Capitol (Pat)riots: A comparative study of Twitter and Parler
Hitkul, Avinash Prabhu, Dipanwita Guhathakurta, Jivitesh jain, Mallika Subramanian, Manvith Reddy, Shradha Sehgal, Tanvi Karandikar, Amogh Gulati, Udit Arora, Rajiv Ratn Shah, Ponnurangam Kumaraguru
1111Capitol (Pat)riots: A comparative study of Twitter and Parler
HITKUL,
IIIT - Delhi, India
AVINASH PRABHU ∗ , DIPANWITA GUHATHAKURTA ∗ , JIVITESH JAIN ∗ , MALLIKA SUBRA-MANIAN ∗ , MANVITH REDDY ∗ , SHRADHA SEHGAL ∗ , and TANVI KARANDIKAR ∗ , IIIT - Hy-derabad, India
AMOGH GULATI ∗ and UDIT ARORA ∗ , IIIT - Delhi, India
RAJIV RATN SHAH,
IIIT - Delhi, India
PONNURANGAM KUMARAGURU,
IIIT - Delhi, India
On 6 January 2021, a mob of right-wing conservatives stormed the USA Capitol Hill interrupting the session of congresscertifying 2020 Presidential election results. Immediately after the start of the event, posts related to the riots started to trendon social media. A social media platform which stood out was a free speech endorsing social media platform Parler; it isbeing claimed as the platform on which the riots were planned and talked about. Our report presents a contrast betweenthe trending content on Parler and Twitter around the time of riots. We collected data from both platforms based on thetrending hashtags and draw comparisons based on what are the topics being talked about, who are the people active on theplatforms and how organic is the content generated on the two platforms. While the content trending on Twitter had strongresentments towards the event and called for action against rioters and inciters, Parler content had a strong conservativenarrative echoing the ideas of voter fraud similar to the attacking mob. We also find a disproportionately high manipulationof traffic on Parler when compared to Twitter.Additional Key Words and Phrases: Social Computing, Data Mining, Social Media Analysis, Capitol Riots, Parler, Twitter
Misinformation of the United States of America’s presidential election results being fraudulent has been spreadingacross the world since the elections in November 2020. Public protests and legal cases were taking place acrossthe states against the allegation of voter fraud and manhandling of mail-in ballots. This movement took a violentheight when a mob attacked the Capitol Hill building to stop certification of Mr. Joe Biden as the 46th Presidentof the United States of America. The incident lead to a mid-way halt of a running congress session and the deathof five people including a police officer. The incident also spread its ripples in the online world with hashtagslike ∗ Authors contributed equally to this research. a r X i v : . [ c s . C Y ] J a n In this report, we collect data from two Online Social Media sites (OSMs) - Twitter and Parler - that wereactively used to discuss the riots. Parler is a free speech social network that garnered attention when PresidentDonald Trump publicly denounced social media giants like Twitter and Facebook for targeting him and otherconservatives. The network has also been used by right-wing extremists to plan the Jan 6th breach of the Capitol. Section 2 goes into details of Parler rules and functioning.We conduct a comparative study of the trending content and users on the two platforms. We observe a violentrhetoric on Parler with a significant portion of the content in support of the Capitol riots and misinformed claimsof fraudulent elections. Content on Twitter, by contrast, denounced the storming of the Capitol and the weakresponse from police to the incident, in contrast to the BLM protests. Users on Twitter shared their concernsregarding the rising violent riots in America. A longitudinal analysis of the content and the users’ joining dateshighlights how the two OSMs were actively used during and after the protests, reaffirming the point that socialmedia has a significant role to play in political events.
Parler is a micro-blogging and social networking service launched in 2018 with headquarters in Henderson,Nevada, USA. According to the About page, the network is a free speech social network , built on a foundation ofrespect for privacy and personal data, free speech, free markets and ethical, transparent corporate policy . Parleradvertises the minimal rules and content guidelines it imposes, explaining its popularity amongst users whoare banned from popular social media websites such as Twitter and Facebook due to their content moderationpolicies. In the aftermaths of the Capitol attack, Parler got suspended by Amazon Web Services hosting leadingto platform being unavailable. The Parler Community Guidelines are based on two principles: 1) Parler’s services cannot be used as a toolfor crime and unlawful acts and 2) Bots and spam are a nuisance and not conducive to productive and politediscourse. While threats of violence and advocacy of lawless actions are prohibited, fighting words and NSFW(Not-Safe-For-Work) content is allowed, under some restrictions. Reported violations of these guidelines arereviewed by a Community Jury which determines whether the content is permitted or not. A point system is inplace to ban repeated and frequent offenders. Parler allows registered users to write parleys , which are posts at most 1,000 characters long. Social Networkengagement features such as comments and votes on parleys written by others are also present. Each user hastheir own feed - a stream of parleys that they can interact with. Unlike other popular platforms the feed is notcurated by Parler. Users curate and moderate their feed using options provided by the platform - reflecting https://parler.com/ https://company.parler.com https://legal.parler.com/documents/Elaboration-on-Guidelines.pdf https://legal.parler.com/documents/Parler-Community-Jury.pdf apitol (Pat)riots: A comparative study of Twitter and Parler • 111:3 Table 1. Terminology used by Parler.
Term Definition
Parley A Parley is a 1,000 word post that can be shared on the Parler Platform.Hashtag A word or phrase preceded by a hash sign ( discover section. The discover section consists ofparleys, people and affiliates . Affiliates are news outlets which are allowed by Parler to post their news articles.Table 1 provides a description of common terms and actions associated with the network.
Our data collection was done on the 7 and 8 January 2021. A list of trending hashtags and keywords were curated,and used as the seed to collect data. Parleys were collected using parler-py-api. Data collected was datedbetween 1 Nov 2020 03:39 am EST to 8 Jan 2021 08:15 am EST. A total of approximately 100,000 parleys from22,000 unique users were collected. For collecting Tweets, we used the official Twitter Streaming API. Our collection period was 7 Jan 8:00 AMEST to 8 Jan 7:15 AM EST, while adhering to rate limits. A total of approximate 4 Million tweets were collectedfrom 1.7 Million users. Table 2 provides a summary of our dataset statistics. https://github.com/KonradIT/parler-py-api/ http://precog.iiitd.edu.in/resources.html Table 2. Summary of dataset statistics
Parler Twitter
Total Parleys 101,945 Total Tweets 4,196,988Echos 35,165 Retweets 3,288,274Unique users 22,326 Unique users 1,720,826 (a) Parler (b) TwitterFig. 1. Ten Most frequently used hashtags. Hashtags on Parler are echoing the ideas similar to the Capitol attacking mob,Twitter is treding with hashtags representing a call for action against the event.
We compared the collected data from Parler and Twitter on the basis 1)
What was being posted, 2)
Who were thepeople posting and creating engagement, and lastly 3)How organic was the traffic generated.
To understand each platform’s topics of trending conversation, we looked at the top ten most common hashtagson respective platforms. Figure 1 shows the ten most frequently used hashtags represented with the percentageof post they appear. Percentage of use for a particular hashtag has been calculated by counting the number ofposts mentioning that hashtag at least once, normalised by the total number of posts on that platform containingat least one hashtag.We observe a stark contrast in the hashtag popularity on two platforms. All the hashtags trending on Parlerrepresent the misinformed idea of voter fraud and echo the ideas similar to attacking mob. Hashtags popular onTwitter are either neutral towards the event, e.g. and or the ones who are calling outfor impeachment of President Donald Trump who is being called responsible for inciting his follower for thisattack. One exception to this result is the presence of on Twitter, which is associated with the PresidentTrump’s election campaign. apitol (Pat)riots: A comparative study of Twitter and Parler • 111:5 Table 3. Sample Parleys using popular hashtags.
Interesting.. The cops should have just went home..We will never call Sleepy Creepy PEDO maker Joe ourPresident. ...they stopped counting in Chatham County at like10:30-11 last night - "resumed" this morning. Repub-licans were leading both seats with over 90% of thevote in... ... MAGA PATRIOTS please support Trump and stormthe Capitol and Congress properly.!! We will installTruml as our Lord and savior, PRESIDENT FOR LIFE...because we won the election!! THIS TIME WE TAKETHE CAPITOL WITH GUNS bcs it is not fair welost DESTORY AMERICA OR TRUMP STAYS PRESI-DENT... its the most patriotic thing to do...
Antifa is in there. They have breached theHouse...
75% ... ...Treason against the United States is taking place sur-rounding the presidential election. The largest cyberwarfare activity in the world... ...Still haven’t seen where Biden and Harris havetaken their vaccines. Has anyone else seenit?...
Despite the crisis, a sense of unity & patriotism afford ...It is also worth noting that most frequent hashtag on Parler is present in above 70% of posts compared to onlyabove 12% in case of Twitter representing an extreme one-sided narrative present on Parler. To provide a betterinsight of context in which these hashtags have been used, Table 3 and Table 4 shows 5 parleys and tweets fortwo most frequent hashtags respectively.To get further insights into the content being posted on both platforms, we repeated the frequency plotsfor terms shown in Figure 2. We have used the word term in this context to mean any uni-gram (excludingstop words) included in a post, but not as a mention of a user or a hashtag. Percentage of use for a particularterm has been calculated by counting the number of posts mentioning that term at least once, normalised bythe total number of posts on that platform. We observe the same pattern as shown by hashtags, terms used onTwitter, indicating a sense of dissatisfaction and disdain towards the US Capitol’s actions. On the other handterms, frequent on parley indicate a strong sense of support towards undermining the veracity of the 2020 USpresidential elections. Parleys also display abundant use of strong language compared to the tweets, which can beeasily attributed to the liberal community guidelines, and supports claims that the platform is used as a mediumfor proliferating violent and offensive content.Apart from this stark difference in opinion, some terms harbour a deeper meaning. On Twitter, terms like advocate President Donald Trump’s immediate removal from office. In contrast, those like black draw a comparison between law enforcement agencies’ response to the demonstrations at the US Capitol and theBlack Lives Matter protests of July 2020, adding a racial dimension to the conversation.Since terms trump and president are common and occur frequently in both the platforms, we compare thecontexts in which two popular terms appear in parleys and tweets in Table 5 and Table 6. These Tables reinforce
Table 4. Sample Tweets using popular hashtags.
Dear @VP @Mike_Pence: In light of the yesterday, there are bipartisan calls for you toinvoke the 25t. . . President and Commander in Chief Trump should beeither removed from office with the 25th amendment,impeached, and/or investigated for criminal charges.
He betrayed the country heswore to protect. The Constitution and Democracymean nothing to him. He. . .The yesterday were underpinned bypure racist hatred nothing less. This started withTrumps birtharis. . . We can’t wait 13 days.
The were a terrorist act incited byDonald Trump, Don Jr, Rudy Giuliani and membersof Congress. Arre. . . The right time to do the right thing is RIGHT NOW. (a) Parler (b) TwitterFig. 2. Ten most frequently used terms. Conservative and election fraud terms most frequently appeared on Parler representingan association with the idea of rioters. However, Twitter is dominated by call for 25th amendment and comparisons withblack live matter protest. our earlier inferences of Parler content supporting the riot and voter fraud narrative, whereas Twitter samplesare against the riot and President Trump. apitol (Pat)riots: A comparative study of Twitter and Parler • 111:7
Table 5. Context Comparison for the term
Trump . Parler Twitter
Plandemic Exposed...
Trump knows that the Dum-mies are already forging their narrative with howthey will stop this plandemic as soon as Chinese Joegets in the Oval Office... Well,
Trump finally got his wall. It’s in WashingtonDC....I urge all
Trump supporters and lovers of freedomto protest at your place of government. We can notlet this nation fall to the hands of foreign puppets... No matter how
Trump now behaves...Listening to
Trump speak.
Trump needs to resign or be removed fromoffice. America has endured enough.
Trump pledging appear personally at the rally of out-right fascists and white supremacists on January 6thas part of his coup efforts. This is an act of intimi-dation that has parallels with Mussolini’s march onRome. It is aimed at intimidating not just the workingclass as a whole but Trump’s opponents...
Trump did NOT immediately send out the NationalGuard. No initial presence. Clashes began ∼ Trump supporters tackledhim in 30 seconds. trump if he’s not immediately removed by the 25thAmendment. LET’S GOOOO!!!!
Users on both platforms were compared across three attributes - Content generated, Mentions and Reposts. Sincethe datasets are vastly different in size, we normalize by number of posts to gauge the differences between theplatforms. We start with filtering out users by the volume of content they generate. We also compare user biosand joining date to get a bird’s eye view on the users’ type present on the platforms. Tables 7 and 8 shows thefive most active users on Parley and Twitter respectively.We once again observe a large contrast between the users and their posts. The users on Parler posted manyextreme Parleys in which they urged people to take part in the protest and later showed support for the protest. Onthe other hand, Twitter users openletterbot and
RogueRiverSun posted extensively against the protest. openletterbot is a bot used to deliver messages from the public to elected officials and posted many anti-protest messagesduring the Capitol storming. On Parler approximately 11% of the total content is generated by only five users thismay indicate that there is heavy traffic manipulation on Parler, we attempt to study this in Section 4.3.Next, in Table 9 and Table 10 we list the top five most mentioned users on both platforms. We observe PresidentTrump’s account getting most mentions on both the platforms. However, there is a stark contrast for the rest ofthe users on the list. We observe predominantly right-wing personalities being mentioned frequently on Parlerwhereas on Twitter, we observe a larger inclination to left-wing leaders being mentioned. The lack of left-wingleaders not being mentioned on Parler may be due to the fact that most left-Wing Leaders do not have a Parler
Table 6. Context Comparison for the term
President . Parler Twitter ...It took a small group of Patriots to enter the Capi-tol Building to get the attention of America aboutthe president elect. Almost 50% of Americansdo not think it was a free and fair election... Why are these swamp creatures in such a rush touse the 25th amendment or to impeach
President @realDonaldTrump? Do. . .I proudly stand with
President
Trump and I will fightto the death to ensure the sanctity of our Republic... The Vice
President and the Cabinet should vote, to-day, on invoking the 25th amendment. Every secondthat Donald Tr...
President
President
Trumpaccountable...GOD BLESS
PRESIDENT
TRUMP!!! The pres-ident says HELL NO! to the PORK STUFFEDBURRITO that CONGRESS calls A STIMULUSBILL!!!...
President has betrayed the coun-try and incited an insurrection against our own gov-ernment. Retweet i...I also stand with
PresidentTrump ... pres-ident in the history of the United States.
Table 7. Most Active Users on Parler. Five users accountfor 11% content generated.
User Posts Percentage
Patriots4US 6320 6.19%TheRealWakeUpMfers 2580 2.53%GameOver 1406 1.37%Billyboy428 1105 1.08%marylandcrabbing 994 0.97%
Table 8. Most Active Users on Twitter.
User Posts Percentage openletterbot 511 0.0001%4Tchat 360 0.0001%Difference30360 333 0.0001%Anime_ABEMA 330 0.0001%RogueRiverSun 317 0.0001%account. The FBI is also mentioned dominantly on Twitter due to a movement in which the identity of the peopleinvolved in the riot was identified on Twitter and reported to the FBI account. In Tables 11 and 12 we list top five users with the highest number of Reposts on both platforms. Most ofthe accounts belonged to people who were involved in Media and Press. However, the content that two sets ofusers posted were extensively different. A majority on Parler posted in support of the “revolution" and believedstrongly in the riots whereas the users on Twitter put up posts which condemned the riot on the platform. Thisstrongly shows the disparity in the content that is spreading on both websites. We further notice an abnormallyhigh repost percentage on Parler coming from a small set of users, this combined with the single-sided narrativeof most user-generated content is a vital sign of a platform level echo chamber on Parler. https://twitter.com/FBI/status/1348283582490546177?s=20 apitol (Pat)riots: A comparative study of Twitter and Parler • 111:9 Table 9. Most Mentioned Users on Parler. Predominatelyright-wing user accounts.
User Mentions Percentage
TeamTrump 3363 3.30%linwood 2802 2.75%SeanHannity 2185 2.14%Marklevinshow 2139 2.09%GenFlynn 1805 1.77%
Table 10. Most Mentioned Users on Twitter. Predomi-nately left-wing leader, President Trump being an ex-ception. The movement of Identify rioter. online led toheavy presence of FBI.
User Mentions Percentage realDonaldTrump 87223 0.023%AOC 53418 0.014%FBI 37317 0.010%HawleyMO 32931 0.009%SpeakerPelosi 30325 0.008%
Table 11. Most Reposted Users on Parler. Twenty percentof repost content being generated by five conservativeaccounts, present strong sign of skewed narrative onParler.
User Reposts Percentage
WarRoomPandemic 11352 11.1%Ryanahlberg 5342 5.2 %epochtimes 2017 1.9 %tjf2020 1538 1.5 %JoePags 1123 1.1%
Table 12. Most Retweeted Users on Twitter. Predomi-nantly left-wing accounts were retweeted on Twitter.
User Reposts Percentage
AOC 42016 0.012%kylegriffin1 26270 0.007%BarkyBoogz 25503 0.007%SethAbramson 24055 0.007%MeidasTouch 23633 0.007%Finally, to garner a sense of user-profiles we generate a word cloud of user bio and time of joining the platformshown in Figure 3 and 4 respectively. Though results shown by the word cloud, do not show a clear separationlike our other plots, mainly because of a heavy presence of trump and maga in Twitter bios which are generallyassociated with conservatives. However, users on Twitter indicate some signs of diversity with presence of termslike love , life , mom and fan versus that of Parler which is only populated with terminology associated withconservatives.While analysing the user joining date, we observed three distinct peaks in Parler. From left to right in the graph,these peaks correspond to the black lives matter protests from May-June 2020, the We calculated the Coefficient of Traffic Manipulation (CTM) [5] for frequently occurring hashtags on bothplatforms. CTM is a relative metric to measure how much traffic of a given hashtag has been manipulated.Equation 1 provides an mathematical representation of CTM. (a) Parler (b) TwitterFig. 3. Wordclouds of user bios. Twitter presents some diversity with presence of terms like love , mom , trump , whereas, Parleronly contains conservative right-wing terms same as those echoed by the attacking mob.(a) Parler (b) TwitterFig. 4. Proportion of accounts created over time. All the user sign-up spikes on Parler are during a political event, explaininga highly polar userbase. In comparison, Twitter have a steady rate of sign-ups with less prominent peaks during politicalevent. However, A large rate of sign-up is observed on Twitter during the time of riots. 𝐶 = 𝑅 + 𝐹 + 𝑈 (1)Here, for a given hashtag 𝑡 :- • 𝐶 is Coefficient of Traffic Manipulation for 𝑡 . • 𝑅 is percentage of 𝑡 traffic created by reposts. • 𝐹 is percentage of 𝑡 traffic created by top fifty users. • 𝑈 is average number of posts per user for 𝑡 .Using equations 1, we find the value of 𝐶 for most popular hashtags on Twitter and Parley shown in Figure 5.CTM for the most manipulated hashtag on Parler is approximately 46 times more than the most manipulatedhashtag on Twitter, indicating a large extent of traffic manipulation on Parler. A similar trend can be seen for all apitol (Pat)riots: A comparative study of Twitter and Parler • 111:11 (a) Parler (b) TwitterFig. 5. CTM on popular hashtags. Traffic on Parler is disproportionately more manipulated as compared to that on Twitterwhere all tags are within the range of organic traffic. Highest CTM score on Parler is times higher than that of Twitter. other frequently occurring hashtags too. Though CTM is a relative scale Nimmo [5] in there report anecdotallyfound that CTM for organic traffic generally was less than 12. Judging on that scale along with the abundance ofpro-riot content observed in our previous experiment indicates the presence of a manipulative rigged electionagenda on Parler, which eventually could have fueled into the tragedy of 6 January 2021. In this report, we analyse trending traffic from Twitter and Parler, in light of the riots at the US Capitol inWashington DC on January 6, 2020. We look at the kind of content being generated and the users involvedand find evidence supporting the claims that a significant proportion of traffic on Parler was in support ofundermining the veracity of the 2020 US Presidential Elections. We also find the traffic on Parler to often beviolent, consisting of hate speech (as a result of Parler’s relaxed community guidelines) and manipulated asindicated by CTM. On the other hand, Twitter users majorly used the platform to condemn the incident and itsperpetrators and show their disdain towards President Donald Trump for his actions. An in-depth analysis ofthese users’ networks and their activity across the two websites are required to understand better the role ofsocial media in incitement and planning of these riots. We are continuing to analyze this data for more insights.
ACKNOWLEDGMENTS
Hitkul funded by TCS Research Fellowship.
REFERENCES [1] Alessandro Bessi and Emilio Ferrara. 2016. Social bots distort the 2016 US Presidential election online discussion.
First Monday
21, 11-7(2016).[2] Gillian Bolsover. 2020. Black Lives Matter discourse on US social media during COVID: polarised positions enacted in a new event.[3] Emilio Ferrara. 2017. Disinformation and social bot operations in the run up to the 2017 French presidential election. arXiv preprintarXiv:1707.00086 (2017).[4] Fred Morstatter, Yunqiu Shao, Aram Galstyan, and Shanika Karunasekera. 2018. From alt-right to alt-rechts: Twitter analysis of the 2017german federal election. In
Companion Proceedings of the The Web Conference 2018 . 621–628. [5] Ben Nimmo. 2019. Measuring Traffic Manipulation on Twitter.[6] Anmol Panda, Ramaravind Kommiya Mothilal, Monojit Choudhury, Kalika Bali, and Joyojeet Pal. 2020. Topical Focus of PoliticalCampaigns and Its Impact: Findings from Politicians’ Hashtag Use during the 2019 Indian Elections.