Introduction

The episodes and events that follow migration, mainly Central American, on its way to the northern Mexican border as a waiting destination and on the way to the United States, constitute, as it also happens in other latitudes, media events that partake in the structuring of a dual discourse prone to polarization. Alongside the narrative of support and solidarity that usually governs the discourse on migration in traditionally expelling countries, it is added, rather as a consequence of the geography of the migration flow, a conflictual rhetoric typical of developed nations that receive migrants. In these contexts of public discourse production, we can ask ourselves if the crisis usually accompanying transnational migratory transit, as in the case of the caravan of transit migrants through Mexico, holds certain specificities and/or similarities with countries with a migratory tradition.

The migrant caravan in transit through Mexico is a conjunction of massive migration events having their conceptual and organizational affiliation within the framework of the ordeals that jumped into the public eye as of 2014 (^{Martínez Hernández-Mejía, 2018}). As a media event, the migrant caravan came to public light during October 2018, putting into the spotlight the journey through Mexican territory towards the United States of several waves of migrants in transit mainly from Central America (^{COLEF, 2018}). Both the lack of capacities to process this important human flow under ideal conditions and the toughening of admission, transit, and stay policies, quickly turned this event into a widely mediated migration crisis (^{Arriola Vega, 2016}; ^{Albicker et al., 2018}; ^{COLEF, 2018}, ^2019a, ^2019b).

Crises in general and those related to migration problems specifically are events that promote the emergence of public discourses and their dissemination through various media channels that allow the propagation of competitive argumentation from the various confronted points of view (^{Boomgaarden & Vliegenthart, 2009}; ^{Khiabany, 2016}; ^{Igartua & Cheng, 2009}; ^{Iyengar & Simon, 2000}; ^{Nail, 2016}). One of the central consequences of the media instrumentation of the migration crisis consists in the polarization of the public debate and the segmentation of its actors into antagonistic groups that favor the escalation of discursive rhetoric (^{Boomgaarden & Vliegenthart, 2009}; Iyengar & Simon, 2000; ^{Nail, 2016}).

Yet this does not mean that all Twitter discussion is monopolized by contending groups since other dimensions of the debate are generally displaced to the periphery due to the overflowing passions of antagonism. In these contexts, if the media happen to be outside the place of origin of the content that inflames the debate, ^{Igartua and Cheng (2009}) assign to them a clear responsibility in structuring the beliefs and behavioral attitudes of the recipient. In this sense, for ^{Vliegenthart, Schuck, Boomgaarden, and de Vreese (2008}) the news provided by the media become a transcendent source that makes it possible to form criteria when knowledge and/or similar experiences are lacking.

From another perspective, the dynamics of technological change in the media landscape sometimes manage to redistribute the cards in a pertinent way among the different factors that participate in the creation and dissemination of public debate. In this sense, from the point in which the internet entered the public sphere, a profound mutation of the detonating means of the accessible, transnational and articulated public debate has been witnessed, going from the face-to-face to the virtual modality (^{Bouvier, 2019}; ^{Paulussen & Harder 2014}). Within the framework of this dynamic, social networks quickly managed to position themselves at the center of the media landscape, getting involved in transcendent controversies as in the case of the migration debate (^{Ferra & Nguyen, 2017}; ^{Koylu, Larson, Dietrich & Lee, 2018}; ^{Nguyen 2016}). As a media technology, Twitter manages to stand out globally for its specificity aimed at instant interaction through short content, making this platform the favorite one of rulers and the ruled alike (^{Bruns & Stieglitz, 2014}; ^{Meraz & Papacharissi, 2013}; ^{Tandoc & Johnson, 2016}). We can see this in how strategic Twitter has become in the communication of President Trump and several other opinion leaders from different fields, both in developed and emerging countries. As for other social networks, even though Facebook has a much larger audience, Twitter is still the preferred one when it comes to political communication. Regarding the public debate on migration, Twitter is generally the ideal media context to “cross swords” between anti-immigrants and pro-immigrants in their desire to construct/deconstruct rhetoric discourses oftentimes conflictual and extreme (^{Ferra & Nguyen, 2017}; ^{Karatzogianni, Nguyen, & Serafinelli, 2016}; ^{Nguyen 2016}).

According to the survey of the Internet Association of Mexico (^{AIMX, 2019}), in 2018 Mexico had, in the case of people over 6 years of age, a social penetration of the Internet of 71%. Of this universe, 82% use the Internet to access social networks, and 49% specifically to access Twitter. In quantitative terms, these figures reflect a partial use of Twitter by the Internet user population; furthermore, as evidenced in other contexts, it can be suspected that this use is equal between different demographic and social groups (^{Sloan & Morgan, 2015}). However, the fact of subjecting this research to the universe of Twitter users is compensated, above all, by the efficiency of this platform in the online monitoring of the migration crisis (^{Sloan & Morgan, 2015}; ^{Stephens & Poorthuis, 2015}).

This article analyzes the content of the messages disseminated via Twitter related to the facts and events of the migrant caravan in transit. Thus, only the public debate developed on Twitter is analyzed. This interest stems from a shared perception among the different observers of this migratory phenomenon, who pointed to social networks as linked to a radical anti- immigrant public discourse. To delve into these assertions, an exploratory goal was set aimed at systematizing and classifying the content published on said network, based on the feelings expressed by the messages sent. From another complementary perspective, to give shape and substance to the media strategies involved, the classified contents were related to the socio- professional and political logics of their authors, their geographical origin, and the dynamics of dissemination. A comparative approach was chosen to assess the findings of this analysis, making use of case studies of thematic and epistemological similarity.

Theoretical framework

Migration crisis and public discourse on twitter

With the mass use of the internet and social networks, the impact of national and regional events in the construction of public discourse acquired new dimensions for the transmission of news through different media. Within the framework of this media diversity, in which the analogue is mixed with the online modality, dynamics of participating actors and forms of interaction are generated for new logics of creating a structured public discourse around major issues and controversies, such as the case of the debate on migration (^{Ferra & Nguyen, 2017}; ^{Koylu et al., 2018}). In this new media landscape that allows citizens and their elites to dialogue with each other on the sidelines and/or in synergy with the mass media (^{Paulussen & Harder 2014}), social networks seem to open alternative paths to produce news, events and express both concurrent and competitive opinions (^{Bouvier, 2019}). As for Twitter specifically, regardless of its social penetration, the opinions expressed emerge as a fusion of sentiments projected in intentions and behaviors (^{Koylu et al., 2018}; ^{Yaqub, Chun, Atluri, & Vaidya, 2017}). It is perhaps this behavioral perspective that is usually prone to the motivation that gives this tool its strategic dimension given its transitional power from the real to the virtual and vice versa.

Online Public Discourse on Migration

The incursion of the public debate on migration in public online media is characterized by a reconfiguration of the public discourse that opens its transnational perspective to the participation of different types of actors and organizations from different locations around the world (^{Ferra & Nguyen, 2017}). Our research appears to show that, compared with traditional media, in the online mode there is little possibility for the creation of a migration narrative emanating directly from migrants, and that this initiative is generally monopolized by other actors (^{Karatzogianni et al., 2016}). However and independently of the aforementioned, the media landscape involved in online public discourse transcends the narratives of the national context, achieving international reach (^{Nguyen, 2016}). Sharing concerns and challenges between actors in a dialogue from different sociocultural poles, enables the development of irreducible divergences in less receptive media, due to the reach and limits of the technology used in media interaction (^{Ferra & Nguyen, 2017}; ^{Nguyen, 2016}).

The importance and effectiveness of online public discourse, according to ^{Ferra & Nguyen (2017}), is to be a reflection of the perceived socio-political and economic interests that fuel the emotions, intensity, and propensity to radicalize the debate on migration. In this sense, once expectations increase in an online debate, as ^{Kingsley (2015}) mentions, the possibility of lowering the level of confrontation becomes futile when what is tragic for some can easily become a source of ridicule for others. In these kinds of heated debates, ^{Karatzogianni et al. (2016}) found that the intervention of pro-immigrant groups is not strong enough to mitigate the incidence of the dehumanizing narrative that accompanies the rhetoric of anti-immigrants. In this sense, the convenient online link of the migration situation to the economic crisis and other difficulties that mainly affect low-income layers of the population constitutes one of the central axes for the strengthening of anti-immigrant rhetoric that is difficult to deconstruct.

Within the framework of online discourse, and due to its technological specificity and reticular structure, social networks play a central role in linking various platforms and content capable of involving different types of users. Among so many social networks that compete in their different focuses to attract a greater number of users, Twitter stands out due to its communicational singularity focused on the creation and linking of short content (^{Bruns & Stieglitz, 2014}). As a whole, its technological and communicational attributes and its ability to position itself as a tool for participation and democratization make it the preferred social network for protest and the release of political agendas (^{Meraz & Papacharissi, 2013}). From another perspective, ^{Tandoc & Johnson (2016}) assert that the ease and speed in the distribution of information give Twitter a comparative advantage when it comes to communication-related to crises and conflicts.

Sentiment Analysis as an Approach to Conflict Level

One of the approaches that allow us to examine the discourse on social networks and specifically its conflictual perspective, consists in the application of sentiment analysis procedures. In the case of Twitter, sentiment analysis was used to monitor governance and public relations tasks (^{Tumasjan, Sprenger, Sandner, & Welpe, 2010}), in the electoral positioning of political candidates (^{Abroms & Craig Lefebvre, 2009}), in electoral prediction (^{Avello-Gayo, Metaxas, & Mustafaraj, 2011}), in the study of information dissemination patterns (^{Ferrara & Yang, 2015}) and the analysis of presidential electoral campaigns (^{Yaqub et al., 2017}), among others. Sentiment analysis characterizes a specialization of computational studies focused on examining at the social network user level the opinions, emotions, and moods expressed with texts, images, and videos (^{Koylu et al., 2018}).

From a technical perspective, the scope and limitations of sentiment analysis mainly gravitate around semantic variations and other linguistic devices, which can yield different results depending on the different socio-cultural contexts. ^{Liu (2012}) prepared a review of these technical limitations, also documenting the mechanisms used to minimize their impact on the analysis of opinion and sentiment. Generally, sentiment analysis processes are carried out automatically, or with the intervention of an analyst who consolidates different steps and learning patterns of the algorithms employed (^{Koylu et al., 2018}). Properly said, the analysis can be approached from one of three possibilities: an examination at the document level, at the sentence level, or the entity level (^{Koylu et al., 2018}; ^{Tumasjan et al., 2010}). Globally, the result of a sentiment analysis yields a result in three main classes: negative, neutral, and positive, indicating the mood of the analyzed message. Yet there are also variations based on the subjectivity of the posting and on the reactions that translate an emotional state perceived through the message sent (^{Koylu et al., 2018}; ^{Zimmerman, Stein, Hardt, & Vatrapu, 2015}).

In the case of Twitter, it is estimated that a rate higher than 80% of the trending topics discussed come from columns and news that are at the heart of the news event of the moment (^{Cheng, Adamic, Alex Dow, Kleinberg, & Leskovec, 2014}). In this sense, the analysis of Twitter sentiment at different moments of the news cycle allows visualizing the evolution of trends in public debate, thus opening the possibility to make adjustments in the subsequent processing of information almost in real-time (^{Yaqub et al., 2017}).

Data and methodology

The present work focused on analyzing the messages issued exclusively via Twitter about the migrant caravan in transit, mostly from Central American territories. This analysis does not cover the entirety of this migration phenomenon in time, which began in October 2018 having its main destination in the city of Tijuana, Baja California (^{COLEF, 2018}). Indeed, the thematic and epistemological interest in undertaking this research arose as a consequence of the social and media interactions that accompanied the temporal and territorial transit of the first caravan. Also, it was not feasible to purchase the historical data of more than eight days, given that obtaining it is conditional on a good practice policy that limits free access to real-time retrieval. Under these limitations, the period of our study begins in January 2019, which corresponds to the gestation of the new migrant caravan that left for the United States from San Pedro Sula, Honduras, on the 15th of the same month (^{COLEF, 2018}).

The tweets upon which this research is based were collected through the programming of a Python script applied online to the application programming interface (API) for Twitter transmission. The Twitter API allows tweets to be filtered by different search methods, such as the use of keywords and directing results, almost in real-time, to the authors of the queries (^{Lomborg & Bechmann, 2014}).

Based on a previous pilot study that allowed to verify and adjust the search parameters, tweets were collected without interruption for a period of 46 days from January 1, 2019, to February 15 of the same year. To focus exclusively on our subject of analysis, we chose to collect tweets written in the Spanish language with messages that contained the words migrant and caravan together. This message-filtering strategy allows to substantially increase the probability that the collected tweets are related to the issue of the migrant caravan at the local and regional levels (^{Abroms & Craig Lefebvre, 2009}). However, this does not guarantee to capture all messages on the matter. From a nomological perspective, messages not expressed in Spanish that refer to our topic, and those that use only one or neither of the two search words, will be left out of reach.

On the other hand, the proportion of tweets collected in relation to those issued by users is another factor to take into account when validating the level of reliability of the collection (^{Bruns & Stieglitz, 2014}; ^{Morstatter, Pfeffer, Liu, & Carley, 2013}). In this sense, authors such as ^{Gerlitz and Rieder (2013}) consider that even a sample of 1% is sufficiently representative of the total tweets issued. In our case, the collection was much greater, since 1'805,814 messages were recovered throughout the analysis period, which constitutes the entire universe produced under the search parameters chosen for the investigation. Depending on the type of analysis and the calculation capacity available for it, sometimes the total volume of messages collected was used and in others, a sample of 5% was randomly chosen, that is, 90,290 messages representing five times more than what is considered by ^{Gerlitz and Rieder (2013)} as sufficient. It is important to mention that Twitter's discursive volume grew from September 2017 due to the doubling of the number of authorized characters in a post, which went from 140 to 280.

From an operational perspective, the application SentiStrength allowed us to perform sentiment analysis of the collected tweets. This application performs a content review and then assigns each of the messages a positive score, ranging from 1 to 5, and a negative score from - 1 to -5 (^{Thelwall, Buckley, Paltoglou, Cai, & Kappas, 2010}). In this sense, a score close to 5 would indicate a highly positive message, and -5 would be very negative. Similarly, a score between -1 and 1 would indicate a message with neutral content. In our case study, and to give a single score to each of the messages analyzed, we followed the procedure described by ^{Yaqub et al. (2017}), which consists in adding the negative and positive scores. Accordingly, the result of the addition would indicate a positive message when fluctuating between 2 and 5, a neutral one between -1 and 1, and a negative one between -2 and -5.

The sentiment assessment mechanism that was carried out using the SentiStrength application was used in several studies on social networks (^{Ferrara & Yang, 2015}; ^{Yaqub et al., 2017}). This application, developed specifically to assess sentiment in short texts, in its Spanish has been distinguished by the reliability of 74.3% under the unsupervised method based on the lexical dictionary, and of 78.8% using a syntactic backup (^{Brooke, Tofiloski, & Taboada, 2009}; ^{Vilares, Alonso, & Gómez-Rodríguez, 2015}). In this case, the lexical instruments in Spanish contributed by ^{Vilares, Thelwall, and Alonso (2015)} were used, making modifications that allowed them to be adapted to the regional linguistic context of our research.

In addition to the characterization of the messages by type of sentiment, the contents collected were defined by four other attributes: a) the type of message, b) the number of followers of the author, c) geographical origin and d) socio-professional profile. The type of message defines whether the collected postings are tweets or retweets and are obtained together with the number of followers through the Python script in charge of establishing communication with the Twitter API. In our case, retweets were discarded for sentiment analysis when their messages duplicated a tweet considered in the same process.

The geographical origin of the posting was preferably generated with the data of the geographical services when this function was activated by the Twitter user and, in the opposite case, the geographical tags present in the metadata of the profiles were used as long as these were filled out. As is generally the case on Twitter, only 1% to 3% of users activate their geographic services and the tagging available in profiles is not always accurate (^{Takhteyev, Gruzd, & Wellman, 2012}). This limitation possibly reduces the reliability of the information collected; previous research indicates that despite these deficiencies, the representativeness of the resulting data, due to its complexity, is still an unresolved issue (^{Sloan & Morgan, 2015}; ^{Stephens & Poorthuis, 2015}). In our case, it was possible to locate 59.24% of the postings collected at the country level, and at the level of Mexican states 46.70% of the messages located at the country level, representing 16.01% of total postings.

The socio-professional profile of the authors of the posting was determined by means of a human review operated on a selection of the 100 authors with the largest posting during the study period. This selection represents 47.96% of the total posting and the authors were assigned to eight types of socio-professional profiles: Political activism groups, political activists, Twitter activists, news media, journalists, government officials, government bodies, NGOs, and others.

Results and discussion

As has happened in other social networks, Twitter was the recipient of an extensive exchange resulting from the issue of the migrant caravan in transit in its different discursive, temporal and territorial contexts (^{COLEF, 2018}). From the perspective of the other media and links of public discourse, as this event developed the perception that social networks have been carriers of negative semantics vis a vis the event and its actors crystallized. Similar to what was observed on Twitter during the European migration crisis (^{Ferra & Nguyen, 2017}), the displays of an extreme discourse towards the migrant caravan were also impregnated with xenophobic rhetoric, both in face-to-face and virtual modes (^{COLEF, 2018}, ^2019a). In this sense, the expressions of empathy and solidarity towards migrants occurred almost at the same time as the manifestations of dehumanization and violence (^{COLEF, 2018}, ^2019a, ^2019b). From this perspective, these manifestations, translated into a measurable dimension, perhaps constitute a first approach towards understanding the narrative structure of the migrant caravan. AS for this research, the virtual axis of this narrative and specifically the discourse that emerged from it was focused on Twitter with the support of the categorization of the sentiment expressed in the published postings.

Taking together what was published during the study period, it was noticed that 85.35% of the analyzed tweets reflected a neutral discourse, 5.50% were positive and 8.64% were negative. This type of sentiment composition reflects at least two possible implications when what is published on social networks is perceived in one way or another as negative. On the one hand, it is possible to hypothesize a hypersensitivity to the negative narrative, even though it is marginal, and on the other, it is possible to think of a significant impact capacity of negative discourse inside and outside of social networks. Nonetheless, beyond these research leads, any attempt at explaining this may require taking into account the dynamics of Twitter content creation in times of confrontation or crisis. Indeed, if the 2016 U.S. presidential campaign is taken as a comparative point, which was considered the most negative of its kind, the proportion of neutral tweets was also the most important, followed, as in our case, by negative postings, and then by the positive ones (^{Yaqub et al., 2017}). However, the proportion of the

company’s neutral tweets is about 30 points down and the negative ones about 25 points up than in the case of the migrant caravan: why are these differences due?

A possible explanation can arise from the levels of polarization of the discussions and the amplitude of the mobilization of actors in a confrontation that nurtures said discussions. Indeed, the prohibition of entry into U.S. territory of citizens from seven “Muslim countries” during 2017, was reflected in a strong confrontation, with high rates of positives and negatives, even in the states that elected President Trump by large (^{Koylu et al., 2018}). In this particular context, ^{Koylu et al. (2019)} indicate that the distribution of positives and negatives was the same before the president's executive order, which came only to expand the number of participants in this polarized discussion. To this, it is important to add that the level of online polarization of migrant rhetoric is similarly supported by the face-to-face mode, which provides it with a narrative style and guides for discursive confrontation (^{Zimmer & Proferes, 2014}). In this sense, as mentioned by ^{Conover et al. (2011}), the polarization of the debate on Twitter largely translates to competitive discussions generated within the framework of the political preferences of the debaters.

In our case study, the context and logic of the immigration debate are different. Mexico is a country with a significant migrant population to the United States, a situation that traditionally encouraged the development of a narrative based mainly on claims for the treatment and conditions of fellow citizens in this neighboring country (^{Yrizar Barbosa & Alarcón, 2010}). This context also places the national migration discussion in the sphere of foreign policy, which has a low power of sociopolitical polarization, compared to issues related to the internal agenda, which tend to ignite public debate inside and outside of social networks (^{Bobes León & Pardo Montaño, 2016}).

In this sense, we can see that the polarization in the face of the migrant caravan in transit does not meet the conditions that allow for a discursive tension at the same level as the migration crisis in Europe, or when Trump’s executive order was issued, two events that were analyzed by ^{Ferra and Nguyen (2017}), and by ^{Koylu et al. (2018}), among others. Indeed, in the case of the caravan, except for some public figures such as the mayor of Tijuana, who tried to cultivate an anti-immigrant narrative (^{COLEF, 2018}), an easy tone generally prevailed in Twitter exchanges, as indicated by the findings of this research.

This type of behavior also stands out in the content forwarding rate (retweets) in relation to content created (tweets), which reflects certain utilitarian manifestations made evident in other studies (^{Tumasjan et al., 2010}). In our case, 55.30% of what was published on Twitter was the forwarding of previously created content (tweets), this proportion is about 15% less than that reported by Yuqub et al. (2017) for the last campaign of the U.S. presidential election. These authors, as in ^{Ferrara and Yang (2015}), also found that positive tweets are more likely to be forwarded. Regarding the migrant caravan, comparable rates were found between tweets and retweets for negative and neutral messages, while for positive ones, forwards were twice as important as content created. These findings once again confirm a debate on Twitter similar to that observed in the other crises, however, with low tension and intensity of confrontation.

The origin of the posting on Twitter can also be an indication of the sentiment permeating the messages sent. Indeed, in accordance with findings from previous works (^{Cheng et al., 2014}; ^{Ferra & Nguyen, 2017}; ^{Yaqub et al., 2017}), our study shows dominant participation of the news media with 54.05% of the total postings by the top 100 most prolific authors (see graph 1). This proportion reaches 62.60% if we add the postings by solo journalists, which is 27 times more than what is published by non-governmental organizations, 15 times more than the government bodies and political activists, and eight times more than what is published by officials and politicians in election posts.

The polarization generated by the media also translates into a certain balance between positives and negatives in postings, an aspect that does not seem to be achieved by other actors. Indeed, keeping the proportions, the most positive authors seem to be officials and politicians in elected positions followed by Twitter commentators, while the most negative are NGOs and other organizations, followed by political activists. This seems to translate into a double logic of content production on the migrant caravan in transit. On the one hand, a logic of political positioning is embodied by the authors of the protesting negative postings, and on the other, a nuanced and pragmatic logic expressed by actors aware that positive messages are disseminated and reproduced the most. Between these two logics, the news media seem to keep a moderate and neutral position that is more conducive to creating mass media events. In this sense, contrary to Twitter commentators, everything seems to indicate that the news media and journalists who report and communicate information on the caravan have favored a professional content less prone to confrontation.

With content focused on the actual news, the dominant messages are directed to the local population, informative, and written with a tone alike the following: Migrant Caravan in transit from Hidalgo City to Tapachula *Be extremely cautious when driving your vehicle in that area*. We also exemplify a posting published on Twitter aimed at a more inclusive audience of different territorial scales:

THE FIRST MEMBERS OF THE NEW HONDURAN MIGRANT CARAVAN BEGIN TO

ENTER. The PNC has already deployed the security plan 3-2019 called “Security Device for the Mobilization and Return of Central American Migrants in Guatemalan Territory” [highlight in capitals from the original message].

These messages break with the content dynamics of some columnists who refer in their messages to other textual or audiovisual content in which opinions and interpretations of the news are expressed.

1. Political activism groups, 2. News media, 3. Political activists, 4. Twitter commentators, 5. Officials and politicians in popular election posts, 6. Government bodies, 7. Journalists, 8. NGOs and other organizations. Source: Own elaboration based on data from the study.

Graph 1. Postings by the 100 Most Prolific Authors, by Place of Origin 

Another dimension of the structuring of the discourse on Twitter is its transnational perspective, which involves actors from socio-political and cultural contexts other than national authors (^{Ferra & Nguyen, 2017}; ^{Koylu et al., 2018}; ^{Nguyen, 2016}). As in ^{Ferra & Nguyen (2017)}, in our study, these actors include large conglomerates and international media (for example, @CNNEE, @Rtultimahora, @CaraotaDigital, @notibomba, among others) that were important both for the creation and the structuring of the public discourse on the migrant caravan on Twitter. Alongside these users, there are also commentators and activists who, in addition to dominating the discourse, seem to sometimes saturate with repetitive content (for example, @KatuArkonada, @DeniseDresserG, @AleLadyBird, and @ConElMazoDando, among others).

Taking into account the difficulties and limitations in determining with certainty the geographical location of Twitter users, only a third of the messages about the migrant caravan came from Mexico, 6.13% from Honduras, 4.29% from Guatemala, 3.50% from Venezuela 3.03% from Spain and the rest from different countries of the world (see graph 2). As ^{Ferra and Nguyen (2017}), and ^{Nguyen (2016)} comment, in our case also the public debate on Twitter seems to exceed the landscape of national authors, asserting actors with the ability to significantly influence the rhetoric and meaning of public discourse. From this perspective, in some postings, entirely concerted rhetoric can be perceived, as in the case of the content sent from Germany and Russia, which was all neutral. In the case of the messages sent from Brazil, Uruguay, and Panama (to a certain extent, also from Spain) the contents oscillate between neutral and negative, while the positive ones are always involved in a posting logic in conjunction with neutrals and negatives. The latter express, as in the case of the study by ^{Koylu et al. (2018}), a relatively contradictory and conflictual narrative as it usually occurs in a competitive public Twitter debate.

Source: Own elaboration based on data from the study.

Graph 2 Postings by Country of Origin 

Closing remarks

As the use of social networks intensified, new contexts and technological platforms were opened for the creation and dissemination of a public discourse intimately articulated with the face-to-face media. Within this framework, the treatment of crises and confrontations between opposing points of view constitutes an energetic catalyst for a new real-time, transnational, and relatively participatory media landscape. Both for the construction and the deconstruction of media narratives, social networks (Twitter, specifically) became an important tool to detonate and distribute segments of a public discourse that is both concordant and competitive.

One of the most prolific contexts for the creation and dissemination of public discourse in developed countries is surely the issue of the conditions of entry, transit, and stay of migrants. Lately, due to the geographical nature of migration flows, several traditionally sending countries have found themselves involved in migrant transit crises and conflicts. These contexts seem conducive to mobilizing the media landscape, triggering a structured public discourse about migration rhetoric. In this sense, the public debate ignited by the migrant caravan in Mexico constitutes a characteristic example of a new discursive dimension of a media landscape articulated between the real and the virtual.

As a media technology platform, Twitter became an important host for the creation and instantaneous dissemination of news, opinions, and controversies that followed the migrant caravan on their way to the United States. During this event, along with the demonstrations of support and solidarity, a consistent perception was also generated that social networks hosted negative and extreme content, similar to what happened in recent events of the European migration crisis. From this perspective, this research arises from the concern to quantify, characterize and explain the sentiment of the messages spread during as the migrant caravan was in course.

This study focused on the analysis of the discursive dimension generated on Twitter from January 1, 2019, to February 15, 2019, within the framework of the migrant caravan. Broadly speaking, the study's findings challenge to a certain extent the perception that social networks had a negative treatment vis a vis the events and actors of the migrant caravan in transit. When it comes specifically to Twitter, although the negative messages slightly outnumber the positive ones, most of the content was rather neutral (85%). Almost the same emotional message structure was found as in the last presidential election in the United States, which was characterized by a higher proportion of negatives, which translated into higher levels of polarization and conflict than in the case of the migrant caravan. Indeed, compared to the U.S. public discourse on immigration, the development of migration rhetoric in Mexico and its political use do not seem to rival that of countries traditionally receiving migrants. This low intensity of discursive confrontation is reflected to some extent in a high rate of retweets (55%), which is usually linearly related to the rate of positive tweets. This link between the wide dissemination on Twitter and the positive sentiment of the content seems to be connected, as in other developed countries, with a dominant and pragmatic intervention of the news media and journalists (about 63% of the messages) that seek greater distribution and reproduction of the news.

From the perspective of the transnationalization of public discourse on the migrant caravan on Twitter, 34% of the messages were sent from Mexico, surpassing by several percentage points Honduras, the country where the caravan originated. The rest came from various regions and continents, with a clear posting of homogeneous and arranged content blocks from certain identified locations. This perspective and others from abroad, in addition to broadening the national media landscape and its contents, also allowed segments that correspond to other contexts of interaction with the crisis and the level of migration conflict to be injected into the local public discourse.

However, regardless of the impact on Twitter of transnationalization in the construction of the local migration narrative, in the case of the migrant caravan, everything seems to point out that a neutral tone prevailed, generally. The latter seems to reflect a local public discourse recently faced with the reality of a migration crisis in a context of relative instrumentalization, mainly by political actors.

The findings of this analysis generate theoretical and empirical impacts that contribute to the construction of knowledge about the discourse on the migration crisis in social networks in general, and Twitter in particular. From a theoretical perspective, this study clarifies that the migration debate on Twitter follows a similar discursive structure in countries with a migratory tradition such as Mexico, a nation characterized by transit migration to the United States, mainly. Broadly speaking, this research corroborates the clear predominance of a neutral tone in the content posted, showing a significant gap between the perception of a migration crisis discourse on social media and a minority of negative messages, at least on Twitter. From this perspective, the migration discourse on the caravan is far from reflecting a homogeneous corpus in a single direction, rather following a determinism that is geographical, professional, and according to ideological commitment.

From the empirical research perspective, the study was made possible thanks to the development of computer programs compiled in Python to carry out the search and systematization of the information. Likewise, a linguistic adaptation was carried out for SentiStrength to be able to analyze Mexican and Central American contents. As a whole, our findings make it possible to reinforce concerted strategies to combat anti-immigrant hate speech, mainly opting for practices that do not allow expanding its reach into traditional media. From yet another perspective, the characterization of the sources of hate in social networks can provide grounds for containment guidelines in the regions where such content originates, to reduce the impact of these discourses on public opinion.

Translation: Fernando Llanas