Content uploaded by Ángel Panizo Lledot
Author content
All content in this area was uploaded by Ángel Panizo Lledot on Jan 30, 2020
Content may be subject to copyright.
DESCRIBING ALT-RIGHT COMMUNITIES AND THEIR DISCOURSE
ON TWITTER DURING THE 2018 US MID-TERM ELECTIONS
A PREPRINT
A´
ngel Panizo-LLedot
Computer Science Department
Universidad Autónoma de Madrid, Spain
angel.panizo@uam.es
Javier Torregrosa
Biological and Health Psychology Department
Universidad Autónoma de Madrid, Spain
francisco.torregrosa@uam.es
Gema Bello-Orgaz
Departmento de Sistemas Informáticos
Universidad Politécnica de Madrid, Spain
gema.borgaz@upm.es
Joshua Thorburn
RMIT University
Melbourne, Victoria, Australia
joshthorburn18@gmail.com
David Camacho
Departmento de Sistemas Informáticos
Universidad Politécnica de Madrid, Spain
david.camacho@upm.es
January 30, 2020
ABS TRAC T
The alt-right is a far-right movement that has uniquely developed on social media, before becoming
prominent in the 2016 United States presidential elections. However, very little research exists
about their discourse and organization online. This study aimed to analyze how a sample of alt-right
supporters organized themselves in the week before and after the 2018 midterm elections in the
US, along with which topics they most frequently discussed. Using community finding and topic
extraction algorithms, results indicated that the sample commonly used racist language and anti-
immigration themes, criticised mainstream media and advocated for alternative media sources, whilst
also engaging in discussion of current news stories. A subsection of alt-right supporters were found
to focus heavily on white supremacist themes. Furthermore, small groups of alt-right supporters
discussed anime, technology and religion. These results supported previous results from studies
investigating the discourse of alt-right supporters.
Keywords polarization, far-right, community finding, topic extraction, political extremism
1 Introduction
In recent years, far-right political parties and candidates have achieved increased electoral success across various
Western states. Furthermore, terrorist attacks by far-right extremists, such as in Norway and New Zealand, demonstrate
the risk that this extremism poses. Although it has a long history, the far-right has flourished in recent years, with the
effective use of social media strongly contributing to this rise. While the alt-right itself is “highly decentralized”, with
no official leaders, ideology or political party, this broad far-right amalgamation has effectively used social media to
attract followers and influence political discourse. Indeed, the alt-right played a key supporting role in the election of
US President Donald Trump. Alt-right supporters have used a variety of social media networks to spread content and
engage in political discussion, including Twitter, Facebook, Reddit, 4Chan and YouTube. Notably, in response to social
media companies increasingly removing hate speech and extreme far-right accounts, alt-right supporters have recently
congregated on less-restrictive websites, such as Gab, an alternative to Twitter.
APREPRINT - JANUARY 30, 2020
While the discourse of other white extremist groups has been previously studied by academics [
1
], there are few articles
concerning how alt-right sympathizers communicate and interact online. However, the alt-right is a difficult group to
study due to the movement’s fragmentation and lack of ideological agreement. With these limitations in mind, the
present research aims to analyze how a set of alt-right supporters communicated on Twitter during the 2018 US midterm
elections, a politically relevant period. To evaluate this, two questions were posed. First, "Are there different subsections
of alt-right followers?", and second, "What are the topics discussed in these groups?".
2 Related work
This section first describes the alt-right discourse and it similarities and dissimilarities with other ultra-conservative
groups, before reviewing common techniques for doing online discourse analysis.
2.1 Alt-Right discourse
Firstly, opposition to immigration, particularly from the Middle East, Africa or Latin American countries, is a central
political stance of alt-right supporters and the far-right. For example, the refugee crisis stemming from the Syrian
Civil War led to far-right groups using hashtags to reject European countries accepting these immigrants, [
2
], such as
#Ausländerraus (“Foreigners out!”), #EuropeforEuropeans or #refugeesnotwelcome. Furthermore, alt-right followers
have been supportive of Trump’s strict immigration policy proposals [
3
]. Lyons [
4
] found that original manifestos from
prominent ideological thinkers of the alt-right were deeply rooted in a rejection of non-white immigration. Similarly,
alt-right supporters commonly use racist language to describe immigrants, much like other far right groups [
5
]. Indeed,
the alt-right has popularised the use of white supremacist hashtags such as “#ItsOkToBeWhite, or #WhiteGenocide
[
6
]. Supporters of the alt-right frequently attack minorities and are especially hostile towards black people, Jews and
Muslims. Consequently, while they are often aligned with Trump, alt-right followers have criticized him for being too
soft against Saudi Arabia and Israel because they symbolically represent Islam and Judaism [7].
Criticism of mainstream media is also common in alt-right discourse [
4
], with this often mirroring President Trump’s
frequent lamentations about “fake news”. A study conducted by Forscher & Kteily [
8
] found that alt-right supporters
tend to have high levels of suspicion towards mainstream media sources (i.e, New York Times, CNN, etc.) and strong
trust in alternative media sources, such as the conspiratorial InfoWars. Conspiracy theories are also commonly believed
alt-right followers. For example, it is commonly believed that certain media companies are actively trying to bring
down the Trump presidency, or that a secretive Jewish elite hold enormous power [9].
Alt-right supporters tend to be very politically active on various online social networks. Through the analysis of user
descriptions from the online platform Gab (an online platform similar to Twitter) it was found that most of its far right
users included references to Trump or his campaign slogans, conservative topics, religion or America [
10
]. Hashtags
against Islam, about the alt-right itself were also found on the website. This analysis of Gab found that there posts
by users contained 2.4 times more hate words than on Twitter, but less than half than those on 4chan. 4Chan itself,
especially its /pol/ (‘Politically Incorrect’) message board, has been strongly linked with providing a fertile breeding
ground for the alt-right to develop as a movement [11].
Although many of the ideological underpinnings of the alt-right are shared with and stem from earlier and contem-
porary far-right movements, the alt-right has differed in some respects too. For example, although some alt-right
supporters are fervently religious, they often differ strongly from the Evangelical conservative Christian right in the US.
Indeed, “cuck”/”cuckservative”, the frequently used alt-right pejorative, derives from a sexual act. This slang is also
representative of another characteristic of the alt-right: its ability to use humour to attract supporters. By using ironic,
dark-humoured satire, the alt-right has succeeded in flippantly spreading white supremacist ideology on various online
platforms [
12
]. Further, the alt-right has used internet memes, as well as creating its own imagery and slogans to attract
supporters.
2.2 Techniques for online discourse analysis
A number of technical approaches can be used to analyse online communications. Depending on the focus of the
research (e.g. text, interactions, frequency of writing, etc.), certain methods are more appropriate. Natural Language
Processing (NLP), which includes sentiment analysis, topic extraction or linguistic inquiry, is a common approach in
analysing large textual datasets. Examples of this includes a study by Torregrosa et al [
13
] where the tweets of jihadi
supporters on Twitter was examined using Linguistic Inquiry Word Count (LIWC), or in a study by Tumasjan et al [
14
]
where sentiment analysis was used to analyze tweets referring to political parties.
2
APREPRINT - JANUARY 30, 2020
Understanding how people and groups interact on Online Social Networks (OSNs) is important when analyzing
discourses. Social Network Analysis (SNA) techniques are especially useful to study these interactions. While some
SNAs are focused on extracting common characteristics of a network, such as the density or its diameter, others focus
on characterizing properties of the users of the network. For example, analyzing the Homophily of the network or
finding cohesive groups of users (communities) can be an effective approach. ‘Community finding’ is one of the most
commonly used techniques to investigate and analyze OSNs.
Several studies have combined community finding with NLP techniques. Among them, three main approaches can be
found in the literature. The most frequent approach combines NLP techniques and community finding into a singular
process. In this approach, communities are not only consistent in a structural manner, but also in the topics they
discuss. Feller et al. [
15
] used this type of method when applying a network analysis to contextualize the users political
preferences during the 2016 Germany elections. The second approach apply both process simultaneously but separated,
comparing the outcomes later. An example of this approach can be found in Surian et al [
16
], where they analyze
the discussions about HPV vaccines on Twitter. Finally, the third approach initially applies the community finding
process, and then applies NLP techniques in each community independently. This was used by Yin et al [
17
] in a study
comparing two datasets, with one containing computer science journals and the other composed of tweets that included
the keywords "Obama" or "media". In the present research, this third approach was applied to examine the tweets and
the communities.
3 Methodology
3.1 Data description
The dataset used in this study contained 52903 tweets published by 123 alt-right Twitter accounts collected between
the
30th
of October and the
13th
of November, 2018. This period corresponds to the weeks before and after the 2018
US midterm elections. In addition to the text of each tweet, information such as publication date, retweets or hashtags
was gathered. The selected alt-right Twitter accounts were chosen from a dataset created by Thorburn, Torregrosa and
Panizo [6]. The alt-right users published between 2 and 2590 tweets during the data collection period. In fact, ≈60%
of the accounts tweeted every day of this period and
≈80%
of the users in the dataset tweeted on at least 10 different
days.
As shown in Figure 1 (a), the number of tweets published per day was between 2656 and 4539 tweets, with increased
activity surrounding election day (November 6, marked as a red vertical line), with this peaking on November 7, the day
immediately after the elections. However, in regards to the number of users posting each day (shown in Figure 1 (b)), it
can be noticed that the same spike in activity is not present. Finally, regarding the daily distribution of tweets published
by each account (see Figure 1 (c)), the median and the whiskers of the box blot diagram indicate that this also increased
surrounding election day.
3.2 Protocol
Once the data was gathered to create the dataset, this information was divided and analyzed using community finding
and topic detection algorithms. The phases of this process is detailed as follows:
1. Data Extraction
: The data collected to perform the analysis of the alt-right discourse was extracted from the
Twitter API. Every tweet published by the users in the selected two time periods was included for analysis.
The extracted tweets are stored in a database (MongoDB) together with their related information, such as
creation date, location, source, retweet count, etc.
2. Data Preprocessing
: Two re-tweet networks of users were created, representing the interactions between all
users before and after the election day. To create each network, users were considered as the network nodes,
and their relationships represented the edges. The relationships were established using the re-tweets.
3. Community Detection of Alt-Right Users
: The detection of relevant groups of users within the OSN can be
addressed by the application of community detection algorithms in the re-tweet networks. Before applying the
community search algorithm, a filter was used to get the largest connected component of each re-tweet network.
Then, the most similar users were grouped together using the Clauset-Newman-Moore greedy modularity
maximization method [18, 19].
4. Topic Extraction by the Communities
: Once the different communities were obtained for each re-tweet
network, a topic modeling technique was applied to identify the topics covered by a collection of texts. With
this aim, the Latent Dirichlet Allocation (LDA) model [
20
] was fitted and applied for each community, using
3
APREPRINT - JANUARY 30, 2020
(a) (b) (c)
Figure 1: Subfigure (a) shows the number of tweets published each day. The x-axis contains the days and the y-axis the
number of tweets published. Subfigure (b) shows the number of accounts that published something every day of the
data collection period. The x-axis contains the days and the y-axis the number of accounts that tweeted on each day.
Subfigure (c) shows the distribution of the number of tweets published by an account for each day. The x-axis contains
the days and the y-axis the number of tweets published. The red line marks election day (November 6).
Categories Example keywords Example hashtags Code
Racial discourse
White, Racist, Black, Hate, Dis-
crimination, Diversity, Race, Israel,
Semitic, Jewish, Jerusalem, Islam,
Iran, Pakistan
#Iran, #WhiteGenocide,
#ItsOkToBeWhite RD
Politics
Trump, Obama, Hillary, President,
Right, Left, Party, Republican,
Democrats, Maga, Antifa
#Maga, #Democrats,
#AmericaFirst P
Immigration
Border, Immigration, Caravan, Mi-
grants, Citizenship, Wall, Asylum,
Wave, Illegal, Invasion, Refugees
#BirthrightCitizenship,
#WalkAway, #StandUp-
ForEurope
I
Media
Media, News, Fake, Gab, Alterna-
tive media, Press, Hivemind, Propa-
gandistic, CNN, Conspiracy, Jour-
nalist
#Killstream, #WSJKill-
sKids, #CNN M
Elections
Election, Vote, Campaign, Senate,
Mid-term, Rally, Ballots
#Midterms2018,
#VoteRed, #ElectionDay E
Policy debate
Money, Taxes, Taxation, Marxism,
Socialism, Communist, Freedom,
Nationalism, Brexit
#TaxationIsTheft PD
Table 1: Categories used to classify the topics covered in the alt-right communities.
only the text of the tweets published by the users belonging to the particular community. In addition, the most
mentioned hashtags for each community were extracted.
5. Analysis of the Social Discourse and Evolution of the Communities
: The analysis of the discourse of each
alt-right community on Twitter was conducted using the evolution of its relevant topics. For this purpose, topics
were manually categorized according to the general themes that were found to be relevant in the discourse of
the alt-right groups in the related work section. A summary of the categories can be found in Table 1.
4 Results
Table 2 summarizes the communities extracted from the sample before the election, along with information relating
to the topic extraction process, the most mentioned hashtags and the categories in which the topics were included.
4
APREPRINT - JANUARY 30, 2020
Nearly all the communities contained politically focused discourse, including references to the elections (#ElectionDay),
right-wing political slogans ("Make America Great Again", #MAGA) or other relevant political groups (#Anonymous).
Most of the groups include words associated with far-right ideologies and the alt-right more specifically. Racist
discourse was identified in many of the communities. This can be subcategorized into white supremacy (communities
1, 3, 4, 10 and 12), antisemitism (communities 1 and 10) and Islamophobia (community 10). Group 12 was found
to be especially extreme in its language, using hashtags such as #WhiteGenocide and #WhiteLivesMatter, which are
associated with the alt-right and white supremacism.
The references to the media were divided between pro-comments (comments defending alternative media) and anti-
comments (comments against traditional media). In the first subcategory, there are references to media related with
far-right movements (#Killstream, #GAB, #RedNationRising). There are also references to Gab (some users presented
a link to their Gab profile on their Twitter descriptions). The second subcategory includes criticisms against traditional
media, with this including hashtag campaigns against CNN ("#CNNCredibilityCrisis) and the Wall Street Journal
(#WSJKillsKids). Criticism of traditional media outlets was especially notable in community 14, where CNN was
mentioned several times.
As expected, immigrations was frequently mentioned, and this can be divided between references to illegal immigration
from Mexico (e.g. communities 8 and 11) or the refugee crisis in Europe (e.g. communities 5, 10 and 11). The words
"Trump" and "Border" appear together in some groups (e.g. community 8), which indicates that users commonly
discussed the Trump administration’s immigration policies.
Interestingly, three groups seem to talk about isolated topics. Community 6 is the only one that focused on Brexit,
taxes and political ideologies (Communism, Socialism, Neoliberalism, etc.). Furthermore, communities 9 and 15
centred around discussion of anime and religion respectively. These outliers illustrate some of the distinctions and
demographics that the alt-right is composed of, which is elaborated on further in the discussion section.
Concerning the week after the election, Table 3 shows changes in some of the communities. Firstly, there was an
increase in tweets about the election itself (communities 1, 2, 3, 8, 9, 12 and 13), with the word “election” and the
#ElectionDay being used more. Notably, #Broward featured prominently, with this hashtag referring to difficulties in
vote counting in the county of Broward, Florida. Various stories from right-wing media and political figures alleged
that there had been a conspiracy to rig the election in favour of Democrats in this county. This alleged conspiracy was
especially prominent in communities 2, 3 and 12, where the #StopTheSteal was frequently used.
While no community maintained their original structure from the previous week, some of them conserved their topics
and their most relevant users, as seen on Fig. 2. This was the case in community 6, which featured political debate
in both weeks; community 7, which was dedicated to discussing anime (like community 9 from previous week); and
community 10, which was the most extreme community in the second week too, using exactly the same hashtags that
community 12 had used previously.
Most of the topics identified in the week before the election were also found in the second week. White supremacist
(communities 2, 3, 8 or 10) and antisemitic (communities 9 and 17) discourse was maintained, but there was no
community that included a clear reference to Islam in the two weeks that our data covers. Still, discussion of
immigration featured in communities 1, 10 and 11. Again, there were comments against traditional media (communities
3, 5, 14 and 16), including with references to "fake news" (Community 16). There were also tweets supporting
alternative media (Community 5, 10 and 16). Interestingly, Gab and the hashtag #KillStream (this hashtag is for
a far-right podcast of the same name) are mentioned again. Lastly, several communities mention politics in their
discussions (e.g. Community 8).
Finally, there were three new categories in the week following the election. First, community 14 contained references
to war and military forces, mentioning the hashtags #Veteran and #Iran. The second was community 15, which was
focused on debates about technology in general (including references to blockchain, crypto currency, AI, etc.). Lastly,
debate about abortion was mentioned by community 12.
5 Discussion and Future work
This study aimed to analyze the communications of a group of alt-right supporters in the context of an important political
event, in this case, the 2018 US midterm elections. Firstly, we demonstrated that there was increased Twitter activity in
the days immediately surrounding the election. The outcomes of this study also demonstrates that sub-communities
from this database were mostly unstable over the data collection period, but that many topics did maintain over this
period.
5
APREPRINT - JANUARY 30, 2020
Id. N.U. Top 3 Topics Top 3 Hashtags Labels
1 909 people white new want america Killstream RD(we, as)
got racist better hate nyt MAGA P
anti gop twitter election black semitic WSJKillsKids E
2 745 trump president obama media campaign MAGA E
trump left border caravan people Midterms2018 P
democrat gillum breaking party florida VoteRed M (anti)
3 531 really trump true girl love WSJKillsKids E
good man bad today voting Killstream M (anti)
women gt hate american twitter HappyHalloween RD (w,b)
4 294 vote democrats red voting trump MAGA E
campaign wow maga know migrants BLEXIT RD (we)
trump president obama democrats america VoteRed P
5 274 vote trump president america florida Iran E
democrats family brownley senate women california RedNationRising M (pro)
shawn company insurance stopped program MAGA I
6 205 good time woman high child TaxationIsTheft PD
read sense world better vegans DonLemon
people socialism hate free marxist mises
7 177 people online love media god VerifiedHate M (pro)
gab speech free hate left Update
world getting think read gt Gab
8 160 just think know trump middle EnemyOfThePeople I
trump media job usa hit wages M
immigration wall country border asylum manufacturing
9 119 starting rain evidently mustang sins jojo_anime Anime
halloween anime game minmod mini azurlane
eat bad hey bro crunch NoTatsukiNoTanoshii
10 115 trump obama white caravan migrant TheTruthCommunity I
trump president caravan migrant midterm FaroeIslands E
democrats iran midterms sanctions kavanaugh Anonymous RD(ai)
11 99 stop country home triangle lemon Halloween2018 I
trump illegal beto cnn caravan texas HappyHalloween2018 E
border invasion campaign southern illegal MAGA
12 67 white hatred sick wshis girls WhiteGenocide RD (we)
diversity itsokaytobewhite defund whitepeople word
code StandUpForEurope I
migrants refugees europe eu border WhiteLivesMatter
13 64 memes hillary tweet trump black MAGA2018 E
timeline country economy train book BTFOMobRule P
voteredtosaveamerica hivemind critical marxist theory
memewarmidterms
14 63 trump campaign candidate gillum navy CNNCredibilityCrisis E
rofl voting sundaymorning mondaymotivation thewalk-
ingdead ElectionDay M (anti)
cnn vote obama gillum rally ElectionDay2018
15 4 american stones demons boy bad church CavalierNationalism Religion
gt christ right blessed strates WhiteHonorKilllings
protestant fascism authoritarian mary collectivist
Table 2: Communities detected before the Election Day and the general topics on which their users have posted
messages.
6
APREPRINT - JANUARY 30, 2020
Id. N.U. Top 3 Topics Top 3 Hashtags Labels
1 784 election county tucker broward antifa MAGA E
white america black live press Editorial I
democrat gillum breaking party florida Broward
2 758 county ballots election votes florida StopTheSteal E
trump acosta white states jim ElectionNight RD (we)
year boom years women stop Broward P
3 571 trump president just vote got StopTheSteal E
broward ballots county florida fraud snipes BrowardCounty M (anti)
white people election democrats steal MAGA RD (w,b)
4 256 time day men fam long WoolseyFire RD (we)
people good don like white electionnight
right new fuck bad community ff
5 215 streaming company home muslim think Killstream M (pro,anti)
white pride thread beagle deadly WSJKillsKids
sargon really youtube just people BREAKING
6 178 socialism marxist nationalism party british MeToo PD
men group make things word AsiaBibi
western christian asiabibi asylum british JimAcosta
7 157 wow evangelion taught looks whoah DELTARUNE Anime
real needs energy avi world
time undertale aha know kinda
8 149 white women vote yes god ElectionDay E
trump election republican ballots vote Florida P
broward county tucker twitter scott Broward RD (we)
9 144 followers help million raise communist StopTheSteal E
watch just media trump kid TaxationIsTheft RD (as)
nationalism broward hitler ballots county Nationalism
10 104 white people problem asylum facebook migrants WhiteGenocide RD (we)
defund discrimination whites sense StandUpForEurope I
gab migrant media free right facebook WhiteLivesMatter M(pro)
11 100 border entry ports story italy 2A I
white fact german ohio language FoxNews
trump european imagine language jewish France
12 96 trump stop american democrat abortion StopTheSteal E
macron women michelle football saudi USMC Abortion
florida election county broward fraud MarineCorpsBirthday
13 95 sounds fuck shit cultural florida VeteransDay2018 E
trump great non cruz stop Nature
got wonder family black white MAHw1d5
14 85 marine love saved children television Iran Military
god bless noticed lady today Veteran M (anti)
press jim acosta stopthesteal trump MAGA
15 63 retired proud crypto jeff major Blockchain Technology
president ai stage crypto blockchain AI
tonight broward skating sponsorship paccoin WashingtonElite
16 28 news talking fake weird fbpe TwoMinuteSilence M (pro)
british little ffs great sure RemembranceDay2018
just white good make ve video ArmisticeDay100
17 10 antifa pay polish fuck jewish MeToo RD (as)
wanking media deficit social migrant PolishIndependenceDayMarch
trump voters punish jobs black jerusalem SouthAfrica
Table 3: Communities detected after the Election Day and the general topics on which their users have posted messages.
7
APREPRINT - JANUARY 30, 2020
Figure 2: On the left, communities detected before the election day. On the right, communities detected after the
election day. Each community is labeled with their topics category, the underscored text represents main topic categories.
An arrow joining two community on the legend indicates the transference of a community between snapshots.
Regarding the communities, it was found that none maintained the same structure of users over both weeks. Some of
the relevant users remained in the same communities from one week to another (table 2), but they did not represent
more than the 20-40% on average of the previous relevant users. Therefore, it was easier to interpret communities using
the topic they discussed, than by the users that were included. If we compare the topics covered by the communities
and their categorization, we can find that some of the communities did maintain over time. A perfect example would be
the community using white supremacist hashtags (community 12 pre-elections, 10 post-elections), or community 6,
whcih discussed the economy and general politics in both weeks. Therefore, while communities of users did change
during the data collection period, similar topics were discussed in many of these groups.
Indeed, the topics that our data identifies were commonly discussed reflected earlier research on the alt-right. For
example, anti-immigrant sentiment [
3
,
2
,
4
] and racially derogatory comments [
5
] were common. Furthermore, hashtags
such as #WhiteGenocide and #WhiteLivesMatter were used frequently in some communities, and these hashtags are
strongly associated with white supremacy. As to be expected, discussion of current political news stories was frequent,
with this frequently mirroring right-wing media talking points. Similarly, criticism of traditional news media (such as
CNN and The Wall Street Journal) was detected, with this often mirroring President Trump’s denunciations of the press.
In contrast, "alternative" media sources were frequently shared or discussed, such as from Kill Stream and Red Nation
Rising, with this reflecting other research on the alt-right [4].
Interestingly, our results indicate that anime and technology were discussed by many alt-right supporters. These findings
provide support to other literature which has linked internet subcultures to the growth of the alt-right. For example, in
Kill All Normies, Angela Nagle navigates the link between #GamerGate, 4Chan and even sections of anime fans with
the rise of the alt-right. Similarly, in Mike Wendling’s book on the alt-right, he agrees with this assessment and also
connects proponents of technologies like cryptocurrencies with the movement. The results of this study provides further
evidence of this link between the alt-right and certain sections of anime fans and cryptocurrency proponents.
The results of this study also indicate that there is a small section of alt-right supporters who commonly discuss
religion. While alt-right supporters are typically hostile towards Judaism and Islam, the movement cannot be broadly
characterised as Christian, as a large contingent appear to be irreligious [
5
]. Still, results here identified a small
community of users where religion was discussed before the election and another where abortion was a central topic
8
APREPRINT - JANUARY 30, 2020
after the midterm elections. This is perhaps evidence of the small, paleoconservative element which is aligned with the
alt-right, where opposition to abortion is a strongly held belief [5].
Considering the similarity on the topics, the increased relevance of elections on the second week and the increasing on
the tweeting rate until the day of the elections, we can suggest that the Alt-Right communicational activity grows when
a political event is near. However, a study picking longer dates or making a deeper assessment on the daily evolution of
the communities could help to confirm this hypothesis. Also, it could be useful to determine why communities are
quite different, how and when do they break into different groups, and why the topics covered by this communities are
maintained.
Overall, this research demonstrated that alt-right supporters were more active in the days immediately surrounding
the 2018 US midterm elections. Furthermore, this study identified a number of sub-communities within the broader
alt-right, that gathered around certain topics. Similar to previous literature on the alt-right, anti-immigrant sentiment,
racism towards Jews and Muslims, as well as support for white supremacy, were all frequently reflected in the results.
Also supporting earlier literature, our results identified smaller, yet still notable segments of the alt-right sample used in
this study discussing anime, technology and religion.
Acknowledgements
This work has been supported by several research grants: Spanish Ministry of Science and Education under TIN2014-
56494-C4-4-P grant (DeepBio) and Comunidad Autónoma de Madrid under S2013/ICE-3095 grant (CYNAMON).
References
[1]
Roderick Graham. Inter-ideological mingling: White extremist ideology entering the mainstream on twitter.
Sociological Spectrum, 36(1):24–36, 2016.
[2]
Ramona Kreis. # refugeesnotwelcome: Anti-refugee discourse on twitter. Discourse & Communication, 11(5):498–
514, 2017.
[3]
Thomas Greven. The rise of right-wing populism in europe and the united states. A Comparative Perspective.
Friedrich Ebert Foundation, Washington DC Office, 2016.
[4]
Matthew N Lyons. Ctrl-alt-delete: The origins and ideology of the alternative right. Somerville, MA: Political
Research Associates, January, 20, 2017.
[5] George Hawley. Making sense of the alt-right. Columbia University Press, 2017.
[6]
Joshua Thorburn, Javier Torregrosa, and Ángel Panizo. Measuring extremism: Validating an alt-right twitter
accounts dataset. In International Conference on Intelligent Data Engineering and Automated Learning, pages
9–14. Springer, 2018.
[7]
Tanner Mirrlees. The alt-right’s discourse on" cultural marxism": A political instrument of intersectional hate.
Atlantis: critical studies in gender, culture & social justice, 39(1):49–69, 2018.
[8] Patrick S Forscher and Nour Kteily. A psychological profile of the alt-right. 2017.
[9]
Joel Finkelstein, Savvas Zannettou, Barry Bradlyn, and Jeremy Blackburn. A quantitative approach to understand-
ing online antisemitism. arXiv preprint arXiv:1809.01644, 2018.
[10]
Savvas Zannettou, Barry Bradlyn, Emiliano De Cristofaro, Haewoon Kwak, Michael Sirivianos, Gianluca Stringini,
and Jeremy Blackburn. What is gab: A bastion of free speech or an alt-right echo chamber. In Companion
Proceedings of the The Web Conference 2018, pages 1007–1014. International World Wide Web Conferences
Steering Committee, 2018.
[11]
Angela Nagle. Kill all normies: Online culture wars from 4chan and Tumblr to Trump and the alt-right. John
Hunt Publishing, 2017.
[12]
Gabriel Emile Hine, Jeremiah Onaolapo, Emiliano De Cristofaro, Nicolas Kourtellis, Ilias Leontiadis, Riginos
Samaras, Gianluca Stringhini, and Jeremy Blackburn. Kek, cucks, and god emperor trump: A measurement study
of 4chan’s politically incorrect forum and its effects on the web. In Eleventh International AAAI Conference on
Web and Social Media, 2017.
[13]
Javier Torregrosa, Joshua Thorburn, Raúl Lara-Cabrera, David Camacho, and Humberto M Trujillo. Linguistic
analysis of pro-isis users on twitter. Behavioral Sciences of Terrorism and Political Aggression, pages 1–15, 2019.
9
APREPRINT - JANUARY 30, 2020
[14]
Andranik Tumasjan, Timm O Sprenger, Philipp G Sandner, and Isabell M Welpe. Predicting elections with twitter:
What 140 characters reveal about political sentiment. In Fourth international AAAI conference on weblogs and
social media, 2010.
[15]
Albert Feller, Matthias Kuhnert, Timm O Sprenger, and Isabell M Welpe. Divided they tweet: The network
structure of political microbloggers and discussion topics. In Fifth International AAAI Conference on Weblogs
and Social Media, 2011.
[16]
Didi Surian, Dat Quoc Nguyen, Georgina Kennedy, Mark Johnson, Enrico Coiera, and Adam G Dunn. Character-
izing twitter discussions about hpv vaccines using topic modeling and community detection. Journal of medical
Internet research, 18(8):e232, 2016.
[17]
Zhijun Yin, Liangliang Cao, Quanquan Gu, and Jiawei Han. Latent community topic analysis: Integration of
community discovery with topic modeling. ACM Transactions on Intelligent Systems and Technology (TIST),
3(4):63, 2012.
[18]
Aaron Clauset, Mark EJ Newman, and Cristopher Moore. Finding community structure in very large networks.
Physical review E, 70(6):066111, 2004.
[19]
M E Newman. Modularity and community structure in networks. Proc Natl Acad Sci U S A, 103(23):8577–8582,
June 2006.
[20]
David M Blei, Andrew Y Ng, and Michael I Jordan. Latent dirichlet allocation. Journal of machine Learning
research, 3(Jan):993–1022, 2003.
10
