Analysis of The Correlation Between Populist Discourse and Tweet Popularity

This paper analyses linguistic features present in populist discourse on Twitter, together with the relationship that these may have with the popularity obtained by the tweets. The preliminary part of this study was conducted on tweets collected from the oﬃcial accounts of four European populist leaders, namely Luigi Di Maio, Matteo Salvini, Marine Le Pen, and Nigel Farage. This phase suggested that particular discursive elements often related to populism, such as emotionalization, simplistic rhetoric and intensiﬁed evaluations, are present on social media as well. However, the main focus was on the possible correlation between these aspects and the number of “likes” and “retweets” that a single tweet receives. Therefore, tweets were ﬁrstly classiﬁed by a popularity value, and then divided in two groups, creating a corpus of most and least popular tweets for each subject. Secondly, tweets were annotated using the Appraisal Framework (Martin and White 2005), in order to observe the existence of a peculiar linguistic behaviour by populist leaders. The same operation was conducted on a control group formed by three establishment politicians, namely Matteo Renzi, François Hollande, and David Cameron. Finally, the annotations of the most and the least popular tweets were compared to highlight features that were particularly frequent in popular tweets. This process showed how speciﬁc “populist” features, such as emotions, negative judgments, or intensiﬁed evaluations are related with the attention received by users on social media. Findings indicate that these features are positively correlated with the tweet popularity, both when considering populist


Introduction
The surge of populist politicians and movements to the detriment of more established leaders and parties has been demonstrated by several recent events.In particular, two of these started at a national level, but then had a global effect: the decision by the United Kingdom to leave the European Union, also known as "Brexit", and the election of the businessman Donald Trump as the 45th President of the United States.The former was strongly encouraged by the actions of the British Eurosceptic and right-wing party named UKIP and its leader Nigel Farage; the latter surprised most of the media and the public, both rather sure about the victory of Trump's opponent, Hillary Clinton (Healy & Peters 2016;Ingram 2016;Greenslade 2016;Tharoor 2016).In addition, further political outcomes confirmed the momentum of populist parties.In France, Marine Le Pen, leader of the nationalist party Front National, was the second most voted choice in the 2017 presidential election after Emmanuel Macron, who then became President of France.In Italy, the general election held in March 2018 showed the success of two populist parties, Movimento 5 Stelle and Lega Nord.
The spread of social issues, such as immigration, racism, terrorism, and economic crisis in the Western world could have played a major role in the rise of populist ideas during these last decades.However, ideological, social, and political conditions have to exist for populist parties to emerge (Taggart 2002;Laclau 2005;Pasquino 2008).In addition, as stated by several authors (Canovan 1999;Weyland 2001;Kriesi 2014), we could suppose that populism arises in the presence of charismatic leaders who use their personalities to gain consensus during political and social crises.Populist politicians are also able to pervade the media with a peculiar language that emotionally appeals to voters through simplistic rhetoric and spectacular claims (Heinisch 2008).In this context, it is interesting to observe the role that social media are having in spreading the populist message.Although politicians from all parties are increasingly using social media to communicate with the public, populist figures are the ones who seem to take more advantage from having a Twitter or a Facebook account.The use of these allows for populists to directly present their discourse without any kind of filter, with the possibility to enhance their texts with images or videos that may better appeal to the online audience.
The aim of this study is to examine the language used by the leaders of four European populist parties (Movimento 5 Stelle, Lega Nord, UKIP, and Front National) on Twitter, and to observe a possible relationship between some prototypical features that may emerge from their discourse and the popularity of their messages (or "tweets") on the social network.The research tries to offer several important insights regarding not only the features of populist discourses, but also the effectiveness that these have on the dissemination of the message on social media.In order to investigate their linguistic behaviour, we analyse a relatively small corpus of 8,000 tweets collected from the accounts of the above-mentioned parties' leaders: Luigi Di Maio, Matteo Salvini, Nigel Farage, and Marine Le Pen.Next, we use the Appraisal Framework (Martin & White 2005) to verify the existence of emotional, simplistic, and intensified features in populist discourse, whose presence is suggested by previous research (Canovan 1999;Heinisch 2008;Bos 2011).Finally, we compare the most and the least popular tweets for every account in order to observe a possible connection between the presence of specific linguistic features and the number of "likes" and "retweets" received by the message, which illustrate the number of people who liked and/or shared the tweet.Essentially, this paper draws from previous studies on political and populist language (Zappavigna 2011 and2012;Wodak 2015;Ekstrom & Morton 2017;Wirz 2018) and tries to define the role that social media may or may not have on the consensus received by populist parties in these years.
The paper has been structured in four parts.The first part gives an overview of the background studies that are relevant to the current research.The second part is concerned with a description of the methodology used.The third section presents the findings of the research.Finally, the paper offers conclusions, limitations, and directions for future studies.

Literature Review
Populism seems to be a complex phenomenon.This is probably due to its various manifestations all over the world, or maybe due to the fact that the word "populist" is often used to belittle political opponents, regardless of their beliefs.However, there is a large volume of published studies that describes the nature of populism (Gellner & Ionescu 1969;Canovan 1981;Di Tella 1997;Taguieff 2002;Laclau 2005), and an interesting definition of it is given by Albertazzi and McDonnell (2008, 3) who describe populism as: "[. . .] an ideology which pits a virtuous and homogeneous people against a set of elites and dangerous 'others' who are together depicted as depriving (or attempting to deprive) the sovereign people of their rights, values, prosperity, identity and voice." A similar perspective was already adopted by Taggart (2000) who underlined the importance of considering populism as "people-based" rather than "class-based", as other authors had suggested (Di Tella 1965;Conway 1978).Apart from its definition, it is also crucial to understand the circumstances that may facilitate the growth of populism.
According to Pasquino (2008), the conditions that likely determine its rise can be of ideological, social, or political nature.Ideological conditions could be summarised with a rejection of "established" politics, politicians, and parties.In this context, populist leaders claim to understand the disappointment of the public towards traditional politics and promise to be the true representatives of the people's will once in charge.Social conditions refer to the possibility for the populist party to find individuals who feel isolated and alienated from politics and the community due to social issues such as unemployment, migration, or identity crises.Therefore, these people find reassurance in anyone who asserts to be different from the 'others' and who affirms to have solutions for their problems.Finally, political conditions include a profound crisis of the structures of political intermediation, the personalisation of political power, and the pervasiveness of media in political life.
In particular, this latter aspect seems to have played a crucial role in the recent surge of populist consensus, as suggested by the considerable number of studies related to it (Mazzoleni 2008;Bos et al. 2011;Moffit & Tormey 2013;Ellis & Riejos, 2018).However, most of these only focus on populist communication through traditional media, such as TV and newspapers, whereas it is interesting to investigate the function that social media are having in this context.This is evident in the case of Twitter which, according to Trump, played a decisive part during his presidential election campaign (cf. McCormick 2016): it is also worth noting that the President of the United States has written a total of 39,746 tweets, with an average of 12 tweets a day, and has 73 million followers1 .Similarly, the social media users who supported leaving the EU during the Brexit campaign in 2016 were more numerous and generally more active than those who were in favour of remaining (cf.Polonski 2016;Hänska & Bauchowitz 2017).It is not by chance that populist leaders such as Marine Le Pen, Matteo Salvini, and Nigel Farage publicly defined social networks as an important political and social resource, as shown by the following tweets: (1) "Les réseaux sociaux permettent de s'adresser directement au peuple.Ma campagne sera innovante en ce domaine."(@MLP_Officiel, January 4, 2017)2 (2) "Je continuerai de monter en puissance dans ma relation directe avec les Français, sans intermédiaire, par les réseaux sociaux."(@MLP_Officiel, May 1, 2016) 3(3) "Finché almeno su Twitter potremo essere LIBERI, ritwittiamo!Di cosa avranno paura nel clan Renzi?Confrontarsi è bello ma... #boschiscappa" (@matteosalvinimi, October 7, 2016)4 (4) "Without the internet, the development and growth of UKIP in Britain would have been far tougher."(@Nigel_Farage, April 6, 2016) On the other hand, it is worth noting what Matteo Renzi, former Prime Minister of Italy and member of PD (Democratic Party), stated about Twitter, suggesting that reality is more complex than social networks: (5) "@KiaraFarigu i Social sono molto utili.Ma chi li conosce sa che la realtà è più complessa di una tempesta di tweet."(@matteorenzi, May 12, 2015) 5Against this backdrop, the body of literature investigating populism has been constantly growing.On the other hand, few studies have analysed the relationship between populism and social media (Engesser et al. 2017a).Although all parties benefit from the use of the internet when communicating with the electorate, populist groups are the ones who particularly take advantage of it (Bartlett 2014).Since they present themselves as representatives of the people (cf.Canovan 1999), they seek a direct connection with the public, aiming to bypass the elites and the journalistic gatekeepers, who are often depicted as untrustworthy (cf.Mazzoleni 2008).This function is granted by a direct, democratic, and rather uncontrolled channel of communication: the social media (cf.Engesser et al. 2017b;Esser et al. 2017).In addition, a further opponent of the people is represented by the "others" who, differently from the above-residing elites, are portrayed besides the people (cf.Jagers & Walgrave 2007;Albertazzi & McDonnell 2008).They typically consist of ethnic, religious, or sexual minorities whose existence both endangers the honest, ordinary, and hard-working people, and promotes in-group favouritism or out-group discrimination (cf.Reinemann et al. 2017).This homophilic behaviour is facilitated by the internet and social media, where users tend, on one hand, to select information and media content that only reinforce their own opinions, creating a so-called "filter-bubble" (cf.Pariser 2011), and on the other hand to be part of social groups who share the same views, amplifying political and social attitudes through an "echo-chamber" effect (cf.Jamieson & Cappella 2008).Another aspect that exemplifies how convenient social media are to populist politicians is the possibility to create a personal account, which guarantees the capacity to provide a personalised communication that may involve users more than websites focused on the whole party (cf.Kruikemeier et al. 2013).Finally, several studies have defined the style of communication that characterises populism (Canovan 1999;Bos et al. 2011;Kramer 2014).As suggested by Engesser et al. (2017a), there are three major aspects that characterise populist discourse: simplification, emotionalization, and negativity.These three factors strongly benefit populist leaders since they allow to catch people's attention on the Internet, a place where users are often overwhelmed by the amount of information and content (cf.Shoemaker & Cohen 2006).
Therefore, it is well established that there is a strong connection between social media and populists, and that the possibilities offered by the former are thoroughly exploited by the latter.However, the influence that the particular discursive features attributed to populism have on the spread of the populist message has not been fully investigated.
Thus, the underlying question that this study tries to answer is whether the popularity of a tweet, intended as the sum of "likes" and "retweets" (the number of people who like and/or quote the message), is correlated to the presence of peculiar discursive aspects, such as emotional language, intensification, and simplified rhetoric.

Methodology
Although there are several social media websites on the Internet, this paper only focuses on the language present in messages written by populist leaders on Twitter.From a methodological point of view, Twitter allows to collect messages (or "tweets") rather easily, keeping at the same time a considerable amount of metadata such as creation date of the tweet, user name of the author, URLs present in the message together with a series of peculiar elements such as hashtags, mentions, and retweets.In order to analyse linguistic features in a wider spectrum of populist discourse, tweets were collected from the official accounts of Luigi Di Maio, Matteo Salvini, Marine Le Pen, and Nigel Farage.
While the first three politicians are the institutional leaders of their parties (Movimento 5 Stelle, Lega Nord, and Front National respectively), Nigel Farage officially resigned from his role as leader of UKIP on 4 July 2016.However, he still has a significant authority both towards UKIP and "Brexiters" (cf. McCrum 2017;Lowles 2018;Cohen 2018).In addition, we noticed that his tweets were still far more popular than those of any other member of UKIP.
The four populist parties and their leaders were chosen for different reasons.Firstly, they meet the definition given by Albertazzi & McDonnell (2008), as they praise the sovereignty of the people to the detriment of institutional politics.These groups also seem to be rather sceptical towards traditional mass media and are represented by leaders that shape how the parties appear and communicate with the public.Moreover, these parties seem to share a rather right-winged version of populism, since they criticise immigration and excessive taxation and prefer nationalism over socialism (cf.Otjes & Louwerse 2015).As a result, they differentiate themselves from left-wing populist parties in Europe, such as Podemos in Spain, the Linke in Germany, Syrisa in Greece, and the Socialist Party (SP) in the Netherlands.
Tweets were collected using FireAnt (Anthony & Hardaker 2016), a software that uses Twitter API to gather messages from one or more accounts.Retweets were excluded as messages written by external authors might have invalidated the research.In addition, tweets included in the analysis do not represent the total number of messages written by an author on Twitter, as the API only allows for a limited number of tweets to be collected.However, the extent regarding the data should be sufficient to cover an acceptable time interval, as suggested by Next, in order to identify peculiar linguistic features, three reference corpora (one for each language) were created by gathering tweets from the official accounts of Matteo Renzi, François Hollande, and David Cameron.These three political leaders were chosen both because of their belonging to established parties and their former positions as Prime Ministers.As with populist messages, reference tweets were collected using FireAnt, filtering out retweets.(Canovan 1999;Heinisch 2008;Bos 2011).The framework is based on Systemic Functional Theory (Halliday et al. 2004) and is designed to analyse levels of evaluation in a discourse.More precisely, it focuses on "exploring, describing and explaining the way language is used to evaluate, to adopt stances, to construct textual personas and to manage interpersonal positionings and relationships" (White 2001, 1).
Moreover, UAM CorpusTool (O'Donnell 2011) was used to annotate tweets according to the three main features of the framework: attitude, which concerns emotional language, ethical judgements, and aesthetic evaluations; engagement, which describes how an author includes, excludes, or ignores external stances in his/her discourse; and graduation, which focuses on how degrees of evaluation are increased or decreased.Before the annotation, tweets of each author were classified by a popularity value, which represents the sum of both likes and retweets of a message, and only the first and the last thousand tweets were considered.These quantities are arbitrary, the reason for their choice being that the Twitter API limits the maximum number of tweets downloaded, amounting to approximately 3,000 tweets per account.In this way, each subject had 2,000 tweets, resulting in a total of 8,000 tweets for the populist group and 6,000 tweets for the reference group.Finally, UAM CorpusTool was also used to compare the annotations between popular and non-popular tweets: the results were classified by their propensity value (also known as relative frequency) in order to understand how great the difference is between two corpora for a particular keyword or feature (cf.Hardie 2014).However, statistical significance was guaranteed by deleting all features with a Chi square value below 3.84, which is the threshold value for significant data with df =1 and p<0.05 (Gries 2013).

Results
In this section, findings are first presented with a pie chart, which graphically compares the most frequent features of the appraisal framework when analysing popular tweets in the populist corpus and the control group.Secondly, detailed results for each author are shown via tables; this would allow for understanding possible inconsistencies in the tweets by one or more authors.As mentioned in the second chapter, the appraisal framework is formed by three main systems: attitude, engagement, and graduation.
Considering that each of these items has several sub-nodes, we decided to consider only the most discrete features, incorporating some of the smallest sub-nodes (especially endnodes) to their parent items.For example, the "monoglossic" feature regards statements presented without any possible dialogical alternatives (i.e."The government has failed.")and consists of two sub-nodes: "presupposition" and "argumentative".With the former, propositions are taken for granted, whereas with the latter, propositions are minimally justified.Since the difference between the two is not the focus of this study, and the frequency of "argumentative" propositions is often low, we only present results for the general "monoglossic" feature.
The following pie chart presents an overall analysis of the most frequent features of the framework in popular tweets when compared to non-popular tweets by the same author.Here, politicians were grouped in order to give a general idea of the differences between the two corpora.The pie charts suggest that, in both corpora, more than 50% of the features that are more frequent in popular tweets when compared to non-popular messages belong to the attitude system.Figure 1 also shows a general similarity between the populist corpus and the reference groups, with a slightly major presence of attitude features in the establishment politicians' texts.Similarly, graduation elements are slightly more frequent in populist subjects, however only by 4%.Finally, engagement percentages are nearly identical.
Although the absence of considerable discrepancies between the corpora may be counterintuitive, the fact that attitude features are the most peculiar trait when differentiating popular and non-popular tweets was already suggested by previous research (Zappavigna 2011;Stieglitz & Dang-Xuan 2013).In addition, we already observed overall similarities when investigating the differences between tweets by populist groups and establishment parties (Carrella 2017).However, findings in this study seem to confirm the relationship between emotionally connoted tweets and their popularity on Twitter.
Next, detailed results are presented in author-specific tables.All tables in this section present the most frequent features in the first thousand popular tweets for each author when compared to the last thousand popular tweets.Results are classified by propensity (or relative frequency), showing Chi square values as well.The relative frequency helps us to understand the degree of difference between two corpora when observing a word or a feature (Hardie 2014).A propensity value of 1 indicates no difference in the relative frequency of the same element in the two corpora, whereas a value of 2 shows that the feature is twice as frequent in the popular tweets compared to the non-popular tweets.
On the contrary, a propensity value of 0.5 means that the item is half as common in the popular tweets.Finally, colours are used to facilitate the interpretation of our results: all features related to the attitude system are in grey, elements belonging to the engagement system are in dark grey, while all graduation sub-nodes are in light grey.As can be seen from Table 3, the first nine positions, with the exception of "textual", are occupied by attitude features.Among these, "non-authorial-evaluation" represents the most distinctive element between popular and non-popular tweets written by Nigel  3, 6 and 7).This may be explained by the fact that members of Movimento 5 Stelle tend to use social networks more than other parties and probably manage to maximise the spread of their messages through the use of particular conventions such as hashtags or mentions.However, it is interesting to notice that two attitude features, "negative-attitude" and "judgement", have a propensity higher than 1.The first element indicates all occurrences of negative emotional language, while the second represents social and ethical judgements of human behaviour.The following tweet contains an example of an implicitly negative judgement:

Fabio Carrella
Turning now to the results obtained with the reference corpora, we could notice a surprising similarity between the key features in populist tweets and the texts by David Cameron, François Hollande, and Matteo Renzi.As with Salvini, the "distance" feature is distinctive in Cameron's popular tweets as well.

Features
Next, we find "negative-attitude", which connotes all occurrences of negative emotional language, while the third position is occupied by "repetition", which represents all instances of intensification achieved by the repetition of the same lexical items or by lists of closely related terms, as in the following tweet: (10) "The Labour party is now a threat to our national security, our economic security and your family's security."(@David_Cameron, September 13, 2015) Finally, the first two positions in Renzi's popular tweets are held by "distance" and "monoglossic", two engagement features that respectively contract and ignore different dialogical alternatives.Next, we find "dis/satisfaction", which exemplifies all instances of positive or negative emotions related to personal satisfaction.The following tweet is an example of a "monoglossic" statement containing a positive non-authorial "dis/satisfaction" occurrence (in bold).

Conclusion
This study was designed to examine the presence and possibly determine the effect of specific discursive features that are said to characterise populist discourse (Canovan 1999;Heinisch 2008;Bos 2011).More specifically, the aim was to investigate particular elements such as emotional language, simplistic rhetoric, and intensified discourse, and to observe whether these features are related with the popularity and with the diffusion of a message on social media.The Appraisal Framework, a tool designed by Martin & White (2005)

Summary of findings
The first major finding was that the linguistic elements that the literature attributes to populist discourse are also present in the social media context.Additionally, these seem to be strictly related to the popularity of a tweet, that is how much a tweet is liked or shared by other users on Twitter.Instances of attitude features, represented by emotions, judgements, or aesthetic evaluations, represent the most distinctive trait when comparing popular and non-popular tweets.This result confirms previous findings which suggested a positive relationship between the presence of emotional language and the popularity of a message on social media (Zappavigna 2011;Stieglitz & Dang-Xuan 2014).Examples of simplistic rhetoric, particularly illustrated by a framework item called "monoglossic", are usually more frequent in popular tweets, with the exception of Movimento 5 Stelle's leader, Luigi Di Maio.Other instances of detrimental language used to discredit political opponents, such as the "distance" feature, are only present in findings regarding Matteo Salvini.Finally, elements of graduation that suggest a tendency to intensify or decrease degrees of evaluation, such as "quantification", "graphical", or "soften", are found in all populist subjects, again with the exception of Luigi Di Maio.
The second major and more surprising result was found in the analysis of the reference corpora, consisting of tweets collected from Matteo Renzi, François Hollande, and David Cameron.From a quantitative point of view, the analysis of the most prominent features in the popular tweets showed strong similarities between the populist and the reference corpora, even indicating a higher number of attitude features in the establishment politicians' messages.From a qualitative perspective, a detailed observation of the more frequent features in the popular tweets also presented a considerable affinity between the two corpora.Items related to emotional language are substantially more frequent in popular tweets, with the "negative-attitude" feature having a higher propensity values in all three reference subjects.Popular tweets by Renzi and Hollande have a high presence of "monoglossic" and "distance" features, both ideally linked to populist discourse.Although the "monoglossic" item is absent in Cameron's findings, we find "distance" and "disclaim" having a high propensity value, this latter representing the will by an author to reject contrary dialogical positions.Lastly, graduation features are distinctive for all reference subjects, with sub-nodes such as "graphical", "quantification", and "repetition".
Several conclusions can be drawn from the present study.On one hand, findings regarding populist tweets complement those of earlier studies: populist politicians chosen for this research confirmed the presence of emotional language, simplistic or dichotomist rhetoric, and outrageous or spectacular claims in their tweets (cf.Heinisch 2008).Furthermore, these elements seem to be more frequent in popular tweets when compared to non-popular tweets, especially when considering instances of emotional language.
On the other hand, findings regarding members of establishment parties raised relevant questions.It seems that linguistic features often related to the populist style are not only present in non-populist discourse, but they are also considerably frequent in tweets with a high popularity.The fact that "populist" elements are present in non-populist politicians' tweets may suggest that the boundaries defining the populist style are more blurred than it is believed, at least on social media.The context chosen for this study probably plays a key role in the implications: it could be hypothesised that particular discursive features, usually linked to populism, easily attract users' interest on social media (Shoemaker & Cohen 2006).Paradoxically, non-populist politicians seem to take more advantage from these dynamics, maybe because their use of populist linguistic features, such as emotionalization, simplification, and intensification is seen as exceptional and, consequently, worth considering.
Overall, the evidence from this study confirms the existence of particular discursive characteristics present in the populist discourse and their relationship with the popularity of the texts on social media.However, it also supports the idea that, at least on the Internet, establishment parties are adapting their style, gaining online popularity and becoming increasingly similar to those populist politicians from whom they usually distance themselves.

Limitations and directions for future research
The major limitation of this study is probably represented by the small sample of subjects taken into consideration, both for the corpus of interest and the control group.
Therefore, the generalisability of these results is relatively restricted.Furthermore, the fact that we chose to consider only 2,000 tweets per author probably excluded further insights regarding the linguistic features used.Finally, there were several ambiguous cases that posed some problems when annotating tweets with the Appraisal Framework, especially considering that the annotation was conducted by a single author.
Notwithstanding these limitations, this paper offers some observations regarding the linguistic behaviour of populist and establishment politicians on social media.Future research may choose to include more subjects in order to broaden or reject findings offered by this paper.In addition, authors may focus on one particular aspect of the Appraisal Framework in order to offer more detailed results.Additionally, other social media may be considered: for example, Facebook offers more space to write and therefore more context in comparison to Twitter.Finally, statistical methods could be used to find significant relationships between the number of likes or retweets obtained by a tweet and the linguistic features present in the text.

Table 1 :
Data for Accounts and Tweets of Populist Politicians

Table 2 :
Data for Accounts and Tweets of Reference Politicians The Appraisal Framework created by Martin & White (2005) was adopted to determine to what extent the linguistic features implied by the literature are present in populist online discourse

Table 4 :
Key Features in Le Pen's Popular TweetsA similar trend is shown in Table4, which regards tweets by Marine Le Pen.Although there are only nine features with a propensity value higher than 1, six of these consist of attitude items.However, the first position is held by "graphical", a sub-node of graduation that indicate intensification of the message through the use of graphical elements such as emoticons or capital letters.The second element is "monoglossic", which is a feature related to engagement: it refers to non-dialogical statements of an author, as in the following tweet:(7) "Dimanche rappelez-vous: le people est la seule chance pour la France!MLP"

Table 5 :
Key Features in Di Maio's Popular Tweets

Table 5 ,
concerned with Luigi Di Maio's tweets, shows slightly different results regarding key features.Findings indicate that hashtags and mentions are twice as common in popular messages than they are in non-popular tweets, whereas they are usually more frequent in non-popular tweets (cf.Table

Table 6 :
Key Features in Salvini's Popular Tweets

Table 7 :
Key Features in Cameron's Popular Tweets

Table 8 :
Key Features in Hollande's Popular TweetsWe can find further similarities comparing results from Marine Le Pen and François Hollande.In fact, the first two positions are held by the same features, "graphical" and

Table 9 :
Key Features in Renzi's Popular Tweets to study evaluation in discourse, was used to analyse the most and the least popular messages of four European populist leaders on Twitter, namely Luigi Di Maio, Matteo Salvini, Marine Le Pen, and Nigel Farage.The same operation was conducted on a reference group constituted by three European democratic leaders, specifically Matteo Renzi, François Hollande, and David Cameron.