#EspañaInvadida. Disinformation and hate speech towards refugees on Twitter : A challenge for critical thinking

Disinformation is not a new phenomenon, but it is widespread in our society since social media have become a media loudspeaker. The outbreak of war in Ukraine has produced a social debate, partly reflected on social networks, about the treatment of Ukrainian refugees compared with other refugees from the South. For this reason, this research proposes a study that follows a qualitative interpretative methodology with an exploratory and descriptive scope that analyses in depth the content of the discourse on Twitter about refugees and uses computer-mediated discourse analysis as a technique for obtaining data. Specifically, it compares the content published under the hashtags #NotRefugees and #Refugees


Introduction
This study aims to analyse the content disseminated about refugees on the social network Twitter to delve deeper into the discursive typologies related to this group. To this end, computer-mediated discourse analysis is used to obtain data. This technique is useful and innovative from a scientific point of view because it analyses the linguistic and pragmatic properties of the content of the interactions on this social network where what prevails is communicative immediacy and the use of colloquialisms (Mancera-Rueda; Pano-Alamán, 2014).
This research is especially important in the Spanish context since, due to its geographical location, there is a constant migratory flow with characteristics that differ from those in other European countries. Furthermore, this phenomenon is a subject of constant public debate, both on social networks and in the conventional media, which has been used on numerous occasions to misinform and manipulate public opinion for political purposes (Lava-Santos, 2021). The following study objectives are proposed: -to describe the characteristics attributed to refugees on Twitter, and -to analyse the discursive typologies about refugees on Twitter.

Review of scientific literature 2.1. Communicative misinformation: Anything goes to get your message across and convince
Society is undergoing a major paradigm shift. The so-called knowledge and communication society is giving way to a society characterised by disinformation and 'infoxication' -information overload. Hoaxes, information excess, and fake news are the main characteristics that define society's access to information in the media (Amores; Arcila-Calderón; Blanco-Herrero, 2020; Magallón-Rosa, 2021).
The veracity of the media is in crisis, especially at a time when society is immersed in a crisis of values, where instability and social polarisation are the order of the day (Belmonte; Muñoz-Álvarez; Bernárdez-Gómez, 2022). As Pari-Tito, García-Peñalvo and Pérez-Postigo (2022) state, the viral spread of fake news has become one of the great challenges of 21 st century communication. The lack of effective mechanisms or strategies to curb its propagation makes it a challenging problem to tackle.
The darkest side of this phenomenon was experienced during the COVID-19 pandemic. Anti-vaccine, conspiratorial, and reactionary discourses posed real risks to health and social welfare because of the virality they achieved (Larrondo-Ureta; Peña-Fernández; Morales-i-Grass, 2021). This expansion of misinformation cannot be explained without considering the potential of its great ally: social networks. As Pari-Tito, García-Peñalvo and Pérez-Postigo (2022) state, even though social networks were created to be spaces for communication between people, they have become one of the main means of exchange of information and news.
Social media possess characteristics that make them perfect allies for misinformation, such as ease of access, immediacy and simplicity of use, and user anonymity (Corbella-Ruiz; De-Juanas-Oliva, 2013; Gamir-Ríos; Tarrullo, 2022). These qualities allow unverified content to go viral as consumers of this information interact with it and share it without effort or analysis. (2019) states that these characteristics of social media have created a new rhythm of information, i.e., rumours and unverified news that have a minimum of logic begin to spread without any type of control. Likewise, the very design of these tools is so simple and intuitive that they make it easy to acquire a basic knowledge of how to use them, both in their consumer and prosumer aspects, i.e., users not only consume content but are also content creators and distributors (Martínez-Sala; Segarra-Saavedra; Monserrat-Gauchi, 2018).

Alonso-González
The social instability caused by the COVID-19 pandemic was followed by the Russian invasion of Ukraine. The latter has led to a further deterioration in areas such as the economy, the political climate, and social coexistence but has undoubtedly also led to a rise in disinformation (Magallón-Rosa, 2021). The war in Ukraine has made it clear that disinformation is a weapon of war because lies change society's perceptions and have been an instrument of war used since the beginning of the conflict (Arrieta; Montes, 2011). Furthermore, disinformation has been both a defensive and offensive weapon. In short, as Navarrete-Barrutia (2020) asserts in his study on foreigners in Spain, disinformation is an ideal instrument for promoting falsehoods and influencing collective consciousness.

Hate speech on social media: The scapegoat for our ills
Hiding the truth, manipulating it, or camouflaging it is a phenomenon that has been used in wars and conflicts since the earliest civilisations, but it was during the Second World War that an infrastructure for the organisation and planning of propaganda aimed at manipulating public opinion first appeared (Vergani et al., 2022). Disinformation is a phenomenon that destabilises and polarises society; lies spread through hoaxes and fake news change citizens' perceptions, and society becomes a victim of this manipulation (Badillo-Matos, 2019; Arcila- Calderón et al., 2022).
Disinformation is a phenomenon that destabilises and polarises society; lies spread through hoaxes and fake news change citizens' perceptions On other occasions, the type of framing offered in a news item has a series of cognitive effects on the reader. Igartua et al. (2008) found examples of framing of news about immigration that linked it to the phenomenon of crime (Igartua et al., 2007). Hate speech in social media has been found to promote discrimination against certain groups in society (Arrieta; Montes, 2011).
Hate speech can be defined as a form of expression that openly promotes, justifies, and disseminates the disparagement and exclusion of certain social groups based on ethnicity, nationality, religion, sexual orientation, gender, disability, and other characteristics (Amores; Arcila-Calderón; Blanco-Herrero, 2020; Wachs et al., 2022). Another definition to consider is that put forward by the European project Preventing Hate Against Refugees and Migrants (Pharm), which aims to prevent hate speech against refugees and migrants. This project defines hate speech as speech that seeks to harm the very dignity of a person or group based on a number of defining characteristics, both inherited and acquired (Pharm, 2019).
This project warns of the importance of cyberhate, since, as research has corroborated, there is a clear relationship between hate speech on the internet and the growth of crimes against the groups towards which this hate is directed. Various elements explain the growth of this phenomenon, where the discursive change that has taken place in the public and political sphere is important. In addition, the negative perception of immigration has increased, with people being warned on many occasions that immigrants have links to terrorism (Pharm, 2019; Sánchez-Holgado; Amores; Blanco-Herrero, 2022).
The FAD Foundation (FAD, 2022) suggests that there are five mechanisms for carrying out the 'fabrication of hate': -creating or exaggerating fear towards a group without justified evidence; -pointing out supposed culprits who, in fact, have nothing to do with wrongdoing; -creating unjustified fear and amplifying it until it becomes a great threat; -predicting the appearance of a great leader appear who will 'save' us; and -acting against the 'enemy' without any scruples.
However, it should not be forgotten that all such incitements to hatred, hostility, discrimination, and violence against groups or individuals based on race, ideology, religion, sexual orientation, gender, or disability grounds may constitute crimes in Spain (Article 22.4 of the Penal Code) and throughout Europe.
Prejudices, as part of the phenomenon of the fabrication of hate, are of great importance as they form the basis of the discourse and become facilitators of the commission of hate crimes (Müller;Schwartz, 2021). Racism and xenophobia cause the most hate crimes in Spain, according to the latest annual study of the National Office for Combating Hate Crimes (2022) (Ondod hereinafter). Disinformation is a key factor in the promotion of hate speech, especially at times of great socio-economic instability, which encourage the search for scapegoats for the causes of society's ills. Blame is often placed on migrants through fake news and disinformation (Narváez-Linares; Pérez-Rufi, 2022).

Educommunication: Critical and ethical thinking in the use of the media
Citizens are defenceless if they do not develop the skills, knowledge, and understanding that facilitate a critical and ethical use of media in general and social networks in particular (Vuorikari; Kluzer; Punie, 2022). Educommunication becomes part of the basic training or literacy that must be accessible to all citizens so that they can access and use information critically and create media content responsibly and safely. This learning cannot be limited to the simple use of tools and technologies but must aspire to provide the capacity for critical and ethical thinking in the use of information and media (Gutiérrez-Martín; Pinedo-González; Gil-Puente, 2022; Vuorikari; Kluzer; Punie, 2022).
As Gutiérrez-Martín and Tyner (2012) argue, this media education should not be reduced to an understanding of the forms of personal or social communication but rather is essential to the development of a critical education in relation to these communicative processes. From the field of information and communication, it is necessary to understand the need for society to develop critical thinking in the use of information and communication technology (ICT) and media, as well as ethical behaviour as responsible and competent citizens. Although this need to develop so-called digital competence in citizens is often mentioned, this training tends to be reduced to the more instrumental and technical dimensions of ICT (Gutiérrez-Martín; Tyner, 2012; Unesco, 2021).
Hate speech seeks to harm the dignity of a person or group based on a number of defining characteristics, both inherited and acquired Racism and xenophobia cause the most hate crimes in Spain, according to the latest annual study of the National Office for Combating Hate Crimes Consequently, information literacy needs to be developed to access and critically evaluate information and make ethical use of information throughout the process (Wilson et al., 2011). Media literacy is also needed to understand the workings of the media and its economic and ideological interests (Wilson et al., 2011).
Authors such as Arrieta and Montes (2011) define digital literacy from three perspectives: -the use of technology; -critical understanding when using it; and -the ability to create and communicate digital concepts in a wide variety of formats.
We focus here on the principle of critical understanding, defined as "the ability to understand, contextualise and critically evaluate the digital media and content with which one interacts" (Montes; Arrieta, 2011, p. 187).
The role of media education should not be limited to instrumental training based on the use of digital media as a medium. An urgent need exists for a critical education of citizens to provide them with the ability and the means to select and contrast the information they access (Zaragoza-Lorca, 2007). The framework of digital competences for citizenship developed by Vuorikari, Kluzer and Punie (2022) within the framework of the European Commission (DigComp 2.2) is a necessary and complete approach in its theoretical bases. This update of the DigComp model provides a framework in which critical implementation is crucial in moving away from a purely instrumental perspective on digital competence. The model is made up of a total of 21 competences, grouped into five large blocks: -search and management of information and data; -communication and collaboration; -creation of digital content; -security; and -problem solving.
These competences reflect the development of digital competence oriented towards not only technological management but also the importance of internet security and the need for critical thinking in the face of phenomena such as disinformation and fake news.
In short, citizens need critical and ethical training in the use of the internet, especially social networks, to combat phenomena such as disinformation that favour social polarisation and negatively affect social coexistence, especially through the promotion of hate speech (Álvaro-Sánchez, 2018).
The design of the study's qualitative methodology is a systematic approach based on grounded theory, which is notable for meticulously following an inductive coding process to arrive at a theory for the phenomenon under study (Jorrín-Abellán; Fontana-Abad; Rubia-Avi, 2020; Hernández-Sampieri; Fernández-Collado; Baptista-Lucio, 2018). A descriptive analysis of the main categories in a sample of analysed tweets was also carried out (Gutiérrez-Braojos et al., 2017).
Citizens need critical and ethical training in the use of the internet, especially social networks

Sampling
This study follows a non-probabilistic convenience sampling approach with two criteria: -the selection of tweets that fit the research objectives in belonging to a clear discourse on refugees; and -the establishment of criteria for the interaction of the tweets to select those that have generated at least a minimal level of engagement among Twitter users.
These criteria are the number of likes (minimum of 10), comments (minimum of five) and retweets (minimum of three).
Data collection was carried out during February and March 2022. A sample of 344 tweets was obtained using the hashtags #NoSonRefugiados and #SonRefugiados.

Procedure
First, a systematic literature search on hate speech in social media was carried out. This search allowed us to identify previous studies of interest in the field and identify various research opportunities.
Second, data collection was carried out using the Tweet Archivist programme, which allowed the creation of a database of tweets that have been published under the hashtags #NoSonRefugiados and #Refugiados. The social network Twitter was selected because its characteristics are congruent with the design of this research: -it allows any user to access tweets that are posted openly; -the content of the tweets can be textual, images, or videos, so an in-depth analysis of the messages intended to be conveyed can be conducted; -hashtags are widely used in this social network (hashtags are texts preceded by the # symbol that help to categorise the subject matter of the tweets, expand the message, and find or detect tweets on a particular topic); and -Twitter reports data that help to quantify the relevance and impact of a tweet, such as the number of likes, comments, and retweets, characteristics that have been taken into account in the selection of the tweets.
These hashtags were selected because of the existence of debates among users about the treatment of refugees, on the one hand, in favour of equal action in the face of the phenomenon (#Refugees) and, on the other hand, the justification of why both groups cannot be treated equally (#NotRefugees).
The qualitative analysis of the data consisted of the following steps (Strauss;Corbin, 2002): -detailed reading of the data and -coding of the data based on thematic areas relevant to the study in order to deepen the understanding of the phenomenon under analysis: the discourse towards refugees on Twitter.

Data analysis
The software Atlas.ti was used to carry out the data analysis. Figure 2 shows the coding process, which consisted of open, axial, and selective components (Strauss;Corbin, 2002). In this coding process, emerging categories were also detected and defined, as well as memos, comments, and quotes that have allowed a central category to be established. From this, hypotheses and theory development were generated.
As Figure 2 shows, during the open coding process, categories and subcategories are detected from the emerging codes.
To facilitate their understanding in the networks presented in the Results section, Table 1, in which the syntax and label of each of the categorical levels are defined, has been elaborated.

Results and discussion
The data analysis follows the coding process (open, axial, and selective) established by grounded theory (Strauss;Corbin;2002). After elaborating the open coding, two phases were developed in the axial part: a first phase in which the relationships between codes were established considering the research objectives and a second, axial-selective phase, in which the emerging theory of the research carried out was generated (San-Martín-Cantero, 2014).
By means of open coding (an inductive process), we sought to extract the content of the tweets analysed and thus describe the ideas, thoughts, and meanings they reveal (Strauss;Corbin, 2002). This first content analysis process resulted in the extraction of 66 codes. Given the large number of codes, it was decided to develop code groups to facilitate the categorisation process. After comparing the codes with respect to their properties and meanings, a summary of generic concepts and their definitions (see table 2), drawn from the comments elaborated in the analysis process, has been elaborated. National security reasons.
Reasons why hatred is expressed: Hate speech towards refugees from the South as a global concept has generated different types, classified into subcategories. Each shows a motive or reason for justifying hatred towards refugees.
The subcategory of "racism" has a peculiarity in that it is not a justification as such, but simply hate speech based on racial adjectives. This process provides a starting point for the next phase: axial coding. At this point, the aim is to identify the relationships between the categories and subcategories obtained from the analysis of the tweets in the open coding (San-Martín-Cantero, 2014). According to Strauss and Corbin (2002), it is important to bear in mind that categories and subcategories represent a phenomenon. Therefore, through the emerging categories, the phenomenon of the discourse on Twitter about refugees is developed in parallel.
Axial coding was carried out in two phases: (1) establishment of the type of relationship between the codes and their subcategories and categories (this process is based on the table obtained from the open coding) and (2) development of the relationship between the categories and subcategories. Comments and memos elaborated throughout the qualitative analysis process are considered. At this point, the elaboration of networks is carried out. Their presentation is organised according to the objectives of this research.

Attitudes and values towards refugees
First, and following the inductive process of grounded theory outlined by San-Martín-Cantero (2014), the first network is obtained (see Figure 3). It shows how codes are connected to positive or negative attitudes and values (subcategories) through an 'is a' relationship. It can then be seen that both subcategories have an 'is a part of' relationship. This shows that both types are part of attitudes and values as a general concept. This dual structuring shows a division in terms of positive and negative values and attitudes. It is posed as a way of polarising and dividing society into good and bad, with the latter being identified as a threat (i.e., invasion, disorder, danger). This content is related to what was previously stated by FAD (2022) concerning negative values and attitudes helping to create a great threat, pointing out culprits that have nothing to do with any wrongdoing, and creating an unjustified fear that becomes a threat or enemy against which we must direct ourselves. The network codes (see Figure 3) are evidence of this (FAD, 2022).

Feelings and emotions towards refugees
The second network (see Figure 4) maintains the same structure as the previous one. All codes distinguish between negative and positive emotions and feelings, although, for example, 'grief' can be linked to both since it is expressed in both senses. The codes have a relationship of possessions to their subcategories. These are linked as part of the first-level category, 'emotions and feelings'.
This content clearly reflects the importance of emotional and sentimental content on social networks. In this regard, Martínez-Rodrigo, Segura-García and Sánchez-Martín (2011) observe that there are three reasons why emotions are so predominant in social networks. The first is immediacy as a dynamic in the functioning of the networks, where emotional content permeates the consumer quickly and without reflection. The second is participation, which arises from the rapid interaction between users, which invites rapid communication. The third is personalisation, as people understand social media as a space for free expression. This last conception of networks as an area where freedom of expression has no limits is related to the conception of unreality in their use (Bustos-Martínez et al., 2019). All of this reflects a perception of impunity, translated into anonymous profiles that make the propagation of hate speech even more feasible (Blanco-Alfonso; Rodríguez-Fernández; Arce-García, 2022).
Consequently, the use of emotional content reinforces the impact of tweets on users. This effect is also detected in the use of visual content as a way of reinforcing and making the message visible, an element that has also been detected in this study, as shown in Figures 5 and 6. In Figure 5, we can read a tweet and see the image that accompanies it; it can be seen that it conveys negative emotional content, e.g., fear, hostility, or anger. In contrast to this tweet, we find  Figure 6, where both the written text and the image aim to convey opposing emotions. On the one hand are positive emotions towards the women and children from Ukraine, provoking tenderness, help, etc., and are the opposite emotions, such as fear and hostility.
This use of visual content as a means of impact corroborates the hypothesis put forward by researchers such as Gamir-Ríos, Tarrullo and Ibáñez-Cuquerella (2021), who argue that audiovisual content is a great mechanism for promoting disinformation by increasing the possibility of penetrating deeper into the social subconscious. Larrosa (2007) defends the need to provide citizens with tools to understand the visual world, given its great potential to promote disinformation and hate speech.
This constant appeal to emotions in the content awakens a collective perception that leads to misinformation and social polarisation (Castillo-de-Mesa et al., 2021). Research by Molina-Cañabate and Magallón-Rosa (2019) suggests that information based on the promotion of hate speech on Twitter can arouse a series of irrational and emotional impulses that contribute to increasing the virality and visibility of such speeches.
These highly emotional discourses make it impossible for information verification agencies, such as Maldita.es, to intervene. This is because they are a type of message that is not subject to verification, connected to emotion and opinion, a channel through which disinformation and hate speech can be transmitted.

Refugee characteristics
In the third network (see Figure 7), the following process of relating categories emerges. The codes are associated with existing types of refugees. This shows a clear segmentation according to refugee type. Consequently, both refugee typologies have an oppositional connection, as they represent two different phenomena in terms of characteristics. Nevertheless, both subcategories are part of ('is part of') the category 'refugee characteristics'. Therefore, this network reflects how, in Twitter discourses, distinct qualities are assigned to each type of refugee.
This network shows that the discourse analysed on Twitter shows a differentiated treatment between two typologies of refugees: on the one hand, those coming from Ukraine and, on the other hand, those coming from other places, especially from the south. The latter are criminalised and seen as 'butchers' and 'criminals', while refugees from Ukraine are viewed more positively, with characteristics such as 'family' and 'women and girls'. Valdez-Apolo, Arcila-Calderón and Amores (2019) conclude that immigrant status is used in the media to discredit or disqualify certain refugees. People from Ukraine are treated more favourably in terms of the characteristics associated with them. The discourse against other refugees has a clear negative focus in the same dimensions as the previous one. These ideas are discussed by Rebollo-Díaz (2021), who identifies social networks as a medium in which a negative discourse about refugees spreads hatred and prejudice towards them. However, the same is not true of the treatment of Ukrainian refugees. As Delfino (2022) argues, the response to these refugees has had a clear focus on solidarity and welcome, which is clearly reflected in public opinion, as well as in social media. However, given that this is such a recent phenomenon, there are no precise studies on how they have been treated in the media. This pronounced dualism can also be seen in the proliferation of hate speech against refugees from the south.

Motives for expressing hatred
The fourth network has a very high complexity, given the large number of codes generated in the open coding, despite having carried out a code fusion process, as in the rest of the networks. For this reason, the exposition of the network was divided into three different figures (Figures 8, 9 and 10) to make it easier to understand them.
In Figure 8, two subcategories emerge in the third network. All codes have an associative relationship with their subcategories, as they are linked to both racist and hateful labels against the Islamic religion. The latter are linked to the first level category, "reasons for expressing hatred", as a way of justifying such discrimination.
The second part of this fourth network (see Figure 9) shows how three subcategories emerge from the codes. In the first, refugees are shown as a 'problem for the national economy'; in the second, there is an appeal to 'nationalist exaltation'; and finally, in the third, hatred is spread towards 'Spanish entities that support emigration'. This means that their codes express economic problems (i.e., looters) or discourse of national exaltation (i.e., nation). Subsequently, there is the subcategory 'Spanish entities that support immigration', where their codes have a possession connection, i.e., both codes are organisations that defend immigration. This is related as a way or justification by which hatred is expressed, linked to the main category that has emerged in the network.
Finally, Figure 10 shows the different codes that give meaning to and are associated with the second-level category 'reasons for the country's security'. This entire volume shows the different emerging reasons that create a homogeneous discourse. This is reflected as a subcategory, forming a reason why hatred is expressed and justified on the social network Twitter.
In short, the three networks observed show the six subcategories defined as the reasons why hate is expressed: (1) racism, (2) hatred against religion, (3) perceived threats to the economy, (4) nationalist exaltation, (5) rejection of Spanish entities that support this immigration, and (6) perceived threats to the country's security. These subcategories that emerge from the analysis of the discourse disseminated on Twitter are closely related to the classification proposed by Santos-Alvarado (2021). The networks obtained in this research show the clear importance of discourses linked to refu- Figure 8. First network of reasons why hate speech is expressed Figure 9. Second network of reasons why hate speech is expressed gees as a threat to security. In Figure 10, the subcategory that identifies these refugees from the south as a problem for security is the one with the highest density (number of related codes). The threat to the economy as a justification for hatred against refugees from the south is another emerging category, which is the second highest in qualitative density. Finally, the threat to national identity, as a variable of analysis and as defined by Santos-Alvarado (2021), encompasses two subcategories: hatred against religion and national exaltation. Both views seek to identify refugees from the south as a problem for cultural values, as they are labelled as Muslim and radical people who threaten national Christian values (see Figure 11). This approach is one way in which Ukrainian refugees are justified in receiving more favourable treatment since they have close identity ties to nationals in terms of religion, physical appearance, and other characteristics.
All this development of refugees as a threat is linked to the findings of Valdez-Apolo, Arcila-Calderón and Amores (2019), whose study found that, in general terms, a representation of these displaced people as a threat clearly predominates. These authors establish the threat in the same areas (economic, security, and identity) that have been detected in this study. The danger in the economic sphere is that refugees are seen as a burden on national finances, given their social class. This is a clear sign of aporophobia, the hatred of poor or low-income people.
The threat to cultural identity has a strong symbolic value, strongly linked to the Islamic religion in our study. However, the idea of refugees as a security problem has a more realistic character, i.e., that the threat poses a risk to the integrity of the country and people, as they are portrayed as aggressive or criminal. As it affects something so direct and close to people's reality, it is probably the discourse with the highest density, as established by the research results. Valdez-Apolo, Arcila-Calderón and Amores (2019) establish that the security of a country is the issue that generates the most hatred since it affects what is most sensitive: the supposed physical integrity and the very life of the citizens of the country hosting the refugees.
The blaming of nationals has a lower, but still remarkable, density (number of codes linked to the subcategory). As Santos-Alvarado (2021) argues, this discursive typology is essential to understand social polarisation as a symptom of the rise of hate speech. This last detail is evident in Spanish society according to Ondod (2021) data, where hate crimes have increased by 43.6% since 2013.
Lastly, we find racist hate speech, which has a low density. This low frequency is directly related to one of the issues raised by Valdez-Apolo, Arcila-Calderón and Amores (2019), where explicit racism is politically incorrect, such as that analysed in the study (for example, apes, whites versus blacks), and is largely rejected in advanced societies. Therefore, this racist discourse often tends to be diluted or disguised with other discursive typologies, such as relating the refugee as an economic or security threat. On other occasions, the tweets themselves claim not to be racist and that they only seek to defend European refugees against those they do not consider refugees, but only illegal immigrants without any right to enter the country (see Figure  12). This is clear evidence of the finding of Valdez-Apolo, Ar-

Summary and relationship between categories and subcategories
Once the results obtained in accordance with the objectives set out have been presented, we proceed to develop the relationship between the categories and subcategories. To this end, the comments, quotations, and memos elaborated throughout the analysis have been considered. As a result, the networks that emerge from the categories and subcategories and the type of relationship between them are shown in Figure 13. It is clear how the motives for hate speech are associated with the rest of the refugees. In addition, they are linked to negative attitudes/values and emotions/feelings. Conversely, Ukrainian refugees are characteristically associated with positive attitudes/values and emotions/feelings.
Despite being considered an axial coding process, the last phase of coding, called selective coding, is already being established. This last stage is not independent of the previous two but rather is an extension of them with a higher level of abstraction (San-Martín-Cantero, 2014). This allows us to develop a theory about refugee discourse on Twitter based on the last network obtained after the axial process. Following Strauss and Corbin (2002), our final network, composed of the relationship between the subcategories and categories, allows us to establish the central category that will define the theory of our study of the discourses.
The resulting theory on selective coding emphasises that there is a differentiated treatment of refugees on Twitter.
Ukrainians are associated with certain characteristics (i.e., family, women, and children), as well as a clearly positive focus on attitudes and values (i.e., humanity, order) and emotions and feelings (i.e., tenderness, love). Other refugees receive an opposite discourse, linked to negative values and attitudes (i.e., danger, disorder) and negative emotions and feelings (i.e., rage, anger). The refugees coming from the south receive a series of hate speeches motivated by various justifications: (1) because of their religion, (2) pointing the finger at Spanish entities to blame for this immigration, (3) racism, (4) problem for the security of the country, (5) problem for the economy and (6) national exaltation.
This typology of discourses shapes the various ways in which Twitter users find justifications for hate speech against refugees from the south. This reflects the urgent need to carry out media literacy training aimed at all citizens from an emancipatory and transformative approach (Barbas-Coslado, 2012). Consequently, it is necessary to rethink the models of digital skills training since, as Gutiérrez-Martín and Tyner (2012) argue, they are often reduced to the technological dimension, where the focus is on the instrumental control of the programmes.

Conclusions
This study has demonstrated the dualism in the treatment of refugees and hate speech towards groups from the South on the social network Twitter. This work is a methodological and scientific novelty, as computer-mediated discourse analysis has made it possible to capture and analyse informal and colloquial content that users of this social network share in their posts in an immediate and probably not very reflective way. The normality with which hate is poured out on social networks is not an innocuous phenomenon, given that hate crimes in Spain have grown exponentially, making it an enormously dangerous phenomenon for social coexistence.
In this way, it should be recalled that the victims of this hatred must be the focus of the debate, as they are the ones who suffer the violation of the most fundamental human rights, such as the right to life, security, and non-discrimination.
In this regard, Spain, from an institutional and legal point of view, is showing significant concern for this phenomenon. The creation of the National Office for Combating Hate Crimes Ondod, which reports to the Spanish Ministry of the The victims of this hatred must be the focus of the debate, as they are the ones who suffer the violation of the most fundamental human rights, such as the right to life, security, and non-discrimination Interior, was an important step forward in this regard in 2013. Regarding this same body, it is necessary to highlight the annual reports that show the increase in figures, as well as the recently approved II Action plan to combat hate crimes (2022)(2023)(2024), where the treatment of the victim of hate crimes becomes the centre of the strategy of action. Added to this is the new Organic Law 15/2022, comprehensive for equal treatment and non-discrimination, which establishes a special emphasis in articles 22 and 53 on aggressions or actions that incite hatred through social networks.
However, the regulatory and judicial structure of the state is not enough if it does not consider the training of citizens in what is known as media education or media literacy, especially regarding the critical and ethical use of the media, which will make it possible to combat phenomena such as disinformation.
With respect to the spread of disinformation on networks and the urgent need for media literacy, this study has detected the need to provide specific training on the analysis of audiovisual content in order to help citizens understand the use of the audiovisual world in the media. This concept is becoming more important, especially in social media, which use and sometimes abuse visual content as a format for information and on other occasions for disinformation. If this approach is not considered, citizens are condemned to become easily manipulated subjects.
It should be noted that this research aims to detect these needs in the digital training models of our society, as well as the fight against misinformation, especially that which involves hate speech. Failure to confront hate and the manipulation of information and communication media means normalising discrimination against many individuals and groups. This is consistent with the so-called Sustainable Development Goals (SDGs) and, in particular, SDG 16, 'Peace, justice and strong institutions' (UN, 2015). The aims are to promote the development of both peaceful and inclusive societies and create justice accessible to all, with an emphasis on the need for effective institutions to meet these challenges.

Limitations and future lines of research
The non-probabilistic sampling of the research cannot guarantee control over how the sample is constructed. Furthermore, the number of tweets that make up the sample may not be representative of hate speech about refugees on Twitter and therefore may be inadequate to generalise. For these reasons, this study should be considered an exploratory approach that can be a starting point in the study of this phenomenon. There are some limitations associated with the characteristics and functioning of the Twitter platform. The anonymity of the accounts does not allow us to know which individuals, groups, or bots are behind these types of publications. The social network Twitter usually uses tags that categorise posts, but given that only the content of two tags (#Refugees and #NotRefugees) has been analysed in this study, it is likely that not all tweets conveying hate speech towards refugees have been accessed.
With respect to possible future lines of research, continuity with the theme of this work is suggested. The main idea is to replicate the study but bear in mind the need to overcome the limitations indicated. Enlargement of the sample size will be fundamental.
It is worth pointing out the possibility of continuing to study hate speech in the media towards other social groups, such as Unaccompanied Foreign Minors, the LGTBIQ+ group, and members of particular religions, ideologies, and social classes. Another possible future line of research is the analysis of why citizens become involved in broadcasting this type of discourse, as this would be very useful information for designing prevention proposals aimed at minimising this phenomenon.
In short, this article suggests great possibilities for research to try to detect the needs of citizens in the face of the challenges posed by the information and communication media, especially with respect to disinformation.