New approaches to the propagation of the antifeminist backlash on Twitter

A year after the #MeToo movement erupted, antifeminism started to retort. The idea behind the backlash was that ‘the men who have been accused are the heroes’ (Tolentino, 2018). Twitter was one of the public spaces where this confrontation occurred; the #HimToo backlash gathered steam in 2018 (Asimov, 2018) and expanded into 2019. Focussing on the reactions against #Cuéntalo Twitter campaign –the Spanish equivalent to #MeToo—, this article examines how the antifeminist backlash proliferates, offering a view of the dynamics driving it. The authors choose an eclectic and interdisciplinary approach that integrates graph theory and CDA (Critical Discourse Analysis), thus connecting social media formulations with offline discourses, and proposing new ways of studying social movements. To examine the backlash’s characteristics, this study utilises Congosto’s typology of Twitter profiles (Congosto, 2018). Findings suggest that both approaches are complementary and necessary, for while graph analysis enables the distinction of antifeminist communities on Twitter and their behaviour and characteristics, CDA allows investigators to uncover their discursive strategies and favoured themes.


Introduction
A year after the #MeToo movement erupted, some people and their bots -autonomous programmes designed to interact as users-started to push back in social media platforms. The principle behind the backlash was that 1 m.gutierrez@deusto.es Universidad de Deusto, España 2 mpando@deusto.es Universidad de Deusto, España 'the men who have been accused are the heroes' (Tolentino, 2018). Others admonished feminists by claiming that harassers can also be women (Segall, 2018). Men decided that they could no longer be silent. #HimToo became part of the 'tweet frenzy', inspiring 'a full-on hashtag-slinging countermovement' (Ellis, 2018).
The #HimToo backlash became active after the #MeToo movement turned into a rolling wave in 2018 (Asimov, 2018). However, this countermovement is not a way of overcorrecting a movement -#MeToo-that had gone too far; it is a reaction to its mere existence (Tolentino, 2018). Faludi noted in 1991 that 'the antifeminist backlash has been set off not by women's achievement of full equality but by the increased possibility that they might win it' (Faludi, 2006, 11). Almost thirty years later, feminism has not achieved equality for women. Backlashers resurface any time these facts are put on the table. The backlash is recurrent and pre-emptive, but unlike previous backlashes, it now occurs on digital social platforms where algorithms disseminate messages massively and in real-time.
Both #MeToo and # HimToo are opposite movements. By definition, antifeminism is a counter-movement surfacing in antagonism with feminism. 'Antifeminists oppose women's entry into the public sphere, the reorganisation of the private sphere, women's control of their bodies, and women's rights generally' (Kimmel & Aronson, 2003).
Antifeminism arose after the first feminist demands were voiced in the 19 th century. Women's struggle for equality gathered steam in the mid-19th century, the 1900s, 1940s and 1970s, relinquishing to backlash every time (Faludi, 2006, 63). News media played a role in supporting the antifeminist recoil; in the 80s and 90s, they published studies to reassure men (Faludi, 2006, 63), reinforcing antifeminist claims that social arrangements among men and women are natural and do not favour men (Clatterbaugh, 2003).
Today antifeminists rely on subtle messages that rebel against the mainstream, institutionalised feminism, political correctness that favours women and 'feminazism' -a portmanteau of 'feminism' and 'Nazism' referring to 'women who shut down their opponents with authoritarian orthodoxies' (Williams, 2015). Backlashers are self-fashioned as liberal feminists and soft right-wing (Lewis, 2019). Whereas in the past antifeminists seemed radical and crude, today they are likeable, to the point that antifeminists are making right-wing stances 'socially acceptable' (Schutzbach, 2019). While extremists' messages are effortlessly dismantled, this antifeminism is challenging to recognise.
Not much has been said about the antifeminist backlash in Spain. Focusing on the reaction against the #Cuéntalo (#Tellit) campaign, this article first examines how antifeminist messages proliferate on Twitter, offering a view of the dynamics driving the manosphere. Second, it proposes a mixed-methods approach to the study of antifeminism. By combining graph analysis with critical discourse analysis (CDA), it bridges the gap between macro and micro levels. Bringing these methodologies together also shows the potential of hybridisation not only by analysing trends but also exploring the quality of the discourse, identifying themes and contextualising the text as a social practice This article is organised as follows: first, we introduce Twitter as a vehicle for campaigning. Second, we define the methods and tools used to explore the #Cuéntalo campaign. Third, we categorise the people who have been sharing messages against the #Cuéntalo campaign. Four, their discursive narratives are critically analysed, identifying and contextualising recurrent themes. Finally, the discussion bridges together the graph analysis and CDA linking the antifeminist backlash with new forms of mobilisation. These integrative approaches mixing methods from different disciplines shed light on how social movements behave in the platform society 4 .

A window to antifeminism
Twitter is a microblogging platform on which users post 280-word messages and multimedia content known as tweets; users can retweet (RT), sharing those tweets with their networks (people who follow them) or 'like' them (Twitter Developer, 2019). Since its launch in 2006, Twitter's usage has grown exponentially; in Spain, it has 6.71 million users (Statista, 2019).
While #MeToo was ignited in the US by sexual-abuse accusations against US film producer Harvey Weinstein, in Spain, the #Cuéntalo movement was prompted by a court verdict in 2018 declaring that the gang rape of an 18-year-old woman during San Fermin in 2016 was a mere sexual abuse. In fourteen days, women wrote 128,400 tweets and 2.6 million retweets, 50,000 of which were first -and second-person accounts of murders (9.8 per cent), rapes (14 per cent), sexual assaults (28 per cent), and abuse (15.7 per cent) (Valdés, 2018).
We are aware that platform datasets can be incomplete, and that relying on social platform analysis to scrutinise social movements is problematic. Schradie lists several failings, including that 'hashtag data are often cherry-picked' (Schradie 2015). In some studies, selecting case studies based on platform usage and data access, instead of the platforms' ability to mobilise people, is a limitation. Due to the concision of the messag-4 The figures, tables and translations are ours. The translated texts into English are inserted in the article, and the original, in endnotes to facilitate readability. es, the fragmentation of meaning over multiple tweets, RTs, biographies, conversation threads and linguistic nuances, and the fact that it is increasingly communicated via images, Twitter poses challenges for analysis (Qian, 2017).
However, observing the manosphere (Hanash Martínez, 2018, 343) is interesting, as it congregates bloggers, YouTubers and Twitter users united by their belief in a world infested by feminism (Landsbaum, 2016;Mantilla, 2015). Social media platforms encourage 'hegemonic discourses and traditional power relations' (Demirhana & Cakır-Demirhan, 2015, 308). Real-time messaging afforded by social media platforms acts as message amplification. When 'these kinds of ideologies are promoted', unhinged men who nurse sentiments of rancour and entitlement can take action, as it has happened with Incels -involuntary celibates-who have even killed people (Mantilla, 2018). Twitter offers a rare window to the world of the antifeminist backlash, because 'if we want to know who is at the other end of the violence, we have to enter the place where the perpetrators are organised and gather' 5 (Hanash Martínez, 2018, 343). Twitter also presents advantages for research, including easy access to its application programming interface (API) and the availability of historical tweets (although this access is limited) (Littman, 2017). Focussing on Twitter to study the antifeminist backlash is relevant, as antifeminism on Twitter can say a lot about how people mobilise online (Ellis, 2018).

Methods
This study uses two research methods, combining graph analysis and CDA to evaluate the antifeminist reaction to the #Cuéntalo campaign. First, we identified who is a member of the backlash community, and based on their tweets, we analysed the characteristics of this community. To overcome the limitations of tweet analysis, we then used CDA to interpret the messages shared within the antifeminist group. These methods reinforce each other, allowing not only to know what has been said but also who are involved in the saying.

Graph analysis
The authors have access to 477.792 Twitter messages connected with the hashtag #Cuéntalo in Spain in Spain on April 26 to 29, 2018 -the three days that concentrate the majority of the activity-by 154.764 users. Tweets were obtained using the Twitter Representational State Transfer API -a Twitter service that allows the harvesting of data-and the t-hoader_kit tool 6 , a set of scripts to extract data from the Twitter API and process them to generate a graph developed by Congosto, which is visualised with Gephi.
Messages on Twitter can be original tweets, replies, RTs or quotes. The level of engagement within Twitter communities can be measured by looking at their RTs (Twitter Developer, 2019). The way people retweet generates information bubbles, as they typically RT what they agree with (Nagulendra & Vassileva, 2014), engendering political polarity (Conover et al., 2010), which can be visualised looking at the communities as they position themselves around particular issues.
Graph analysis -or the analysis of networks and relations-is employed here to study RTs of messages containing the hashtag #Cuéntalo and to identify, within the retweeting groups, the community of backlashers. Using Gephi's modularity algorithm (Blondel, Guillaume, Lambiotte, & Lefebvre, 2008), which can measure networks, we identify communities with the most RTs in common. Gephi is an open-source network analysis and visualisation software (Bastian & Heymann, 2009). Its modularity algorithm groups the users that have connections in common. The more interactions a user has, the easier it is for the modularity algorithm to classify them. The more a user interacts, the more precisely he or she is ranked by the modularity algorithm. In the case of a user's account with similar connections to clashing groups, the algorithm assigns the user randomly to a community, which can lead to possible false positives. The accuracy obtained in the classification of the backlashers is 96.56 per cent. Due to the high reliability of the classification, the analysis includes all the results.
Once the groups have been identified, original messages are classified according to the group to which the author belongs; RTs are connected back to the retweeted user. This way, it is possible to segregate the backlasher community's tweets. Next, the original tweets, replies and quotes are extracted, assigning them the number of received RTs. Downsampling reduces the amount of tweets to which the CDA is applied, outlining the messages with more significant impact (i.e. more RTs).
We visualised the users that published or retweeted at least one tweet with the hashtag #Cuéntalo and the RTs represented as connections (lines) linking retweeter and retweeted. The nodes are coloured by community and user, while their size is proportional to the number of RTs received to highlight the users with the most impact. We named each community based on the most retweeted users within it, exercising a degree of subjectivity.
The shape of the visualisation is given by the algorithm force-atlas-2 (Jacomy, Venturini, Heymann, & Bastian, 2014), which clusters nodes with more connections and repels those with fewer links. This algorithm makes the distance between the nodes inversely proportional to the commonality of their connections. Given one node, its neighbours are those that are connected directly to it. The distribution of the communities shows two macro groupings or filter bubbles that exhibit a degree of consensus within each: one is in favour and the other against #Cuéntalo. Then, we gathered the last 3,200 tweets of each of the 6,628 users identified as backlashers to characterise them.
The backlash group's members are typified by their relationships; that is, how they are connected among them and by their declared and dynamic bonds. A user's declared relationships are the accounts that he or she follows, while the dynamic relationships are those people with whom the user interacts. These relationships were obtained with the Twitter API using the t-hoader_kit tool. That is, looking at how accounts follow others (declared relationships) and interact with others (dynamic relationships), we can obtain the structure and strength of the connections between the members of a given group.
For each member, we have calculated their in-degree (number of followers), their out-degree (number of following) and their clustering coefficient. The in-degree determines the popularity of the user; the out-degree determines their interest in listening to group messages, and the clustering coefficient indicates the cohesion among their contacts. The clustering coefficient is obtained by dividing the number of connections among the neighbours of a node by the maximum number of possible connections. When the clustering coefficient of each node is 1, it is called clique (i.e. a group with strong links). A 0 clustering coefficient indicates weak connections.
This study is also based on a taxonomy that identifies profiles on Twitter, taking into account their activity and impact. The taxonomy includes six types of profiles: 'speakers', whose messages get disseminated widely; 'networkers', who tweet frequently and receive the same number of RTs; 'retweeters', whose main activity is dissemination; 'replicators', who mostly reply to messages; 'monologists', whose tweets are hardly retweeted; 'common', who do not retweet or get retweeted Congosto, 2018).

Speakers
Their messages get disseminated widely; there are three levels (low, medium and high) depending on the dissemination of their tweets Networkers They tweet and RT frequently and receive the same number of RTs

Retweeters
Their main activity is dissemination; 70% of their messages are RTs

Replicators
They mostly reply to messages; more than 70% of their messages are comments

Monologists
Their tweets are hardly retweeted; around 30% of their messages are not retweeted Common They do not retweet or get retweeted; their activity is low as they hardly interact with others Normal Accounts that not meet any of the defined characteristics Source: Congosto (2016;. From the dynamic relationships, we have obtained an impact measure that we call h-index, which is typically employed to measure the online impact of scholars (Hirsch, 2005). Replacing citations by RTs and the number of articles by the number of original tweets, the h-Index can calculate the productivity and impact of each account.

Critical discourse analysis
Discourse analysis is employed to evaluate the most retweeted messages, allowing investigators to study naturally occurring language to reveal the social and individual characteristics of people (Fairclough, 1992;van Dijk, 2015). The main counter-discourses of the #Cuéntalo campaign are analysed as social practices that require examination, since discourses produce and reproduce power relations in society (van Dijk, 2015). This analysis is especially useful to look at 'noisy, unstructured text, like the tweets', which are challenging to analyse using automatized methods (Mukherjee & Bhattacharyya, 2012).
CDA works as a complement to graph analysis. Applying different methods to address the complexity of social media discourse is gaining pace as platforms require new approaches. The communication on these platforms differs from the exchanges on conventional offline media and offers some methodological challenges. While traditional offline media (e.g. written and audio-visual) rely on specific genres and stick to conventions of authorship and authority, communication on social media platforms is based on a shift in the quality of the message. Concise content, immediate, multiple and open distribution channels and diverse speakers' identities (e.g. anonymous writers) are some of the singularities of Twitter that challenge conventional discourse analysis ( CDA is characterised by its focus on both text and context. This approach considers discourse as constitutive of social order and, at the same time, constituted by social structures (Chouliaraki & Fairclough, 1999, p. 4). Social interactions take place through language, which expresses the tensions and power relations emergent within societies. CDA provides an analysis of discursive practices that moves beyond quantitative linguistic content. It explores the relationship between these practices and other social structures (Blommaert & Bulcaen, 2000, 452). In this light, tweeting is a social practice that involves communities beyond the community, producing meaning at a very rapid pace. Tweets often work as a reaction to other tweets, thus embedding the message in the broader network. Twitter creates meaning out of the interaction between text and context in a dialectical movement where elements of dominance, difference and resistance can be identified (Fairclough, 1992(Fairclough, , 2012Wodak, 2013). Adapting CDA to social media implies that the discursive practice in this platform is mediated and ideologically conditioned, and narratives are embedded in a context that requires as much attention as the message (KhosraviNik, 2017;KhosraviNik & Unger 2016;KhosraviNik & Esposito, 2018;Unger, Wodak & KhosraviNik, 2016).
Ideology manifests discursively; sexism, racism, anti-migratory movements, and populist right-wing movements often build similar discursive practices around strategies of polarisation and resistance (Fairclough & Wodak, 1997;Wodak, 2013;Wodak, de Cilia, Reisigl & Liebhart, 1999). CDA embraces the principle that ideology has a discursive form and that it is revealed through textual strategies that are not always explicit (Wodak 2008, 56). Following Wodak's argumentation, racism can be replaced by sexism, since exclusionary discourses operate in similar ways. Being able to identify contents and strategies helps to understand the power of these movements as well as the symbolic and textual mechanisms (e.g. prejudices and anxieties) that contribute to building backlash communities.
In sum, this article's innovative methodological approach combines graph analysis and CDA to study the propagation of antifeminism on Twitter. Mixing these different approaches results in an innovative method that seeks to overcome the problems derived from superficial quantitative analysis and a limited qualitative one. In this case, the limited number of tweets associated with the backlash (256) facilitates a comprehensive and insightful critical analysis.

Meet the backlashers
We visualised the communities of users that published or retweeted at least one tweet with the hashtag #Cuéntalo to identify which ones supported or were against #Cuéntalo. We have only looked at the communities larger than 2 per cent in terms of number of users, finding that twelve communities positioned themselves in favour of the feminist campaign and one against it. Table 2 details each of these support communities, their size and colour in the visualisation shown in Figure 1. The distribution of the communities in Figure 1 shows two macro groupings that exhibit a degree of consensus: one is in favour and the other against #Cuéntalo. On the right-hand side, the visualisation shows the communities that supported #Cuéntalo; on the left-hand side, the backlash group who rejected it. The backlasher community represents 4.08 per cent of the total nodes. This proportion indicates that this is a minority of users, showing a uniform structure.
Tweets classified according to the different communities show that the backlashers had participated from the beginning of the campaign, and remained active during the period analysed here.
We characterised the backlashers according to their declared and dynamic relationships. We have obtained the data from each of the user's account, following the last 3,200 tweets published per user. Not all backlashers have been able to be characterised here because their profile status has changed a year later. Twitter has suspended some users; others can no longer be found because they have been deleted; others have been reported or are being evaluated at the time of writing, and others have closed their accounts, a common practice to avoid being reported for abuse. Twitter can suspend users' accounts when they are found spamming, are fakes (e.g. misrepresentations), are abusive (e.g. threating), or they introduce 'security risks' for Twitter or its users (Twitter Help Center, 2019).
Looking at whether the backlashers' accounts are active at the time of writing offers the possibility of identifying the users that have broken Twitter rules or are afraid of being suspended. A year after the #Cuéntalo campaign, 89.98 per cent of the 6,628 backlashers detected in this study have an active account, 4.25 per cent are suspended by Twitter, and 5.76 per cent have deleted their accounts ('not found'). The percentage of those suspended by Twitter is lower than the percentage of suspended accounts identified in other studies, including research on the 2017 German elections (9.58 per cent suspended) (Majo-Vazquez, Congosto, Nicholls, & Nielsen, 2018). 6.62 per cent of the accounts do not have declared relationships, so we can only analyse their dynamic data. We have been able to obtain all data from 80.70 per cent of the accounts. Table 3 shows a summary of the statuses of the backlashers' accounts.  Figure 2 shows the backlashers nodes whose accounts were active and their tweets open at the time of writing. The connections indicate the declared relations between them. We have used Gephi to visualise it and to obtain the communities and network parameters degree, in-degree, out-degree and its clustering coefficient for each of the nodes. Most of them are grouped into dense interrelated communities of neoliberal or nationalist ideologies; they self-identify as such in their profiles.
Using an inductive method, we have employed different labels to identify seven distinctive sub-communities based on their most active speakers' biographies or self-awarded designations. The idea is to add another analytical layer, which is then complemented by the CDA. These labels identify sub-communities character-ised by their most prominent speakers, whose biographies, self-descriptions, backgrounds and messages we have investigated. Table 4 shows these communities.  Figure 2 shows a compact structure representing the backlashers' community and two peripheral communities, one comprising ultra-Catholics from Peru and another one, prison officers.

Figure 2. Graph of the declared relationships of the backlashers
We have calculated the h-index and the role of each of the backlashers to understand these accounts' roles and typify them according to their activity and impact. The distribution of the backlashers' functions ( Figure  3) indicates that the most frequent type is the retweeter (60.57 per cent), followed by the monologist, the networker, the low speaker, and others. Typically, the high speaker -the most influential voice-is the rarer.

Figure 3. Type of backlashers
The distribution of the accounts according to the number of followers shows that the most common account has between 100 and 999 followers (54.46 per cent), followed by that with between 1,000 and 9,999 followers (27.73 per cent), between ten and 99 (13.06 per cent), zero to nine (3.76 per cent) and 10,000 to 99,999 (1.95 per cent).

Figure 4. Distribution of # of backlashers' followers classified by role
Most retweeters and monologists have between 100 and 999 followers, although there is a significant group of retweeters (13.40 per cent) in the range 1,000 to 9,999 followers, which implies a considerable scope for message-spreading.
Finally, we have compared the clustering coefficient and h-index. The clustering coefficient follows a normal distribution with a mean and median of 0.29. This indicates that the most frequent profiles are those that have a third of their contacts connected. We find that the backlashers' accounts with the most significant impact (that is, high speakers) have less cohesive networks (see Figure 5). Retweeters and some speakers with an h-index below 50 have a clustering coefficient above the average; however, the clustering coefficient drops dramatically for those with higher h-indexes. This drop is due to a fan effect -lots of followers with loose bonds-in which people show interest in popular accounts but do not establish relationships with these accounts' followers.

Backlashers protest
The corpus of the analysis are 590 tweets; still, we only analyse the discourse of those retweeted more than twice (255 tweets), ranging from those retweeted 20 times to two of them that got more than 1,200 RTs, in the understanding that monologists posted the rest 7 .
An inductive method was employed to identify key ideas in the tweets. This way, researchers avoid using preconceived categories, allowing insights to flow from the data instead (Glaser and Strauss, 1999). Since finding patterns in data and interpreting those patterns is up to the researcher (Bernard, 2006, p. 453), labels and hierarchies for codes come directly from the text or the text in its context. As Hsieh and Shannon note, one risk of this type of analysis is failing to develop a complete understanding of the background (Hsieh andShannon 2005, p. 1280); therefore, to assure the credibility of the study, we have relied on the contextualization of the data.
Based on a first observation, to classify this corpus, two perspectives were adopted: thematic and modal. The selected tweets were arranged according to their thematic affinity responding to questions such as 'what is the discussed issue?' According to the answer (e.g. false accusations, unfair trials, feminazism, males' vulnerability), these recurrent themes resulted in a first classification. Thematically, four prominent families emerged:

•
The first (coded as FEM) gathers messages denouncing feminism as a tyrannical movement. This category groups together tweets that criticise feminism as contrary to the general interest and feminists as over-reacting women that destabilise cultural values, abuse their power and demand unfair discrimination of men.

•
The second (MAN) refers to controversial court cases of sexual abuse as the high numbers of tweets on this issue deserved a separate category. Here contextualisation was vital for the classification of these tweets: the social impact of these cases was apparent in demonstrations, media coverage and social media platform activity.

•
The third (LEF) groups tweets that establish connections between feminist policies and left-wing parties. Here the context plays an important role, as these messages often mix up with other issues in the political agenda.

•
The last (MEN) is concerned with men who feel victimised by abusive feminism and denounce discrimination and arbitrary treatment.
A second classification was carried out to identify the mode. While themes are easily spotted, modes of communication are more subtle. Discussing xenophobic discourses, Krzyzanowski and Wodak note that racist messages are considered taboo and no longer embraced by any community: 'racism now often takes more pervasive, diffuse forms, even to the point of being expressed as the denial of racism' (Krzyzanowski & Wodak, 2009, p. 2). The same could be applied to antifeminist discourses. In our sample, there is little explicit misogynistic talk. Instead, the antifeminists encode prejudices through discursive strategies that can be identified in the mode of communication. Classification in mode clusters helps unravel the different mechanisms that operate in the discursive practice and are enacted through strategies, like legitimation/de-legitimation, justification, victimisation, polarisation or shift of responsibility (Wodak et al. 1999). By mode, we understand the epistemological status of the text based on two main axes: opinion and evidence. Tweets labelled as 'opinion' are a reaction to journalistic stories (classified as O.N.) or reveal the user's assumptions (O.S.). In some of these cases, the tweet works as an intertext that finds its meaning in allusion to previous publications. Meanwhile, those labelled as 'evidence' are supported by data, e.g. statistics published by news media organisations (E.D.), or direct testimonies (E.T.). Figure 6 shows how messages are distributed according to themes and modes. 7 In a second round of analysis, we found that most of these accounts had been cancelled either by the users or Twitter.

Figure 6. Themes and modes in backlashers' tweets
Most of the tweets, 76 per cent, are opinions, of which most are opinions based on prejudice or stereotypes; meanwhile, only a few of the tweets rely on the evidence of direct testimonies and facts. It is interesting that, unlike the #Cuéntalo campaign, which is based on the evidence of direct testimonies, backlashers tend to state their opinions or react to news stories. It is only under the theme of 'men being abused by women' that testimonies boost to almost 70 per cent.
Resistance to feminism springs from many different positions, both men's and women's. The thematic content of antifeminist discourses in the examined tweets revolves around four major topics. FEM indicates tweets claiming that feminism is an authoritarian ideology and feminists can be feminazis. Here, the antifeminist sentiment encompasses the idea that feminists exert some tyranny upon all men and the women who do not subscribe to their dogmas. The second group (LEF) revolves around the idea that the ideological bias of the left -which also affects other conflicts such as migration and nationalism-clouds the feminists' judgement. Feminist activism is associated here to leftist parties (to Podemos in particular). The third group of tweets (MAN) denounces the double standards of the left with regards to two male gang attacks against women: one in 2018 in Alicante perpetrated by three Algerian men and the other in 2016 in Pamplona by four men who called themselves La Manada (the wolf pack). According to this group, feminists consider these sexual attacks differently depending on who the attackers are; if the aggressors belong to a deprived, migrant community, feminists are accused of being more lenient. The fourth (MEN) includes messages that assume than men are also ill-treated by women, who often take advantage of their privileged position as the so-called weaker sex.
As is often the case, categories are not clear-cut. Many tweets could be categorised in two or even three of these thematic clusters; for instance, some antifeminist tweets are illustrated by a judicial case and associated with leftist activism. We have labelled them according to the most salient topic. Next, we explore the features that characterise the antifeminist discourse.

Tyrannical feminism
The alleged authoritarianism of feminism is one of the most prevalent ideas in the tweets we have studied. They are more than one-third of the total messages analysed (36.5%). Figure 12 shows how the themes are distributed across tweets and retweets; tyrannical feminism is the most tweeted and retweeted.
One of the most potent expressions of antifeminism comes from women: I endure bullying. Every day, they tell me that because I am a woman, I will never arrive where men do. Every day, they tell me that I need help because being a woman is a limitation. Every day, they tell me that men look at me as an object. All this, feminists tell #Cuéntalo 8 This user, Tweeter 1, is prolific, and her tweets are successful rhetorically speaking, as they gather up to 500 RTs. Tweeter 1 is among the women who accuse feminism of overlooking how some women abuse others. The first person emerges here as a message enhancer and a replacement for data and facts. The strategy of homogenising feminists as one single group appears here, creating a polarisation that can be identified as a positive-self and negative-other representation (Wodak, 2008, 56;Krzyzanowski & Wodak, 2009, 22-23). One individual voice gains strength by speaking in the first person of her experience as a woman. 8 Yo sufro bullying. Cada día me dicen que por ser mujer nunca llegaré a donde llegan los hombres. Todos los días me dicen que necesito ayuda porque ser mujer es una limitación. Todos los días me dicen que los hombres me ven como un objeto. Todo me lo dicen las feministas #Cuéntalo.
The backlash relies on the same strategies as the #Cuéntalo campaign, employing testimonials that speak of abuse enacted by feminists on other women: I have never been violated or anything. But I have suffered harassment, insults, I've been bullied at work… By girls. They've always been girls. #Cuéntalo 9 In this case, intimidation equals sexual abuse and women are more dangerous for other women than men, thus undermining the sorority encouraged by the #Cuéntalo campaign and contributing to deactivating the feminist mobilisation.
According to these users, feminism is also doing little justice to women as the #MeToo movement is having a perverse effect: #Cuéntalo The negative consequences that the 'MeToo Effect' is having on women's employment https://t. co/BQ1bpkBIeJ 10 In this case, the tweet includes a link to a story by the Canadian Financial Post, which announces 'Why #MeToo is not a positive development in the workplace' (Levitt, 2018). This is an example of the kind of tweet that relies on controversial opinions published in newspapers. The strategy is to put all responsibility on the abused woman and not on the abuser, what Wodak defines as a shift of blame (Wodak et al., 1999, 36). Antifeminist resistance is shown through a mechanism of self-exoneration, making women responsible for creating a hostile atmosphere for other women, lessening the responsibility of harassers and eroding women's fighting spirit. This is called argumentum ad baculum (Wodak, 2008, 65), a strategy that challenges the effectiveness of social mobilisation. Here, the meaning-making process of these messages is no longer a straightforward business; this is not an us-versus-them, conventional antifeminist discourse. As Faludi contends, the backlash 'pursues a divide-and-conquer strategy…, elevating women who follow its rules, isolating those who don't' and using rhetorics that 'charges feminists with all the crimes it perpetrates' (Faludi, 1993, 14).
Antifeminist tweets dwell on the idea that women's protests are an exaggeration and that feminism creates false alarms: Gender-based violence complaints have multiplied x3 just when economic compensation has increased, and currently, only 1 in 5 complaints end in a guilty verdict. #Cuéntalo 11 This claim is illustrated through the speech-making of fear, with the support of data that supposedly speak the truth. 5 th country in the world in safety for women according to 'The WomanStats Project'. #Cuentalo https://t. co/YSC0KpKqBT 12 By presenting numbers about Spanish women's high standards of living, as compared to other countries, Tweeter 2 questions the legitimacy of the protests and the necessity of action.

The ideological bias of the left
Frequently, the resistance of backlashers is shown through a rejection of left-wing parties. Antifeminist discourse aligns with a right-wing rejection of progressive policies, whether they deal with migration and asylum laws, separatist protests or minorities' rights. Feminism and left-wing policies work in tandem for backlashers, and feminist demands are associated with biased positions. Three topics can illustrate this anti-left/antifeminist resistance: denials of abuse under scrutiny, the double standards of the left, and the unreliability of leftist leaders.
Associating denunciations of abuse with leftist positions finds ample reception among backlashers: I am surprised that almost all #cuentalo women who have suffered from abuse and rape hang around left-wing circles. I've always treated women with respect and women in my circles haven't been raped by their male friends 13 .
Other than the simplistic generalisation carried out through the synecdochical identification of the 'Podemists' (totum pro parte) with those political agents who support multiculturalist policies, two more strategies operate here: by raising suspicion on the testimonies by leftist women, their protests are being neutralised as deceitful. And by associating #Cuéntalo with other social demands, such as multiculturalism and inclusion policies, the #Cuéntalo campaign is consigned to the political agenda of the left, thus turning #Cuéntalo tweeters into partisans with an ideologised discourse.
The anti-left backlash can be directed against their leaders, mostly Podemos's head Pablo Iglesias. The alleged duplicitous discourse about sexual abuse of leftist leaders has been the focus of many tweets: #cuéntalo In my country, there is a political leader that passes himself for a feminist and wants to beat a woman until she bleeds 15 .
The strategy used here is that of an argumentum ad hominem; that is, a verbal attack of the antagonist's integrity and credibility (Wodak 2008, 64). The tweet on the declarations of Iglesias about a well-known journalist had more than 660 RTs. It works as a synecdoche that explains the deceitful nature of the left, their double standards and unreliability. The atomisation and decontextualisation of the event are also significant techniques in producing an intensification effect. In the end, do not trust the testimonials of sexual abuse because leftist women denounce them and, therefore, are ideologically orchestrated.

Court cases
The interpretation of La Manada case in these tweets illustrates how backlashers create a conglomerate of interrelations by mixing feminist protests with the double standards of the left on law reforms. One of the most retweeted texts highlights these alleged double standards: If you demand 1,500 years in prison for the Algerian gang in Alicante, you're fascist and racist. If you demand 1,500 years in jail for the San Fermines gang, you're the Dalai Lama. Explain this to me because I don't get it. #Cuéntalo 16 As in other tweets on legal processes, the antifeminism intersects with racism and nationalism. Out of the 66 tweets under this label (MAN), 34 make direct reference to racism and the case of the Algerian gang. Nationalism emerges when these cases are confronted with other political cases about separatists in the Basque Country. Even though feminism has no relation at all with separatism, these messages bind them as expressions of the same political agenda. The protests aroused by the lenient guilty verdict of La Manada are used here as a foundation to create a backdrop of antagonism about xenophobia. This is an example of what Van Dijk and Wodak call the 'denial of racism' (Van Dijk 1989;Wodak 2008): prejudice is denied by antifeminist tweeters, who adopt a benevolent position based on objectivity, as they denounce the bigotry in others (Wodak 2008, 65). The antagonism in this strategy works in the symmetrical formulation of the statements: the tweet exposes that harsh sentences are assessed in opposing terms depending on the identity of the perpetrators (Algerian versus Spanish). At first sight, what this tweet decries is unfair. Beneath the rhetorics of this symmetrical structure lies a fallacy in the absence of a real subject: who is calling you a racist? Who is calling them the Dalai Lama? Whose is that voice that equates both realities? Comparison by opposition allows simplified associations, and meaning is inferred from effective polarisation.
False allegations are another recurrent topic in antifeminist discourses. In this case, the user provides numbers of supposed false claims that do not end up in a sentence; however, he does not offer any source. Associating false allegations of abuse with economic benefits reinforces the misogynistic attributes of women as deceitful and greedy, and erodes the effectiveness of the law against gender violence.
According to Tweeter 1 and other antifeminist accounts, women's demands are harmful to men, who become victims: You too, man, we know you exist. You who suffered a wrong, are afraid to speak up, were abused by your partner who does not allow you to see your kids, lost everything in a divorce due to a false complaint of gender violence. Cheer up, you too #Cuéntalo 18 Discursive strategies in many tweets mimic the original #Cuéntalo campaign, reversing roles and adopting a testimonial tone. They highlight the assumption that false gender violence accusations criminalise men: Shit! I almost feel guilty for not having been raped, for meeting only respectful men and for having brought up GENTLEMEN incapable of doing wrong to any women. #Cuéntalo 19 This woman, who has never been abused and raised her children to be gentlemen, emerges as a powerful counter-figure that echoes a nostalgia for affable models of masculinity, a woman, in Faludi's words, that 'in the cultural imagination shrinks to a manageable size' (Faludi, 2006, 84). Clatterbaugh (2003) identifies three principles that are undermined by the antifeminist backlash: first, the social arrangements and roles assigned to men and women are not natural or dictated by God; second, these arrangements favour men; and third, collective action is needed to make these arrangements fairer and equitable (Clatterbaugh, 2003, 21). The complaints of men here react to what Clatterbaugh identifies as the second principle of feminism. Men's fears and anxieties result from what they consider is a feminist conspiracy.

Discussion
The short period of the data collection and the limited size of the corpus cannot be translated into scalable results; however, this Twitter study presents a glimpse into the world of the antifeminist backlash. The graph analysis points to four main characteristics of this community: first, the backlash community is compactly structured in comparison with the dispersed communities that support the #Cuéntalo campaign, even though some backlashers surface distinct voices; second, these antifeminist Twitter users are tenacious as they have tweeted persistently during the three days of study; third, one type of backlashers (retweeters at 60.57%), emerge as most popular, which points to a minority of backlashers who create antifeminist messages and a crowd of followers who disseminate them; and fourth, success in the dissemination of tweets in this community is granted by the effectiveness of the messages mostly by high speakers who are successful in disseminating their messages.
Meanwhile, the discourse analysis shows that the backlashers operate as a tight community that aligns antifeminist discourses with right-wing politics. Right-wing assumptions on migration, justice and nationalism emerge as consubstantial to the antifeminist resistance. Backlashers react in opposition against a supposed fanatical feminist movement, political correctness and the prejudiced activism of the left. Backlashers mostly retweet messages, which are based on opinions, occasionally referring to some news information or facts. These tweets are formulated as statements that delegitimise women's allegations, position men as opposed to dishonest feminists, and portray them as victims of a witch-hunting orchestrated by feminist groups and encouraged by leftist parties. Four main themes emerge from the antifeminist community's criticism: feminism is an oppressive movement that subjugates men, the ideological left behind feminism makes it a biased, double-standard movement, the arbitrary position of feminists on different court cases demonstrates their prejudice and, finally, that men are also being abused by women. These themes do not occur in isolation but reinforce each other.
From both perspectives, the backlashers can be qualified as a community composed of a majority of retweeters and a handful of speakers producing few original messages ( Figure 5), which favour opinions, most of which are not based on facts and news stories. In other words, this community offers a shortage of evidence 17 #Cuéntalo Una chica denuncia a un chico por haberle tocado el culo en #Sanfermin, sabiendo ella que era falso. El chico pasó 2 días en el calabozo, pero se hizo justicia, condenada a pagar 15651€ por denuncia falsa !!! ninguna feminista se pronunció al respecto. https://t.co/3jPCOeJV3k 18 Tu tambien hombre, sabemos que existes. Tu que sufriste una injusticia, que temes decir algo, que fuiste abusado por tu pareja, que se te impide ver a tus hijos, que perdiste todo en un divorcio por una denuncia falsa de violencia de genero, Adelante, tu tambien #Cuéntalo (https://twitter.com/@ PensadorLibre91/statuses/989991341374533637) 19 Joer! Casi me siento culpable por no haber sido nunca violentada, por haber topado siempre con hombres respetuosos y por haber criado a CABA-LLEROS [sic] incapaces de hacer ningún mal a ninguna mujer. #Cuentalo and first-hand testimonies (in comparison with the #Cuéntalo messages based on experiences of abuse). It shows a preference for retweeting and commentary. Surprisingly, men being abused -the #Cuéntalo mirror theme-generates few tweets and little impact in the form of RTs. Demonising feminism is the most popular theme for tweeters and retweeters.
Integrating both graph and discourse analysis, Figure 7 shows the impact of the topics by mode. Figure 7. Dissemination (RTs) of themes by mode Figure 7 shows a consistent behaviour for the first three themes; only the messages about men being abused (MEN) behave differently; these messages mostly provide evidence and show a low degree of dissemination.
Opinion leaders in this community -the high speakers-focus mainly on demonising feminism (FEM) and the left (LEF), getting a massive response in RTs (Figure 8). By tweeting much more, med speakers also generate many RTs for their La Manada-related tweets (MAN).

Figure 8. Backlashers' role and discourse
The 'unknown' users (i.e. suspended, not found or closed accounts) are the most aggressive. They tweet almost evenly about the four themes, getting wide dissemination of their messages related to La Manada and antifeminism; they are less successful with tweets about the left.
The findings of both methods are compatible and show some correlations, summarised in Table 5. The table is offered here as a heuristic tool to think about other instances where this methodological approach can be applied.

GRAPH ANALYSIS DISCOURSE ANALYSIS
The backlash community is compact and acts separate from the pro-#Cuéntalo supporters on Twitter Shows coherence in their messages and an intersection of themes and modes; unlike the testimonial dimension of #Cuéntalo campaign, backlashers rely on opinions.
Eye-catching messaging is more successful than frequent messages The loudest messages get more RTs; the most impactful messages conceal prejudiced opinions under the guise of qualified information.
Neoliberals, ultra-Catholics, and Spanish nationalists make most of this community Neoliberal attitudes denouncing a lack of individual freedom or negative consequences of women's protest for the economy; association of feminism and left activism, and of antifeminism with racism and anti-separatism.
Only a few create messages; most retweet Opinion dominates the discourse; users react by amplifying antifeminist messages.
The backlash community's cohesion identified in the graph analysis runs parallel to the coherence of their messages and the backlashers' inclination to rely on opinions, instead of facts, to support their claims. The success in the propagation of loud messages matches the impact of the most aggressive tweets, as seen in the CDA. The makeup of this community (i.e. neoliberals, ultra-Catholics, Spanish nationalists and prison officers) is in line with the messages supporting right-wing ideals. And the profile of the backlashers (a minority of high speakers followed by a majority of retweeters) corresponds with the way a small number of ideas -i.e. tyrannical feminism, court cases, feminism's leftist leanings and abused men-get disseminated on Twitter by an extensive crowd of followers.
This article also offers a framework for integrating innovative methods, contributing to overcoming the limitations of a single approach. Graph analysis poses the possibility of distinguishing antifeminist communities on Twitter and examining their behaviour; CDA allows investigators to uncover their discursive strategies and favoured themes, enriching the study by proposing a view of their beliefs and bonds.