=Paper=
{{Paper
|id=Vol-1395/paper_12
|storemode=property
|title=To Be or Not to Be Charlie: Twitter Hashtags as a Discourse and Counter-discourse in the Aftermath of the 2015 Charlie Hebdo Shooting in France
|pdfUrl=https://ceur-ws.org/Vol-1395/paper_12.pdf
|volume=Vol-1395
|dblpUrl=https://dblp.org/rec/conf/msm/GigliettoL15
}}
==To Be or Not to Be Charlie: Twitter Hashtags as a Discourse and Counter-discourse in the Aftermath of the 2015 Charlie Hebdo Shooting in France==
https://ceur-ws.org/Vol-1395/paper_12.pdf
To Be or Not to Be Charlie: Twitter Hashtags as a
Discourse and Counter-discourse in the Aftermath of the
2015 Charlie Hebdo Shooting in France
Fabio Giglietto Yenn Lee
DISCUM, Università di Urbino Carlo Bo SOAS, University of London
ITALY UK
fabio.giglietto@uniurb.it yl22@soas.ac.uk
ABSTRACT portrayals of the Prophet Mohammed. Within hours following the
attack, the hashtag #JeSuisCharlie [I am Charlie] began trending
Following a shooting attack by two self-proclaimed Islamist
on Twitter, in a show of condolences for the victims, solidarity,
gunmen at the offices of French satirical weekly Charlie Hebdo
and support for the magazine’s right to satirise any subject
on 7th January 2015, there emerged the hashtag #JeSuisCharlie
including religions. Reportedly created by an artist named
on Twitter as an expression of condolences for the victims,
Joachim Roncin, who lived in the neighborhood of the shooting
solidarity, and support for the magazine’s right to free speech.
site, the hashtag was used over five million times by 9th January
A l m o s t s i m u l t a n e o u s l y, h o w e v e r, t h e r e w a s a l s o
and became one of the most repeated news-related hashtags in
#JeNeSuisPasCharlie explicitly countering the former,
Twitter’s history [22]. In the initiator’s own words, ‘je’ in this
affirmative hashtag. In this paper, we analyse 74,047 tweets
context was important as it offered a vehicle through which each
containing #JeNeSuisPasCharlie posted between 7th and 11th
individual expressed themselves vis-à-vis threats to the freedom
January. Our network analysis and semantic cluster analysis of
and tolerance underpinning the participants’ world (Roncin,
those 74,047 tweets reveal that the hashtag in question
interviewed by Sky News, 2015). ‘Je Suis Charlie’ (and by
constituted a form of resistance to the mainstream framing of the
extension ‘Nous Sommes Tous Charlie’ [We are all Charlie]) also
issue as freedom of expression being threatened by religious
served as the principal slogan during the vigils and marches that
intolerance and violence. The resistance was manifested through
took place in central Paris on Sunday 11th January.
three phases: sharing condolences but indicating a reservation
against the mainstream frame (Grief); voicing out resistance However, there too emerged #JeNeSuisPasCharlie [I am not
against the frame (Resistance); and developing and deploying Charlie], explicitly countering the former, affirmative hashtag.
alternative frames such as hate speech, Eurocentrism, and Since the former hashtag entailed a tragedy of twelve deaths and
Islamophobia (Alternatives). The hashtag in this context served support for the universal value of freedom of expression,
as a vehicle through which users formed, enhanced, and declared #JeNeSuisPasCharlie carried an inherent risk of being viewed as
their self-identity. opposing accepted social norms. Despite the risk, the negative
hashtag was used more than 74,000 times over the next few days
Categories and Subject Descriptors since 7th January. Against this backdrop, we set out to unpack a
J.4 [Social and behavioral sciences]: Sociology. complex relationship between the willingness to speak up on
sensitive topics and identity formation on Twitter. More
General Terms specifically, we aim to address three interlinked questions as
Human Factors. below.
1. What are the characteristics of the network formed
Keywords around the #JeNeSuisPasCharlie hashtag and the
counter-discourse, freedom of expression, hashtag, identity, material shared through that network on Twitter?
semantic cluster analysis 2. How did users of the #JeNeSuisPasCharlie hashtag
position themselves discursively with regard to the
1.INTRODUCTION #JeSuisCharlie hashtag?
On 7th January 2015, two gunmen forced their way into and
opened fire in the headquarters of satirical weekly magazine 3. How did the activities under the #JeNeSuisPasCharlie
Charlie Hebdo in Paris, killing twelve staff cartoonists and hashtag evolve as the broader public discussion of the
claiming that it was an act of revenge against the magazine’s shooting attack developed?
2.LITERATURE REVIEW
In order to address the research questions above, the present
study draws upon a combination of three strands of work in the
Copyright c 2015 held by author(s)/owner(s); copying permitted
current scholarship: the network characteristics of Twitter-
only for private and academic purposes. mediated discussion; the roles of hashtags in such discussion; and
Published as part of the #Microposts2015 Workshop proceedings, the expressions of identity in social media activism. First, recent
available online as CEUR Vol-1395 (http://ceur-ws.org/Vol-1395) years have seen a fast-growing body of literature concerned with
#Microposts2015, May 18th, 2015, Florence, Italy. buzzing discussions on the microblogging platform Twitter and
how to examine them systematically. Given the range and amount
of data that researchers could mine from the platform, a keen
· #Microposts2015 · 5th Workshop on Making Sense of Microposts · @WWW2015
interest has been shown in employing network-analysis during one TV programme [1] or a series [7]. More relevantly to
approaches for a ‘bird’s eye view’. Himelboim and Han [10] the purposes of the present study, identity may refer to
argued, through their case study of cancer-related discussion on individuality that used to be blended and lost in the presence of
Twitter, that communities emerged from such discussion with the collectivity required in activism in the pre-social media era
clusters of interconnected users and the information sources on [18].
which they relied most. A 2014 special issue of American
Behavioral Scientist, particularly the contributions by Dubois and 3.METHODOLOGY
Gaffney [5] and Xu et al. [23], showed that opinion leaders and Our dataset consisted of 74,074 tweets containing the hashtag
influencers could be metrically identified in Twitter-mediated #JeNeSuisPasCharlie and published by 41,687 unique users
political discussions. The links formed between political between 7th and 11th of January 2015. Due to the known limits
discussants on Twitter turned out to be considerably different of Twitter free API [17], the data was purchased from Sifter, a
from those observed in the Web 1.0 environment or in web application that provides, in partnership with Gnip, search-
blogosphere, at least in the South Korean context, according to and-retrieve access to every undeleted tweet in the history of
Hsu and Park [11]. Mapping the landscape of Twitter activity has Twitter. The data gathered via Sifter was automatically imported
provided unique insights into various issues of international into a new DiscoverText project. It was then exported in CSV
relevance. Lotan’s study of the 2014 Israel-Gaza conflict [12], for format from there and was analysed using R.
example, visually demonstrated a distinct polarisation between
the pro-Israel and pro-Palestine sides with a negligible number of 3.1.Typology of contents and network
bridging actors in-between. By tracing the Twitter network of The first tweet in the dataset was dated 7th January 2015, 1:46
Western-origin Jihad fighters, Klausen [14] identified that certain PM in local time. The hashtag #JeSuisCharlie was reported to be
strategic roles were assigned to those fighters’ Twitter accounts. created at 12:59 PM on the same day, immediately following the
Discussions on Twitter are speedy and unstructured and, shooting that took place at around 11:30 AM. Tweets in our
consequently, the organisational usefulness of hashtags has dataset were written in various languages. Using the text
attracted practical as well as academic attention. Bruns [3] categorisation engine based on n-grams provided by the textcat R
detailed out his methodological experiences and reflections of package [6], we discovered that French (30%), English (25%)
handling Twitter data around a hashtag and highlighted that and Spanish (12%) accounted for the majority of the tweets. It
hashtags are ‘shared conversation markers’, which require users was unsurprising that French was the most frequently used
to include them in their posts deliberately if they wish to take part language, but the proportion was smaller than expected,
in established conversations. Based on a comparison of various indicating its reference to #JeSuisCharlie. Another interesting
hashtag-based communications, Bruns and Stieglitz [4] characteristic identified was that 1,488 tweets (2%) were made of
concluded that different hashtags are associated with different nothing but the #JeNeSuisPasCharlie hashtag. 70% of the 74,074
patterns of user behaviours. While crisis- and emergency-related tweets were retweets and 41% included URLs. Since retweets
hashtags (such as #tsunami for the March 2011 tsunami in Japan account for almost three quarters of the dataset, we computed and
and #londonriots in 2011) have seen a dominant proportion of visualised a retweet network with a view to identifying central
retweets and URLs pointing outside Twitter, spectacle-oriented users and their clusters if any. We also identified the most
hashtags (such as British #royalwedding in 2011 and #eurovision recurring external sources (URLs).
for the Eurovision Song Contest in 2011) seem to elicit more 3.2.Topics
original tweets from users. Indeed, such findings from hashtag
In order to understand the main topics addressed, we applied the
studies are in line with the studies focusing on unravelling the
text mining techniques provided by the textcat R package [16] to
network properties of Twitter communications discussed earlier.
the textual corpus of all tweets in the dataset. We lowered the
Siapera’s work on #Palestine [19] and Lorentzen’s work on
case of all terms in the corpus and cleaned it up by removing
#svpol (for Swedish politics) [8], for example, point to
auxiliary words in French, English and Spanish, as well as
homophily and polarisation in hashtag-based discussions,
punctuation marks and whitespaces. Additionally, we also
resonating Lotan’s findings cited above.
removed ‘jenesuispascharlie’, ‘charlie’, ‘charliehebdo’, ‘hebdo’,
However enthusiastic the participants in Twitter-mediated ‘jesuischarlie’ and created a document term matrix to calculate
political discussions may be, whether their participations lead to the associations between the remaining words (N=36,030). After
any concrete outcomes is still an ongoing question. On the one removing sparse terms (i.e. the sparsity of a term is defined as the
hand, some offer encouraging anecdotes of how Twitter has percentage of documents with 0 occurrence; in the present study
facilitated protests in different parts of the world, such as one a term was removed if its sparsity was higher than 98%), we
against police brutality in Ferguson in Missouri, US, in 2014 [9]. identified the most frequently used terms (N=17) and their
A cautious voice, on the other hand, is that Twitter and other such Euclidean distances, and created clusters of frequently co-
platforms make social movements ‘easier to organise but harder occurring terms.
to win’ by pushing them to scale up before they are ready for it
[21]. Nevertheless, what social media including Twitter can 3.3.Evolution over time
certainly provide is a space for accommodating expressions of To better understand the evolution of the topics discussed, with
identity at multiple layers. Bennett and Segerberg [2] suggested particular reference to our third research question, we created a
that, in today’s large-scale ‘connective action’ (in distinction to by-minute time series (N=6,444, AVG TPM=11.5) of activity. We
the traditional concept of ‘collective action’), political content is also used the Breakout Detection R package, which had recently
often presented in the form of easily personalised ideas such as been open-sourced by Twitter [13], to identify breakouts or shifts
‘Put People First’ (PPF) during the 2009 G20 London summit in the mean of tweet per minute (TPM).
protests or ‘We Are the 99 Percent’ during the Occupy Wall Street
movement in the US in 2011. According to the two authors, these
personal action frames are particularly inclusive and can be easily
passed across different platforms. ‘Identity’ here can be a
collective identity expressed within a limited time span like
34
· #Microposts2015 · 5th Workshop on Making Sense of Microposts · @WWW2015
chart, the case of #JeNeSuisPasCharlie is noticeably closer to the
second cluster characterised by more retweets and more
inclusions of URLs (Figure 2).
!
Figure 1. Twitter by-minute activity on the hashtag (dashed
lines indicate a breakout)
The Breakouts tool (used with the following parameters:
min.size=5, method=’multi’, beta=.001, degree=1, percent=0.25)
detected 14 breakouts (Figure 1), out of which it identified three !
moments of high user engagement (Table 1). Figure 2. User’s activity patterns comparing different Twitter
hashtags (size indicates total number of contributor)
Table 1. Moments of high user engagement A closer analysis of retweets (Table 3) and URLs provided more
insights into the nature of #JeNeSuisPasCharlie hashtag.
AVG
from to tweets rt @replies Table 3. Top 5 most retweeted posts
TPM
07/01 18:07 07/01 23:44 9,194 7,392 150 50.00 User Text of the tweet N
08/01 11:42 08/01 23:37 16,048 11,688 472 23.56 Les dessins du dessinateur brésilien
Carlos Latuff #JeNeSuisPasCharlie
khurramabad0 #Charlie_Hebdo #islamophobie http:// 1,785
09/01 11:55 10/01 00:44 10,159 6,899 465 13.57
t.co/a6qrL6pdPt
Last August, The Sydney Morning
Finally, on each subset of tweets created during one of the three Herald was forced to remove, apologize
RanaHarbi 868
moments, we calculated, using the same procedure applied to the for this #JeNeSuisPasCharlie
entire dataset, a document term matrix of the most frequently #JeSuisAhmed http://t.co/O7zASRLpD1
used terms. We then grouped those terms according to their co-
Pr moi ce n’est pas Charlie Hebdo qui
occurrences. est mort mais 2 policiers et des
CoraaantinM 794
journalistes. L’hommage est à eux, pas
Table 2. Moments of high user engagement au journal #JeNeSuisPasCharlie
A cartoonist with integrity & intellectual
from terms Max sparsity Most frequent terms consistency – Joe Sacco on Charlie
MaxBlumenthal 774
Hebdo #JeNeSuisPasCharlie http://t.co/
07/01 18:07 5,009 96% 29
5uIRwE2wIu
08/01 11:42 11,327 96% 22
Bizarrement quand je dis
09/01 11:55 9,735 95% 27 #JeNeSuisPasCharlie on m’insulte mais
SinanLeTurc quand Charlie insulte notre prophète ça 729
devient de la liberté d’expression.
4.DISCUSSION OF ANALYTIC FINDINGS
Adopting the methods suggested in Bruns and Stieglitz’s study In the aftermath of the shooting, many well-known cartoonists
[4], we used two standard Twitter metrics (i.e. ratio between expressed their condolences and solidarity for Charlie Hebdo by
retweets and tweets and ratio between tweets with URLs over all displaying tribute drawings [20]. Two of the most frequently shared
tweets) to compare #JeNeSuisPasCharlie with other previously tweets in our dataset also contained links to drawings, but in this
studied hashtags. As also discussed in the Literature Review case one by the Arab Brazilian freelance political cartoonist Carlos
section, Bruns and Stieglitz observed the emergence of two Latuff and another by the Maltese–American cartoonist and
clearly distinct clusters: media events (e.g. #royalwedding, journalist Joe Sacco. The two drawings represented a take on the
#eurovision) and crisis/emergency events (e.g. #tsunami, incident that was different from the one put forward by the
#qldflood, #londondriots). In the former case, original tweets are mainstream community of cartoonists in response to the tragedy of
common and URLs are mainly used to share further stories about their colleagues at Charlie Hebdo. Both Latuff and Sacco pointed
the media events at hand. In the latter case, during an urgent out that the magazine had been publishing, in the name of the
situation, it is more important to share vital information such as freedom of speech, images often considered to be offensive for the
emergency numbers; hence, a characteristically high proportion Muslim population and that the same concept of freedom of speech
of retweets and URLs were observed. When mapped on the same had not been invoked in the case of an anti-Semitic satire earlier.
35
· #Microposts2015 · 5th Workshop on Making Sense of Microposts · @WWW2015
!
Figure 3. Most frequently used words and their association across the three main phases
Along the same line, another heavily retweeted message recalled English words) were prominent in all three moments, confirming
the story of Australian newspaper The Sydney Morning Herald that the freedom of expression and its contested limits were the
[15] being forced to issue an apology and remove a drawing that real leitmotif across the entire dataset. Terms such as racism and
was considered anti-Semitic. This tweet also included the hashtag racist stood out in the second and third moments since users of
#JeSuisAhmed, with reference to a Muslim police officer, Ahmed #JeNeSuisPasCharlie started to approach Charlie Hebdo’s satires
Merabet, also killed during the Charlie Hebdo attack. Many from different angles than free speech.
Twitter users indeed joined the #JeSuisAhmed hashtag.
According to Topsy, it was used over 150,000 times in the days 5.CONCLUSION
following the attack in a show of condolences for all victims of Using a combination of various quantitative techniques, the
the shooting. present study explored the structure of the discussion around the
The most frequently shared external sources (URLs) were all #JeNeSuisPasCharlie hashtag. First, the discussion had a high
images. Links pointing to news sites were rare. This is because proportion of retweets (70%) and URLs (41%). Compared to
#JeNeSuisPasCharlie was not about the news. It’s primarily goal some previously studied hashtags, #JeNeSuisPasCharlie behaved
was instead to mark and declare an identity by distinction. To that more like crisis/emergency hashtags than media spectacle
end, 2% of the retrieved tweets were made up of nothing but the hashtags. That said, our analytic results also highlighted the
hashtag. heterogeneity of the viewpoints and arguments aggregated under
the hashtag in question. Users of the said hashtag showed
As mentioned in the previous section, the first tweet with resistance to the mainstream framing of the Charlie Hebdo
#JeNeSuisPasCharlie was published less than an hour after what shooting as the universal value of freedom of expression being
was reported as the first tweet containing #JeSuisCharlie. While threatened by religious intolerance and violence. In this context,
the hashtag started as an immediate reaction to #JeSuisCharlie, retweeting something that would justify their resistance was a
nevertheless, its nature changed over time. way of marking their identity as distinct from what was accepted
The Breakout Detection tool developed by Twitter engineers in the mainstream. Given the sensitivity of the subject, such
helped us identify three moments of higher user engagement retweets also helped the users protect themselves from the risk of
(Table 2). Besides the words related to the most retweeted posts being viewed as endorsing the violence. We also observed a
(such as Latuff’s cartoon and the Sydney Morning Herald case) unique practice of tweeting nothing but the hashtag, amounting to
discussed above, there are a few noteworthy dynamics in Figure 2% of the dataset. This is a strategy that can be explained in a
3. First, the clusters of words including désolé [sorry] (N=388), similar vein.
familles (N=564), victims (N=628), and compatis [sympathise]
Over time, there were three distinguished phases in the
(N=409) were present in the first dendrogram but not in the
manifestation of this resistance: Grief (i.e. joining the mourning
following two. Liberté and expression (and their corresponding
36
· #Microposts2015 · 5th Workshop on Making Sense of Microposts · @WWW2015
for the victims of the attack but indicating a reservation against hashtag-activism-why-social-media-matters-to-protestors/
the proposed frame); Resistance (i.e. starting to voice out the 384215/. Accessed: 2015-02-07.
resistance); and Alternatives (i.e. fully developing and deploying [10] Himelboim, I. and Han, J.Y. 2014. Cancer talk on twitter:
alternative frames). In this study, the hashtag was not a community structure and information sources in breast and
conversation marker as previous studies identified but a prostate cancer social networks. Journal of health
discursive device that facilitated users to form, enhance, and communication. 19, 2 (2014), 210–225.
strategically declare their self-identity. [11] Hsu, C.-L. and Park, H.W. 2010. Sociology of Hyperlink
Networks of Web 1.0, Web 2.0, and Twitter: A Case Study
Our quantitatively oriented methodology here allowed us to of South Korea. Social science computer review. (Sep.
identify the topical and network structure of the discussion 2010).
around #JeNeSuisPasCharlie and its evolution over time. We also [12] Israel, Gaza, War & Data: 2014. https://medium.com/i-
suggest as an avenue for further research to delve more data/israel-gaza-war-data-a54969aeb23e. Accessed:
qualitatively into the ways in which individual users coped with 2015-02-07.
the sensitive nature of the issue at hand and challenged the [13] James, N.A. et al. 2014. BreakoutDetection: Breakout
mainstream perspective. Detection via Robust E-Statistics.
[14] Klausen, J. 2015. Tweeting the Jihad: Social Media
6. REFERENCES Networks of Western Foreign Fighters in Syria and Iraq.
Studies in Conflict and Terrorism. 38, 1 (2015), 1–22.
[1] Anstead, N. and O’Loughlin, B. 2011. The Emerging [15] Meade, A. 2015. SMH cartoon criticised as antisemitic
Viewertariat and BBC Question Time: Television Debate found to breach press council standards. The Guardian.
and Real-Time Commenting Online. The International [16] Meyer, D. et al. 2008. Text Mining Infrastructure in R.
Journal of Press/Politics. 16, 4 (Jul. 2011), 440–462. Journal of statistical software. 25, 5 (Mar. 2008), 1–54.
[2] Bennett, W.L. and Segerberg, A. 2012. THE LOGIC OF [17] Morstatter, F. et al. 2013. Is the Sample Good Enough?
CONNECTIVE ACTION. Information, Communication Comparing Data from Twitter’s Streaming API with
and Society. 15, 5 (2012), 739–768. Twitter's Firehose. Seventh International AAAI Conference
[3] Bruns, A. 2012. HOW LONG IS A TWEET? MAPPING on Weblogs and Social Media (Jun. 2013).
DYNAMIC CONVERSATION NETWORKS ON [18] Sauter, M. 2014. The Coming Swarm: DDOS Actions,
TWITTER USING GAWK AND GEPHI. Information, Hacktivism, and Civil Disobedience on the Internet.
Communication and Society. 15, 9 (2012), 1323–1351. Bloomsbury Academic.
[4] Bruns, A. and Stieglitz, S. 2013. Towards more systematic [19] Siapera, E. 2014. Tweeting #Palestine: Twitter and the
Twitter analysis: metrics for tweeting activities. mediation of Palestine. International Journal of Cultural
International journal of social research methodology. 16, 2 Studies. 17, 6 (Nov. 2014), 539–555.
(Mar. 2013), 91–108. [20] Telegraph, T.D. 2015. Cartoonists show solidarity after
[5] Dubois, E. and Gaffney, D. 2014. The Multiple Facets of Paris Charlie Hebdo attack, in pictures. The Daily
Influence: Identifying Political Influentials and Opinion Telegraph.
Leaders on Twitter. The American behavioral scientist. 58, [21] Tufekci, Z. 2014. Online social change: easy to organize,
10 (Sep. 2014), 1260–1277. hard to win [Video file]. https://www.ted.com/talks/
[6] Feinerer, I. et al. 2013. The textcat Package for n-Gram zeynep_tufekci_how_the_internet_has_made_social_change
Based Text Categorization in R. Journal of statistical _easy_to_organize_hard_to_win.. Accessed: 2015-02-07.
software. 52, 6 (Feb. 2013), 1–17. [22] Wendling, M. 2015. #JeSuisCharlie creator: Phrase cannot
[7] Giglietto, F. and Selva, D. 2014. Second Screen and be a trademark. BBC Trending.
Participation: A Content Analysis on a Full Season Dataset [23] Xu, W.W. et al. 2014. Predicting Opinion Leaders in
of Tweets. The Journal of communication. 65, 2 (2014). Twitter Activism Networks: The Case of the Wisconsin
[8] Gunnarsson Lorentzen, D. 2014. Polarisation in political Recall Election. The American behavioural scientist. (Mar.
Twitter conversations. Aslib Journal of Information 2014).
Management. 66, 3 (2014), 329–341.
[9] Hashtag Activism Isn’t a Cop-Out: 2015. http://
www.theatlantic.com/politics/archive/2015/01/not-just-
37
· #Microposts2015 · 5th Workshop on Making Sense of Microposts · @WWW2015