Voter Fraud refers to #VoterFraud, #VoterID, #DemandVoterIDNow, #VoterIDNow, and #ElectionFraud.

2. Who Are They

Usage of #VoterFraud hashtag around 2018-07-19

07-05 07-12 07-19 07-26 08-02 0,000 0,200 0,400 0,600 0,800 1,000 1,200 1,400 1,600 1,800 2,000 2,200 Posts containing #VoterFraud hashtag

Accounts from the sample 200 participating in the 2018-07-19 spike

Usage of #VoterFraud hashtag around 2018-08-10

07-27 08-03 08-10 08-17 08-24 0,000 0,500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 5,500 6,000 Posts containing #VoterFraud hashtag

Accounts from the sample 200 participating in the 2018-08-10 spike

Usage of #VoterFraud hashtag around 2018-08-28

08-14 08-21 08-28 09-04 09-11 0,000 0,500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 5,500 6,000 6,500 7,000 7,500 Posts containing #VoterFraud hashtag

Accounts from the sample 200 participating in the 2018-08-28 spike

In our investigation of what was behind the amplification of Voter Fraud, we found that there was a group of accounts consistently tweeting and retweeting narratives about #VoterFraud. We also noticed that many of these accounts seemed to be connected because they a) follow each other, b) mention or retweet each other often, and c) talk about many of the same topics. There are also aesthetic similarities between many of these accounts. For instance, this network of accounts uses lots of emojis and similar hashtags in their bios and tweets.

One way that we analyzed the interconnectedness of accounts was by investigating the accounts involved in pushing Voter Fraud over the last 12 months, and specifically focusing on days when there was a spike in the number of mentions. To get a broad sample we looked at three key spike days and identified the accounts that were key in amplifying the Voter Fraud narrative through tweets or retweets. Through this analysis we were able to identify a group of interconnected accounts is tweeting and retweeting each other's tweets, building up the discourse around Voter Fraud. Specifically, we observed many accounts that mentioned and retweeted each other and had similar aesthetics and narrative content, all actively tweeting about Voter Fraud during the same spike dates. While some of these spike dates amplified around organic events (i.e., newsworthy event), many appeared to be synthetic attempts to amplify the Voter Fraud narrative. All of these factors taken together indicated further evidence of possible coordination.

3. Surge Patterns

Posts by and mentions of @battleofever

2017-10 2017-12 2018-02 2018-04 2018-06 2018-08 2018-10 0,000 0,500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 5,500 6,000 6,500 7,000 Posts by @battleofever Mentions of @battleofever

Accounts from the sample 200 mentioning @battleofever

Posts by and mentions of @bbusa617

2016-10 2017-01 2017-04 2017-07 2017-10 2018-01 2018-04 2018-07 2018-10 00,000 02,000 04,000 06,000 08,000 10,000 12,000 14,000 16,000 18,000 20,000 22,000 24,000 26,000 Posts by @bbusa617 Mentions of @bbusa617

Accounts from the sample 200 mentioning @bbusa617

Posts by and mentions of @girl4_trump

2016-07 2016-10 2017-01 2017-04 2017-07 2017-10 2018-01 2018-04 2018-07 2018-10 00,000 01,000 02,000 03,000 04,000 05,000 06,000 07,000 08,000 09,000 10,000 11,000 12,000 13,000 Posts by @girl4_trump Mentions of @girl4_trump

Accounts from the sample 200 mentioning @girl4_trump

Delving further into the individual profiles and tweet history of the accounts tweeting about Voter Fraud, we noticed that the majority of these accounts exhibited a surge in activity at one point of their life cycle. The surge pattern describes the moment in an account's history when it goes from an average cadence of activity, which one would expect of a Twitter account whose follow grows organically, to a high volume of tweets, retweets and mentions, unexplained by the status of the account (ie. the account is not a verified Twitter user or public figure). While not all of these accounts come online at the same time, we saw this pattern consistently repeated by the group of accounts tweeting Voter Fraud related messages. The fact that such a significant majority of the accounts talking about Voter Fraud were tweeting at the same time, mentioning each other and displaying similar surge patterns further indicated the possibility of a coordinated influence campaign.

4. Replicating the Surge and Growing the Network

Posts by and mentions of @battleofever and @thebeasmith

2015-01 2015-04 2015-07 2015-10 2016-01 2016-04 2016-07 2016-10 2017-01 2017-04 2017-07 2017-10 2018-01 2018-04 2018-07 2018-10 0,000 0,500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 5,500 6,000 6,500 7,000 Posts by @battleofever Mentions of @battleofever Posts by @thebeasmith Mentions of @thebeasmith

Accounts from the sample 200 mentioning @battleofever and @thebeasmith

Posts by and mentions of @bbusa617 and @pinkk9lover

2015-10 2016-01 2016-04 2016-07 2016-10 2017-01 2017-04 2017-07 2017-10 2018-01 2018-04 2018-07 2018-10 00,000 02,000 04,000 06,000 08,000 10,000 12,000 14,000 16,000 18,000 20,000 22,000 24,000 26,000 Posts by @bbusa617 Mentions of @bbusa617 Posts by @pinkk9lover Mentions of @pinkk9lover

Accounts from the sample 200 mentioning @bbusa617 and @pinkk9lover

Posts by and mentions of @girl4_trump and @1776hotlips

2016-01 2016-04 2016-07 2016-10 2017-01 2017-04 2017-07 2017-10 2018-01 2018-04 2018-07 2018-10 00,000 02,000 04,000 06,000 08,000 10,000 12,000 14,000 16,000 Posts by @girl4_trump Mentions of @girl4_trump Posts by @1776hotlips Mentions of @1776hotlips

Accounts from the sample 200 mentioning @girl4_trump and @1776hotlips

To determine how these accounts were surging, and gaining mentions and influence at such a high velocity we investigated the daily activity of accounts discussing Voter Fraud. Through this analysis, we observed a pattern of influence that was replicated by a number of accounts, indicating possible coordination.

Here's the network pattern we saw:

  • User signs up for an account.
  • User starts replying to multiple accounts—some known verified Twitter users and many other accounts that are also on our list of actors, or that fit a similar profile.
  • The replies tend to contain: text, memes, hashtags, and @mentions of other accounts.
  • At some point the pattern shifts from being all replies to original tweets. Those original tweets contain the same types of content as their replies do.
  • Our current hypothesis is that this pattern cycles and repeats when other new accounts come online and start replying to these newly minted influential accounts, and carry on with the same order of operations for gaining influence. We call this stacked surging.

Though there is no definitive data to show which method is pervasive in instigating this coordination pattern, we have a few hypotheses:

a) new accounts join Twitter rooms, are included or become involved with train accounts, or coordinate off Twitter in other online gathering places or offline, connecting with networks that accelerate their growth and influence,

b) when other users see the new accounts using the same hashtags as they are, it signals alignment and that these accounts should be amplified, and/or

c) there is a larger, well coordinated, and well funded effort to wield influence about topics like Voter Fraud.

Regardless of whether some or all of these methods are being used to help accounts grow their following and influence on Twitter, it is critical to understand that these coordination tactics appear to bypass many of the detection methods of existing computational propaganda research. Meaning, that while these potential coordinated influence networks are growing, they are becoming more difficult to detect.

Conclusion

Initial research around accounts tweeting, retweeting and mentioning #VoterFraud and adjacent hashtags over the last 12 months demonstrated that co-spiking was occurring, amplifying the influence of this messaging. By expanding this dataset, we observed that this pattern of mentions of #VoterFraud has persisted for at least the last three years with the largest spike in mentions occurring around the 2016 election.

Our analysis of dates when mentions of #VoterFraud spiked showed that there is a group of accounts consistently tweeting, retweeting and mentioning each other, which share similar aesthetics, bios and narrative frameworks. The fact that the majority of these accounts also exhibited a surge of activity during their lifespan, where they grew in following and influence at an exponential rate, further indicated possible coordination.

Through this analysis, we were able to identify a group of accounts that demonstrated a co-spiking pattern, a surge pattern and in addition to having similar aesthetics, were all mentioning, tweeting, retweeting and following each other. Taken together these patterns indicate possible coordination. When combined with the data regarding clustering and the density of mentions among all accounts tweeting about Voter Fraud over the last three years, there’s an indication of a possible network of accounts dedicated to promoting division and conspiratorial narratives around Voter Fraud online.