INTRODUCTION

One of the bases of democracy is the freedom of public opinion, and for society to express its opinion freely, it needs free and true information (UNESCO, -). Until a few years ago, the pillar for that freedom was the written press, television, and radio. Nowadays, however, the way we inform ourselves, debate, and form our opinions has profoundly changed with the arrival of online media and especially social networks (Törnberg, 2018: 1).

Social networks, due to their structure, have the potential to bring citizens closer to the political sphere and its influence (Castells, 2011) as they promote more active citizen participation in political affairs and the creation and dissemination of content (Matakos, Terzi, & Tsaparas, 2017; Rodríguez, Aguilar, & Hung, 2010). Social networks also increase the amount of information we have at our disposal and the variety of options to choose from (Sunstein, 2003: 65). Gathering, in the same space, individuals from different countries and cultures who can expand their horizons with new points of view (Papacharissi, 2002: 23).

However, even those tools that offered the opportunity to democratize social relations and bring together people in the discussion who would hardly come into contact without them, have shown areas of weakness when it comes to meeting the created expectations (Bakshy et al., 2015: 1130; Vaccari et al., 2016: 2; Törnberg, 2018: 1; ) (Borgesius et al., 2016).

Social networks can also be used with other intentions, fostering tension, division, and social fracture, using unverified facts or subjective statements that do not seek more than the simple disqualification of the political adversary’s ideas, or manipulation (Rojas, 2007). And it is that, rather than rational and slow deliberation, social networks work by modulating and distributing an atmosphere, a public feeling, which in turn conditions users, often imposing the majority conditions that circulate online (Maldonado, 2016).

OBJECTIVES

The objective of this study is to analyze polarization on Twitter. To carry out the research, the debate on the Euthanasia Law of Spain voted in the Congress of Deputies on December 17^th, 2020 has been taken as an object of study. It is not the first time that Congress has voted on a proposal to legalize Euthanasia, in fact, it was voted and rejected up to four times before its approval on this last occasion (Bueno & Martínez, 2019). The last one just three years ago, in 2018, presented by Unidas Podemos, in a session in which a dignified death law proposal by Ciudadanos was also rejected. The law proposal, which was worked on for months, also generated an important controversy both in the traditional media and among the political parties themselves (Hermida, 2020), so that as expected, the discussion around the approval of the law in the lower house also had a great reach on Twitter, as evidenced by the tens of thousands of tweets written on the subject of the Euthanasia Law, some of which are analyzed in these pages.

METHODOLOGY

The material consists of tweets collected on December 17^th, 2020 with the hashtag #LeydeEutanasia through the NodeXL platform. It is a software that collects the latest tweets in strictly chronological order (the latest tweets written on the social network) related to the requested hashtag (Kuz et al., 2015: 206). It has been a widely used program in recent years to collect data on social networks for subsequent analysis ( ; Wu, 2013) (Rim, Lee, & Yoo, 2020; Weber, Garimella, & Batayneh, 2013). In total, 22,986 have been collected, the maximum figure that the software allows being collected in its version for the academy, which allows an adequate sampling to be carried out to analyze the rest of the network participating in this debate through it.

These tweets have been visualized in the Gephi.9.2 software, which is used, on the one hand, to perform the calculation of statistics related to the network such as the density of the graph, its diameter, or the entry and exit level of users (Mcsweeney, 2009), and, on the other hand, it allows, through various algorithms, the calculation of clustering coefficients (Cherven, 2015), which allows visualizing the various groups involved in the debate, classified based on the interrelation between them, both before the debate (who follows whom and vice versa) and the resulting one in the debate itself (the interactions they have between them). This software has been used in several research pieces on the same topic (Gaol, Maulana, & Matsuo, 2020; Gualda & Borrero, 2015). Specifically, the OpenOrd algorithm has been used (ibid., 2015: 35) that helps to generate network graphs very quickly, and is more suitable for very large networks due to the loss of precision for the sake of greater speed.

Of these tweets, 5691 tweets have been randomly chosen using the statistical program SPSS (a sample of 25% of the data) for subsequent content analysis. These tweets have been manually analyzed to determine the perspective of users within this debate. For this, the emotional framing classification of Meneses et al. (2018: 46) has been used. Based on this table, we have formulated the following categories for better adaptation to the objective of the study: support (clearly maintains a favorable position), repudiation/dislike (explicitly rejecting the norm), negative offensive (opposed to the approval of the law but it uses offensive terms for the target person of the message), positive offensive (in favor of the approval but when supporting its opinion, it uses offensive language for other users), informative tweet (it is limited to informing about some aspect regarding the law but without evaluating it), non-aligned (those users who do not clearly show an opinion opposed or favorable to the law), and others (for those tweets that do not fit into any other category).

ANALYSIS AND RESULTS

Graph analysis

Three main groups are reflected in the analysis: the first group –in figure 1 in color red/magenta– located on the ideological left, includes the various left-wing sensitivities and political forces, groups, and the media. In total, this group represents 32.35% of the nodes and 35.13% of the edges. We consider a group to be the sum of several clusters which in turn constitute a set of nodes.

In the second group –in figure 1 in color green– located to the right of the graph, are the groups of the ideological right along with other groups and media, and members and associations of the Catholic Church. They are 33.16% of the nodes and 30.93% of the edges.

And lastly, there is a third group in the lower part of the graph –in gray. Two clusters are mixed in this group. In the first of them, Chilean journalists, politicians, and users are located, who, due to the circumstances of the Andean country, where, at the same time that Spain was debating the Euthanasia law in congress, are also included in the hashtag on euthanasia. In the second cluster, one more group is collected that includes various left-and-right-wing organizations and media. As it is not possible to quantify the weight of each cluster within the group, it has been decided not to count this group in the classification of left and right groups. The Chilean cluster Nº. 354 represents a total of 5.76% of the nodes and 4.30% of the edges. While the other group (Nº. 636) includes various left-and-right-wing organizations and media.

Source: Self-made

The first group known as the ideological left includes 5 of the main clusters. By size in terms of the percentage of nodes in the entire network, from largest to smallest, are the number of the clusters are: 531 (8.98% of total nodes), 2 (7.8%), 1 (7.04%), and 32 (4.79%).

The second group is the one that we classify as the one that includes 6 clusters related to the ideological spectrum of the right: taking into account the percentage of network nodes, from largest to smallest, it includes clusters number: 372 (9.82 % of total nodes), 365 (9.58%), 255 (5.13%), 387 (4.19%), 272 (4.12%), 485 (4%), and cluster 0 (2.55%).

As far as users within these clusters are concerned, it happens that the referential actors of the clusters are mostly users who do not hold public office.

Entry-level

In this section, the users are broken down in terms of the entry-level, which records the number of inquiries received by each of the users that make up the network. The total number of mentions, likes, and retweets received by the user is included among the interpellations.

Source: Self-made

In ascending to descending order in terms of entry-level: the largest representative of the network is Marcial Cuquerella (@cuquemar), who defines himself as a media and technology entrepreneur, has an entry-level of 1130, and is a member of cluster nº365. Along with him, in the same cluster, are other Twitter users such as @adrianbelaza -entry-level 113-, or Alberto Asensi (@albasenve) -entry-level 96-.

The next most prominent user is @derechoamorir, the platform Derecho a Morir Dignamente, the association defending the law, with an entry-level coefficient of 842. It is followed by the vice president of social affairs of the Government of Spain, Pablo Iglesias (@PabloIglesias), with a coefficient of 793. This is at the center of the debate, but this does not mean that he has a centralist position, but that Pablo Iglesias is a reference node among the users who make up the position favorable to the approval of the law.

They are followed, in cluster nº531, by Fernando Barea (@Fgarea), deputy director of the digital media outlet El Confidencial (entry-level of 508) and other people who, due to their interactions in the debate, do not appear in the graph directly related to any specific political party in this debate. Behind, cluster No. 2, which corresponds to the Podemos political party (@PODEMOS) -entry-level of 497- and its parliamentary spokesman Pablo Echenique (@PabloEchenique) -entry-level of 402-. This cluster is more related to the political party and its groupings.

In cluster nº372 we find the inverse phenomenon, the main references are people not directly related to a political party @rimbaudarth (entry-level 471) or @josemanuelsoto (entry-level 247), except for the councilor of Madrid of Vox Fer Martínez Vidal (@fmartinesvidal_) with an entry-level of 276.

The next most striking representative is the Archbishop of Madrid, Carlos Osoro (@cardenalosoro) -entry-level of 255-. And next to him are other members of the diocese such as the general secretary of the episcopal conference, Luis Arguello (@MonsArguello) -entry-level of 45-, or the archbishop of Toledo, Francisco Cerro (@Obispofcerro) -entry-level of 39-.

The referential user regarding entry-level in cluster nº272 is the political party Ciudadanos (@CiudadanosCs) -entry-level of 238-, and around it are Edmundo Bal (@BalEdmundo) -entry-level of 117- and Begoña Villacis (@begonavillacis) -entry-level of 224-.

In a new cluster, we find Luisa Carcedo (@luisacarcedo) and several members of the Socialist Party (@PSOE), this group located at the top includes both the political party (with an entry-level of 154) and the parliamentary socialist group and several of its references such as Adriana Lastra (@Adrilastra) -entry-level of 121-, or Odón Elorza (@odonelorza2011) -entry-level of 50-.

Group nº387 is made up of anonymous users such as @eugeniodors, the user with the highest level in the cluster -193-, people from other fields such as Elena Postigo (@PostigoElena), professor of philosophy and bioethics -entry-level 174-, or @princesatesa, with an entry-level of 41.

Cluster nº485 is encompassed by @Vox_congreso, the account of the VOX parliamentary group that has an entry-level of 142, Emilio del Valle (@edelvallerod) with an entry-level of 122, Vox's Deputy in Congress, and next to him there are other politicians and media such as Macarena Ollona (@Macarena_Olona) -entry-level of 48-, Jorge Buxade (@Jorgebuxade) -entry-level of 44-, or the media outlet La Gaceta (@gaceta_es) -entry-level of 48-.

In the last group, nº0, there are several members of the Popular Party, with its president Pablo Casado (@pablocasado_) at the head -entry-level of 18-, although in this group the President of the Government of Spain, Pedro Sánchez (@sanchezcastejon), also stands out. This is because members of the popular party, politically opposed to Sánchez, repeatedly question him in the debate, as can be seen in his high entry-level value -107-.

Table 1: Entry-level

Table 1. Entry-level Cluster reference users according to their entry-level
Cluster Number	Percentage of nodes	Reference users	Entry-level
365	9.58%	@cuquemar	1130
1	7.04%	@derechoamorir	842
531	8.98%	@FBarea	508
2	7.8%	@PODEMOS	497
372	9.82%	@rimbaudarth	471
255	5.13%	@cardenalosoro	255
272	4.12%	@CiudadanosCs	238
32	4.79%	@Luisacarcedo	219
387	4.19%	@eugeniodors	193
485	4%	@Vox_congreso	142
0	2.55%	@sanchezcastejon	107

Source : Self-made

Exit level

We call the exit level the interactions made by the users of the network with the rest of the users, understanding by interactions the mentions, retweets, and likes made to the rest of the users. We must point out that, regarding the exit level, the difference in the number of outstanding users in each group and the rest of the users in the group is much greater than in the entry-level. The graph very visually shows that only a few users from each group participate very actively (they make multiple interpolations towards other users) in the debate. Each of the clusters has one or two users whose activity -visually represented by its size in the graph- is exponentially greater than the rest, clearly standing out visually in the graph. In turn, it should be noted that the classification of the clusters does not vary concerning their size in the general network, so that in this section of the exit level only the users with the highest exit level will be identified, without highlighting the positioning of the cluster.

From largest to smallest concerning the exit level of users, Podemos Alicante Oeste stands out, from cluster nº 2 (@p_alicanteoeste), with 37. Behind it, @ 1denmadrid (cluster nº 1) -exit level 23-, @prensacee from cluster nº 255 with 22, @pilargonzalezl1 from cluster nº 365, @meme_montero from cluster nº 32 and with exit level 17. It is followed closely behind, with 17, by @joselisa17 from cluster nº 372. The next on the list are @martinaguirrecm with 14, @pinkiewarriorpw from cluster nº 272 has the next highest exit level -13-, followed by @juanbasangulo from nº 0 -12-, @jgdeaguilar from nº 387 with 11, and, lastly, @mfedorral from nº 531 with 8.

Table 2: Exit level

Table 2. Exit level Cluster reference users according to their exit level
Cluster Number	Percentage of nodes	Reference users	Exit level
2	7.8%	@p_alicanteoeste	37
1	7.04%	@1denmadrid	23
255	5.13%	@prensacee	22
365	9.58%	@pilargonzalez1	20
32	4.79%	@meme_montero	17
372	9.82%	@joselisa17	17
485	4%	@martinaguirrecm	14
272	4.12%	@pinkiewarrior	13
0	2.55%	@juanbasangulo	12
387	4.19%	@jdeaguilar	11
531	8.98%	@mfedorral	8

Source: Self-made

DISCUSSION AND CONCLUSIONS

The analysis of the results offers an easily identifiable visual map in which an enormously complex debate is simplified, and in which a large number of factors, actors, and variables interact. Using several techniques, it was possible to identify the most relevant subjects within the object of study, using quantitative techniques, and their subsequent content analysis through a qualitative technique, which allowed the sequencing of tweets and their opinion orientation, in favor or against, regarding the analyzed debate.

Several of the hypotheses formulated at the beginning of the study are confirmed both by the relational graph of the clusters, as well as by the subsequent content analysis and its translation to the emotional framing. On the one hand, the hypothesis that the debates between the different political parties regarding the approval of the euthanasia law, outside of social networks, has a direct bearing on Twitter is validated. And that this discussion does not only take place between the parties themselves or their representatives but extends to other users, as well as to social groups or the media. Furthermore, these results on the ideological/partisan cluster divide on Twitter are also in agreement with the conclusions of Conover et al. (2011:95).

We can also affirm that the network clearly reflects the presence of at least two groups of clusters differentiated from each other in terms of their opinion regarding the debate on the euthanasia law, two prominent echo chambers, which although they have a close relationship between them, and a direct interrelation in the form of edges that unite the nodes with each other, they do not have such close contact with the opposing opinion group, creating a communicative vacuum between them, giving rise to an evident polarization between the opinions favorable and opposed to the approval of the law. The results are consistent with previous studies conducted by: G Garimella and Weber (2017), Vaccari et al. (2016), Q Quattrociocchi et al. (2016), Colleoni, Rozza, and Arvidsson (2014), Hahn, Ryu, and Park (2015), among others.

Furthermore, this is supported by the main idea of Barberá, Jost, Nagler, Tucker, and Bonneau (2015), where they stressed that the nature of the event would change the course of the conversation and polarization and that this would have ideological nuances. Besides the event itself, the phenomenon of polarization also varies from country to country and depending on the circumstances and political conditions of each of them (Urman, 2020). This polarization also occurs in a very even correlation of forces in terms of the percentage of tweets in favor and against, with a narrow margin between users who show their support for the approval of the law, 42% of the tweets, compared to those who oppose it: 35%.

However, it is noteworthy that this structural polarization in the face of the debate does not translate to the same extent to the language used by both groups when interacting with each other. The percentage of favorable and opposing tweets is much greater than those that address the ideological adversary in offensive terms: only 4.25% of the tweets made by users against the approval of the law are written in offensive terms, while those offensive tweets written in favor of the approval of the law are only 1.79%. These results are in line with Buder et al. (2021: 7), pointing out that the language used is a precursor to the division of opinions, marking polarization.

The methodological approach designed for this research offers a static photograph of a discussion that, however, has a dynamic development. Although the study of the debate at the time of the approval of the law offers relevant information about its structure, and the discursive terms in which it occurs, this does not prevent said structure or the discursive framework from being able to evolve or vary over time. Interesting, therefore, are those studies that give continuity over time to the study, to compare the position and forms of the speeches of each other and the motivations behind that change process, to have a greater and more complete understanding about the phenomenon to be studied.

At the same time, the classification of tweets in the emotional frame has been made based on a favorable, contrary, or neutral categorization, without assessing the motivation behind each position. Although throughout the content analysis of the tweets, it has been observed that a large percentage of the tweets in favor denoted the desire for an expansion of social rights, and many of those issued against it argue religious or moral reasons to oppose the approval of the law, it would be interesting to include a methodological analysis that allows classifying the motivations behind each position. The development of these questions raises new research as pertinent on the phenomena of polarization and the motivations that lead to it.

REFERENCES