Few weeks before the presidential elections in Brazil, some feminist movements raised their voice across different web platforms and grounded protests (29 September 2018). The tag #EleNão (not him) becomes not only synonymous of a massive movement against the presidential candidate Jair Bolsonarobut it also gives the political polarization in Brazil a new format: the fight is no longer about the representation of right or left wings, but the ideas and beliefs of one single person. The noise is not very much on Federal or State deputies, but there is a struggle between democratic and modern ideas versus nationalist and extremist ideas. It did not take long for a backlash; #EleSim (yes he). Roughly speaking, part of Brazilians fear to be ruled by a far-right (neo-fascist figure for some) and authoritarian leadership regime (Jair Bolsonaro), while others discredit the Worker’s Party (PT) as if its 12 years ruling the country was a complete disaster. In this dispute, we may face a similar argument structure cross-platforms (rule n.1: to attack the weakest points of the opposition) and a typical composition of dominant voices (e.g. well-known politicians, public figures, activists, artists and musicians, journalists, comedians).

In fact, and since the June Journeys in 2013, the political debate in Brazil was never so lively and polarised. Discussions between daughters and fathers, disagreements among friends and work colleagues, verbal and even physical aggression; it seems Brazil is falling apart. The image of the flames ripping through the main building of the National Museum in Rio de Janeiro, besides representing an incalculable loss of the 200-year-old institution, it is, unfortunately, something that serves to illustrate the current situation in Brazil. In this background, there is also this feeling of political protests (in and out web platforms) as a sort of superficial mechanisms (for some) or a sort of structures of feelings (for others). Something that would not go beyond the act of changing social media profile to support a program or actually going to the streets; re-posting what matters; using particular hashtags to express opinion; making comments on subjects that could not be ignored, or also quickly reacting with likes because the busy routine does not allow you to go further than that. For me, these actions deserve attention in their own forms.

What I propose in this post is a journey on particular aspects that surround the political polarization in Brazil shaped by the tags #EleNão and #EleSim, with analysis framed by Tumblr, Google Images search engine, and Cloud Vision API. Data collection advanced by TumblrTool and Google Image Scraper. It is an oligoptic vision of society [1], indeed. Preliminary findings and insights are distributed in three parts: i) what   hashtag networks can tell about political polarization? (before Brazil’s presidential elections – 7 October 2018); ii) the site of image circulation (one part before and another after the first round results); and finally, Brazilians reactions’ on the announcement of the next Brazilian president. The data analysis follows the digital methods (Rogers, 2013) perspective which approaches the digital from the inside out by following the methods-of-the-medium. In parallel to that, but not less important, this post aims also to raise possible questions that may serve as guidance to digital research.

Key findings: 

HASHTAG NETWORKS: i) While the profile of #EleNão on Tumblr focuses on civil rights, especially women’s and LGBT’s, #EleSim raises the personal attributes of Jair Bolsonaro (honest, a family man, patriot) showing also preoccupation with the candidate health state after the stabbing attack; ii) The adoption of automated systems to spread content across platforms is more easily detected in #EleSim related content than in #EleNão; iii) while the visuality of the protests in #EleSim is mainly composed by official and unofficial political propaganda, and also anti-workers party memes, in #EleNão, images of the protests and women in leadership is what we primarily see.

IMAGE CIRCULATION: i) Between September 17 and October 3, 2018, supporters of Jair Bolsonaro were the actors responsible for generating the images that most circulated across web platforms, those that hit at least 5 and at maximum 10 different URLs; ii) Instagram is definitely the platform in which #EleNão and #EleSim political debate takes place, and with a great possibility of having the flow of images dominated by the far-right’s imagery and visual culture; iii) the suggestions of Google Images, on one side, are supported by new media websites (#EleNão), on the other side, we see the presence of social media platforms (#EleSim), especially, Twitter and Facebook.



Hashtag networks serve digital research with different purposes; from positioning particular interests or strategies to boost likes and followers. In the case of political polarization, event-based tags serve as a path i) to read ideas and conversation;  ii) to analyse the textual and visual content; iii) to have a general perspective of the whole and an in-depth view of its parts; as well as iv) to detect to influential actors. What follows is an attempt to benefit from the affordances of different types of hashtag networks – namely co-occurrence tag, image-label, and image-hashtag; it is a close look at both correlations and specific grammars afforded by these networks. The dataset corresponds to posts published on Tumblr between September 15 and October 3, 2018.

#EleSim and #EleNão co-occurrence tag networks

The following co-occurrence hashtag network shows the profile of the presidential candidate Jair Bolsonaro in social media, giving us a hint on how his beliefs and ideas have been spread and also the use of automation across platforms in his political campaign. Besides the use of key tags such as #Bolsonaro and #Bolsonaro2018, the core topics addressed in this network relates to Bolsonaro stabbing (and with messages of optimism); a far-right “women with Bolsonaro” movement; the value of the traditional family and faith in God. Bolsonaro is depicted as a leader, a hero, an honest and faithful person. Interestingly, and in parallel to such a perfect image, we also see an attack against the Worker’s Paty (tags as ‘Nazi Petista‘ and ‘Get out Worker’s Party‘) and strong positionings such as “a good thief is a dead thief“.

This is an undirected graph with 415 nodes, 5157 edges. Layout algorithm: Force Atlas 2. Gephi. The thickness of edges shows co-occurrences mentions and the node size note count. Clusters related with #EleSim are highlighted in yellow, and #EleNão in red.

Another characteristic to be highlighted is framing the far-right (via the tag ‘right wing’) as a subject of art, entertainment, fun, love and happiness, as well as linking the right wing with LGBT movement. For such new framing, I have a counter-response, which is also based on previous studies [2]; the use of automation to spread political ideas. This can be easily detected through the use of lists of hashtags adopted to boost popularity on Instagram and also cross-platforms, such as #instagood, #repost, #instadaily, #follow, #likesforlikes and further on. In what concerns LGBT, there are actually some representants publicly supporting Jair Bolsonaro – namely the make-up artist Agustin Fernandez. Another layer of conversation within this network is the presence of #EleNão in a more humorous sense (see the clusters at the top), and the force of religious’s tags in opposition to Bolsonaro.

Concerning #EleNão co-tag network (with 1.714 nodes, and 19.531 edges), the analysis was based on the topics of discussion which are strongly connected with the ‘not him’ node. Rather than deleting the main tag or node (a very common pathway to analyse this type of network), I opted to keep with #EleNão and then detect the key hashtags positioned around it [3]. This is how the network exploration starts, and how important topics of discussion were detected. There is, of course, no surprises about it: we see a strong presence of women’s and LGBT’s movements, followed by elections-related topics, and these are also accompanied by anti-dictatorship regime concerns, such as ‘tortura nunca mais’ (no more torture). The claim for equal rights together with Women’s and LGBT’s rights is also one of the main topics of discussion. Additionally, the full support of Fernando Haddad for president and tags related to the city of São Paulo.

Everything reported so far, is part of what we call the main layer of debate around #EleNão on Tumblr. There is, however, the second layer of conversation that encompasses not only aggressive expressions against Jair Bolsonaro but also accusations against homophobia and fascism – for instance, ‘ele nem fudendo‘, ‘Bolsonaro é meusovo‘, ‘homofobia não‘, ‘anti facismo‘. Cities in Brazil (emphasis on São Paulo and Rio de Janeiro) and abroad show the connection with the protests of September 29. Moreover, the use of hashtags’ lists to boost engagement (and commonly emerged from Instagram) are also part of this secondary conversation. Last but not least, the opposition to #EleNão is strongly detected here with Bolsonaro’s hashtags forming clusters that are very likely to respond to automated strategies.

  • Following the natively digital: what hashtag lists can tell about political positioning and automation?

The lists of hashtags are commonly adopted to boost views and engagement, and used by different actors with different interests, politics is not an exception. Either political activists or marketing professionals may follow the hashtag lists as strategies to campaign for a presidential candite, as well as ordinary people like myself.

What you see below is an attempt to follow the natively digital, here in the shape of the grammars of  #EleNão and #EleSim, and in which incidentally I had some hints on political positioning and automation strategies. The first insight refers to the adoption of ‘boost engagement’ hashtags (e.g. #instagood, #instadaily, etc.) and choice of words: while we see a more diverse list of hashtags in #EleNão, only three options are used in correlation with #EleSim. The choice of tags such as ‘instagay’ and ‘instagirl’ are left behind. Bolsonaros supporters spread, however, an opposite idea in social media; that is the candidate has nothing against the LGBT community. The very conservative profile of the Bolsonaro is also reflected in the absence of particular words (tags) to spread his beliefs across platforms. Here, the lack of something can be meaningful.

following the natively digital_hashtags lists_jjo

Another signal of the use of automation strategies is detected within #EleNão co-tag network (see above): the same groups of hashtags used in #EleSim co-tag network happen to appear again (see the blue nodes). One may say that this dataset is not substantial, but such findings (or insights) are about patterns detection and consistency over time [2]. Perhaps, what may change in the proceeding of analysis in larger datasets is the possibility of amplifying this scope. To finish this part, here you can verify who are the most active users of #EleNão #EleSim related content on Tumblr, except cleberjr67 (28 posts) corrupcaomemes (13 posts) and jrdamasceno216 (12 posts), the overwhelming majority of Tumblr bloggers have published between 1 to 3 posts. It will definitely be valid to have a further look at these accounts.

 #EleNão #EleSim image-label network

The image-label network below stresses the site of the image itself and its visual composition. The visual content corresponds to posts published on Tumblr between Sep.15 and Oct.3, 2018.  The visual content (a total of 566 images) was automated labelled by the machine learning models of Google’s Cloud Vision API [4]. What follows is actually an exercise of relabelling the machine: posters, news and banners related to #EleNão #EleSim are what we see at the bottom, and also food-related images at the top right (humourous content). The imagery of September 29 protests is massively represented by #EleNão; crowds in the protests, selfies, or people posing for pictures with posters. While, on a small scale, we find the visualities of #EleSim, protests through images of the protests, selfie or banners. This network can offer a starting point for a more qualitative visual content analysis for the study of political polarization in Brazil.


Since I´ve been following the visual representation of the political polarization in Brazil, I may say that this image-label network is nothing less than a typical depiction of the politically related movements in the country. It may serve as a historical register. This type of network could easily be an end in itself and, of course, accompanied by a well-written academic explanation to convince others of its own purpose. However, we must think that “the fact that a given method is technologically feasible is not a good enough reason to use it”, quoting Noortje Marres (2017, p.113). Image Networks can be visually impressive and substantial indeed, but they also can be meaningless.

#EleNão #EleSim image-hashtag network

The image-hashtag networks, well-suited for small networks, offer a privileged view of the complexity of hashtagging, and these give us a glimpse into the use of political automated beings. It is a look at the relationship between the visual and the textual and vice-versa. What informs the visual? What informs the textual? What are the claims and visual identities that we can relate with #EleNão and #EleSim? These are some questions that image-hashtag networks afford researchers to answer.

Below we see a clear division between #EleNão (on the left) and #EleSim (on the right) forces [5]. ‘Brasil’, ‘love is’, ‘mulheres’, ‘democracia’, ‘eleições’ are the terms that split the image-hashtag network in two, but, at the same time, they connect both sides; because of their position in the middle of the network. Along the working process of data visualization, the first things that caught my attention were the exaggerated use of tags for #EleSim images – another indication of the adoption of automated social media accounts by Bolsonaro team and campaigners.

I do not intend to describe the whole network here, which it is an important step anyway, but I´d rather move attention to general disparities and some peculiarities within this network. What I was particularly expecting to see was the images of Brazilian women who are supporting the far-right presidential candidate Bolsonaro, which I thought to be somehow achievable, accounting also that both sides (against and pro Bolsonaro) went on street protests all over Brazil (and abroad) on the same day (September 29, 2018). What we see instead is a set of official and unofficial election propaganda, memes, anti Worker’s Party content. At first, I thought to have chosen the wrong platform, but, however, a quick look at #mulherescombolsonaro on Instagram indicates the same pattern (with the exception of the profiles @Bolsogatas – created in October 2017) and @Bolsolindas. On the other side (#EleNão), we see images of the protests (with people) and women actually protesting.



We should also start considering what for some is taken as disposable tags, for instance  ‘insta’, ‘phography’, ‘canon’, vasco’, ‘smile’ or ‘love’ related hashtags. These hashtags can either be relying on automation or not, but the fact is they tend to have connections with the subject of study (e.g. see at the top the visual content linked by #instagood or look at the top right side #follow). “Electoral bullying” is a topic clearly out of discussion within this network (I found one related image at the bottom and close #eleições). There is no mid-term in this binary dispute, or the right to agree with one or more points of view or political proposals of the opposition. Either you are #EleNão / #EleSim or you are not. Other interesting perspectives may be revealed in a further analysis of this image-hashtag network. (Please feel free to share your views by commenting or emailing me).


This part encompasses the sites of #EleNão #EleSim image circulation before and after the first round of presidential elections in Brazil. For the period before, the main point was to primarily detect which Tumblr images most circulated across different URLs and what link domains have those images circulated on. To do so, I have looked at Cloud Vision API’s module Web (entities) – namely the ‘web full matching images’ parameter which shows where given images show up online. For the period after the first-round results, I queried Google Images search engine to see what images are suggested when searching for #EleNão #EleSim and to verify the sites of circulation of these images.

#EleNão #EleSim Image Circulation: Tumblr as a starting point platform – a total of 774 images was verified by Cloud Vision API (223 images for #EleSim and 551 images for #EleNão). In this visualization: the 75 images that have hit from 10 to 5 different URLs, followed by the main link domains in which these images have circulated on.


The images above are chronologically organised, each one of these has hit from 10 to 5 different URLs in the period between September 17 and October 3, 2018. The first image reflects what I call double sense hashtags, which especially concerns polarised debates (pro and anti programs); when the original meaning of a tag is shifted to support the opposite cause. In this case, #EleSim is used to support the Worker’s Party presidential candidate Fernando Haddad instead of Jair Bolsonaro. The last image suggests a sort of super-heroes fight; a green strong hand, with the tattoo of the Brazilian flag on the wrist, is stopping a red strong hand that tightly holds a dagger.

An important note to be said is that the total amount of images collected from a platform (here 551 for #EleNão and 223 for #EleSim on Tumblr) has little to say when what matters concern the sites of image circulation. Proof of that is the powerful capacity of the image circulation linked to #EleSim, which elected 33 images to circulate among 10 to 5 different URLs (but with only 223 images listed as starting points), in comparison with #EleNão, which elected just a few more images to circulate among 10 to 5 different URLs (but with much more images listed as starting points: 551).

Apart from that, the affordances of the study of image circulation can be associated with the detection of the lines of arguments, their authors, and sites of circulation. For instance, and beyond the spread of the smiling face of Bolsonaro accompanied by #EleSim, the argumentation in favour of the candidate is presented through humour, fallacies or the endorsement of public figures, in particular, Donald Trump. A similar strategy is embraced by #EleNão; Fernanda Montenegro (Brazilian actress) and Madonna endorse the campaign ‘not him’; the comic aspect is on Cebolinha, one of the famous Maurício de Souza’s characters in ‘Turma da Mônica’. However, the images of the protests and people protesting are again what we do not see in #EleSim, but is pretty well documented in #EleNão.

Supporters of Jair Bolsonaro were the actors responsible for generating the images that most circulated across web platforms, those images that hit at least 5 and at maximum 10 different URLs, namely cleberrjr67corrupcaomemesclaudiaborindelvalleidy2014, and medeixafalar. instagram.com and pbs.twimg.com (an image host) are the main link domains in which the total images collected from Tumblr have most circulated on. Instagram is definitely the platform in which #EleNão and #EleSim political debate take place, and with a great possibility of having the flow of images dominated by the far-right’s imagery and visual culture.



The presidential elections first round result surprised many people in Brazil and around the world: Jair Bolsonaro with 46% of votes, and Fernando Haddad with 29.3%. The week after the shocking results, I queried Google Images to see what images the search engine would suggest when searching for #EleNão #EleSim. Additionally,  to verify what would be the sites of circulation of these images (see the visualization below).


google images_elesim vs. elenao


In short words, and leaving behind the description of the images, on one side we see the power of new media websites (#EleNão), the dominant source is Jornal GGN. The images used in an article signed by Grauninha appears four times in the raking suggested by Google Images. On the other side, a more strong presence of social media platforms (#EleSim), in particular, Twitter via Trendsmap platform to explore trending topics of Twitter, and Facebook via the page Ele Sim – Somos Todos Bolsonaro.


PART III --> [to be continued]





[1] Please don´t blame me, this is all on Bruno Latour and his view on Tarde’s idea of quantification.

[2] See the Chronological Network of #foratemer, #grevegeral and #diretasjá, analysed in March 2017, and how the debate around Bolsonaro started to emerge. It was a period when actually nobody would consider him as a possible presidential candidate.

[3] This is also a form of exploring the affordances of the layout Algorithm Force Atlas 2, and its logic of spatializating the network: more connected nodes are, more central and close to each other they will be positioning.

[4] Here you have two python scripts to work with Google Vision API: MemeSpector by Bernhard Rieder, and the version made by André Mintz (I used the second option). There are, of course, other options of images analysis’ APIs, for instance, Amazon Rekognition, Microsoft Azure, Clarifie, and Imagga.

[5] A special thanks to Gine (Ginevra Terenghi) and her friend Jack for helping me out with the final step of this image-hashtag network. Thank you! 🙂