1
Tweeting Populist Sentiment
A study of Forum voor Democratie’s use of emotional language on Twitter
Abstract: Social media have been praised for bringing political actors in direct touch with the electorate. This direct linkage provides opportunities for actors new to the political landscape, that can now bypass traditional media. The characteristics of social media such as Twitter line up with the ideology of political populists. However, not much research focuses on the way populist politicians use social media and its effects. Generally, Tweets with positive emotions are more likely to be spread. However, populist parties are expected to be more negative online, especially toward elites and the establishment. This negativity may lead to further balkanization of the Twittersphere, thereby influencing an increasingly important political news source. In this research, sentiment analysis is used to answer the question how and to what result the Dutch populist party Forum voor Democratie uses emotional language on Twitter. The findings show that sentiment is present in a majority of Tweets, and that the presence of positive sentiment in Tweets and a lower popularity of said Tweets are correlated. These findings should provide the necessary spark for further, comparative analysis to establish whether Forum voor Democratie and their use of sentiment in Tweets are anomalies in the political Twittersphere or not.
Written by Dunja Lacroix 1044581
MSc Student Political Theory Graduate School of Social Sciences University of Amsterdam Supervised by dhr. dr. G. Schumacher
Assistant Professor in Political Science Social and Behavioral Sciences University of Amsterdam
2
Content
1. Introduction ... 3
2. Theory ... 6
2.1SOCIAL MEDIA AND POPULISM ... 6
2.1.1 Social Media and Populism in the Netherlands ... 10
2.2EMOTIONS AND IDEOLOGY ON SOCIAL MEDIA ... 12
2.3FRAGMENTATION AND ECHO CHAMBERS ON SOCIAL MEDIA ... 14
3. Research Questions And Hypotheses ... 17
4. Research Design ... 18
4.1CASE ... 18
4.1.1 Forum voor Democratie... 18
4.1.2 Twitter ... 19
4.2DATASET ... 21
4.3METHOD &ANALYSIS ... 22
4.3.1 Sentiment analysis check ... 23
5. Results ... 26 5.1SENTIMENT ANALYSIS ... 26 5.2CORRELATION TEST ... 27 6. Discussion... 28 7. Conclusion ... 30 8. Bibliography ... 32 9. Appendix ... 36
9.1SENTIMENT ANALYSIS CHECK ... 36
9.1.1 Tweet 1 ... 36
9.1.2 Tweet 2 ... 37
3
1. Introduction
The Dutch political party Forum voor Democratie surprised many with their electoral success. A mere year after its introduction, the party was elected to two seats in the Dutch parliament and counted over seventeen thousand members. And according to polls, if elections were held now, they could expect up to sixteen seats, making it the third largest party in the Netherlands (de Hond, 2018). On a party congress in November 2017 Henk Otten, co-founder of the party, explained the key to the party’s success: social media. “Forum voor Democratie has emerged to be the biggest online movement of the Netherlands. That is the reason we managed to get elected.” In this paper I will research whether and how the use of emotional language contributed to the success of the online Twitter presence of the Dutch populist party Forum voor Democratie (FVD).
Over the last decade, social media have become an integral part of millions of people’s daily lives. More and more people rely on social media to get their news and trust it more than news found through traditional news sources (Krasodomski-Jones, 2016). Trust in the mainstream news has fallen, and mainstream politicians have seen their majorities deteriorate (Krasodomski-Jones, 2016). In response, old and new political parties have taken to the internet for their campaigns. Because politics are a polarizing topic, political communication often features emotional language. Recent studies positively link the role of emotional language in social media content to information diffusion (Stieglitz & Dang-Xuan, 2013). The use of emotional language triggers cognitive and arousal effects, resulting in higher levels of attention, participation, and social sharing behavior (Stieglitz & Dang-Xuan, 2013). Exposure to emotional language may in this way lead to both emotional and social contagion, spreading the mood and moral judgement in emotional content (Brady, Wills, Jost, Tucker, & Van Bavel, 2017). In sum: social media provide political parties with a platform that gives their content the potential for virality as well as social contagion, which can mobilize a constituency. These are the ingredients for a successful online campaign. There is, however, a reverse side to this coin.
While social media were lauded for their democratic potential at their emergence, recent research has indicated that social media may also undermine the democratic process through the formation of “echo chambers”, in which individuals are only exposed to conforming opinions (Dubois & Blank, 2018). Such politically fragmented echo chambers may stimulate political segregation, and are linked to both populism and extremism (Groshek & Koc-Michalska, 2017; O’Hara & Stevens, 2015). For populist politicians, the effect may provide an opportunity to find
4
and target an audience where their ideas are salient. Therefore, it might be of particular interest to newer populist parties that wish to build up a constituency, especially when targeting groups that are overrepresented online, such as males and youths (Barthel, Gottfried, Mitchell, & Shearer, 2016). Because using emotional language influences group-level communication behavior in terms of information sharing as well as moral judgements, in particular among the in-group, it may influence the formation of such echo chambers (Stieglitz & Dang-Xuan, 2013).
In this paper I will answer the question: “To what extent and effect do Forum voor Democratie members use emotional language on Twitter?” I will argue that their great online success can be partially explained by their use of emotional language, which in turn may facilitate an echo chamber effect. I will show that Tweets by FVD that express negative emotions gain more traction, gaining greater popularity with the in-group of FVD-supporters. This finding opposes emotional sharing theories that presuppose positive messages are more likely to be spread. While no exhaustive conclusions can be based on this research, it should encourage further delving into this highly relevant and overseen topic.
This paper will contribute to contemporary understanding of the political implications of social media in twofold: societally and scientifically. Scientifically, the research aims to add to the growing literature on contemporary campaigning strategies of populist political parties. Understanding how FVD has achieved their success provides insight in similar success stories worldwide. The roles the echo chamber effect and emotional content play in this success have so far been underexposed.
Researching echo chambers has been controversial: as social media are a new field, there is no consensus on how to best research echo chambers. And while some scholars have found evidence for echo chambers and political fragmentation on social media, others claim the mechanism is overstated, or that social media largely discourage this fragmentation (Dubois & Blank, 2018; Karlsen, Steen-Johnsen, Wollebæk, & Enjolras, 2017; Krasodomski-Jones, 2016). All studies have in common that they are limited to American or British politics. Adding a European perspective will shed new light on the generalizability of these findings. By focusing on the dynamics between a political party and social media, this paper also aims to contribute to our understanding of contemporary political campaigning and political communication. And while some research on the online presence of political parties has been done in recent years, there has
5
barely been attention to strategies specific to populist parties, or to what extent and effect parties embed emotional language in their content.
From a societal perspective, this research contributes to an evaluation of the potential adverse effects of online political communication on democratic values. Scholars have pointed out that echo chambers may contribute to a growing knowledge gap between those that are politically interested and those who are not. They may also encourage partisan segregation and polarization (Bright, 2018; Garimella, Morales, Gionis, & Mathioudakis, 2018). Furthermore, the effect has been linked to both extremism and populism (Groshek & Koc-Michalska, 2017; O’Hara & Stevens, 2015). This paper mainly also adds to the young but growing literature on this link to populism. Mapping online fragmentation around a political party is the first step in assessing whether and to what extent we can interpret echo chambers as a threat to our democratic values.
The rest of this paper will be built up as follows: first, I will outline the theoretical framework. I will introduce the concepts of fragmentation and echo chamber, how emotional expressions online may contribute to them, and what part politicians play in this polarization. This theory is necessary to understand fragmentation and how to study it. I will deduce three hypotheses that will be tested in this paper. Second, I will introduce the case in depth and describe the data and research method used. Third, I will display and discuss the results stemming from the research. Finally, I will make recommendations for future research, and discuss drawbacks that were ran into during this research.
6
2. Theory
In this section, I will provide an overview of the current state of the scientific debate on echo chambers. The buildup is as follows: First, I will discuss contemporary political campaigning on social media. How do politicians use social media to campaign? And how do populist parties like FVD use social media in comparison? Second, I will zoom in on the effect of using emotional language in social media content. How does sentiment affect our political views and amplify fragmentation? Finally, I will further discuss the negative effects emotional political content on social media can have in the “echo chamber effect”. From these theories, I will deduce three hypotheses which I will verify through data analysis, as described in the next chapter.
2.1 Social media and populism
More and more politicians use social media platforms to spread their message. It has become a more important communication channel to politicians than radio and television (de Boer, Puts, Livestro, & van Brussel, 2014). The main reason for this is captured in normalization and equalization theories. Equalization theories state that social media provide political actors with a way to put themselves on the political landscape without needing traditional resources (Jacobs & Spierings, 2016). No longer do political actors have to rely on having the expertise, finances, or existing constituency to participate in the political debate. Social media provide political actors with a channel to promote themselves and directly communicate with their electorate through. The equalization theory thus refers to the way social media restructure existing power relations in the favor of newcomers. The normalization theory refers to politicians joining social media as it becomes the norm (Jacobs & Spierings, 2016). In a country like the Netherlands, these theories have somewhat lost relevance as it is a social media frontrunner, with nearly all Dutch MPs being active on Twitter (Jacobs & Spierings, 2016). Twitter has become a legitimate and frequently used politician communication channel for both political institutions and citizens (Stieglitz & Dang-Xuan, 2013).
Less research, however, has considered what the online presence of politicians looks like, or how this may affect (potential) voters. In particular, how populists use online media is mostly still a mystery. Where studies on online communication of populist politicians and political actors on social media are both frequent, combinations are rare. In this section I will review the available
7
literature on political populists on social media. First, I will define the concept of populism. Then, I will discuss how populist logic may influence the social media use of populist politicians.
Populism has been a highly contested concept within the field of political science. Nowadays, most scholars accept Cas Mudde’s definition of populism as “a thin-centered ideology revolving around the antagonistic relationship between two homogenous constructs: the (good) people and the (evil) elite” (Mudde, 2004). Populist parties, especially those on the radical right, are usually characterized by a concentration of power in the leadership to a greater extent than other parties (Jacobs & Spierings, 2018). Consequently, populism is often connected to charismatic leadership (Taggart, 2004).
The relationship between populism and online media was first brought to attention in the late 1990s, where the internet was thought to “restructure political power in a populist direction” (Ernst, Engesser, Büchel, Blassnig, & Esser, 2017). Social media arguably fit this mold even better, and particularly Twitter’s key attributes complement populist ideological characteristics according to Jacobs and Spierings (Jacobs & Spierings, 2018). They list four main attributes: unmediated communication, accessibility, engagement and virality.
First, Twitter allows for unmediated communication (Jacobs & Spierings, 2018). Twitter allows politicians to directly tap into “the voice of the people”. This is important, because populists need to be in touch with the grievances of the people they claim to represent. It also allows populist parties to bypass hostile traditional media elites and distinguishes the parties from the establishment. Twitter’s microblogging format centers around short messages, making it excellent for one-liners, which in turn contrasts with elitist and complex language usually used on the topic of politics in traditional media and in parliament itself (Engesser, Ernst, Esser, & Büchel, 2017). All in all, Twitter’s unmediated communication may emphasize that politicians are “simply one of us”. However, this unmediated aspect may clash with populist parties’ focus on centralization and the party leader. Jacobs & Spierings theorize that because of this some populist parties may restrict party members’ social media use (Jacobs & Spierings, 2018). Another way to successfully combine the unmediated aspect of Twitter with the populist’s value of centralization and the party leader is displayed by Dutch populist party Partij voor de Vrijheid (PVV). Unlike all other Dutch political parties, they do not have an official PVV party Twitter account. Instead, party leader Geert Wilders’ account is considered the party account. The frequency of Wilders’ Tweets is
8
almost twice as high as the number two on PVV’s list Fleur Agema, showing a significant difference in approach.
Second, Twitter is cheap and requires no expertise to use (Jacobs & Spierings, 2018). Where traditional media are characterized by professional gatekeepers such as journalists and a passive audience, “network media logic” revolves around interest-bound and likeminded peer networks in line with equalization theories (Ernst et al., 2017). In practice, this means that even those lacking financial and technical resources or support from gatekeepers can build a constituency (Jacobs & Spierings, 2018).
Third, Twitter is built on engagement, enabling direct interaction and dialogue between politicians and citizens (Jacobs & Spierings, 2018). This human-contact opportunity can be further built upon as unlike other media channels, social media are personalized, potentially offering a look behind the scenes into the personal lives of politicians (Jacobs & Spierings, 2016; Krasodomski-Jones, 2016). This may be humanizing and makes politicians seem more approachable and less elitist (Jacobs & Spierings, 2016). This low threshold of connecting and conversing with people on Twitter raises questions on how it meshes with populist ideology. Populist parties claim to represent the people, and are said to stay away from those who reject their views (Stieglitz & Dang-Xuan, 2013). And Mudde righteously points out that “if citizens want politicians to know rather than listen to the people, there is no incentive for populist parties to connect and interact with the people” (Mudde, 2004). Still, it is a common presumption that the Internet stimulates homophily, a “tendency of similar individuals to form ties with each other” (Engesser, Fawzi, & Larsson, 2017). And unlike other media channels, social media like Twitter allow the opportunity to connect with “like-minded others” (Jacobs & Spierings, 2016). While Twitter users with opposing views cannot be prevented from chiming in in political conversations, network media logic dictates that they are not part of the likeminded peer network of populist parties. Such a network in which sentiments are largely similar, may respond harshly to deviating opinions (Engesser, Ernst, et al., 2017). The mechanism behind this echo chamber effect will be discussed in section 2.3.
Fourth and finally, Twitter is characterized by speed and virality, with content reaching many users and spilling over to traditional media in no time (Jacobs & Spierings, 2018). This can be a double-edged sword, as negative content going viral can be damaging to political parties.
9
While virality can rapidly spread ideas to both wider mainstream and broader likeminded audiences, it can also put in-group dissent under a magnifying glass (Jacobs & Spierings, 2018).
In sum, Twitter’s features can assist populist parties in showing they are anti-establishment and identify with and are in touch with their electorate. These are crucial factors for their (good) people-centric and (evil) elite opposing ideology to flourish (Ernst et al., 2017).
In a recent research on how populist parties manifest their ideology on social media, scholars emphasize the important opportunity social media provides populist parties to attack elites (Engesser, Ernst, et al., 2017). Populist parties use very harsh language to frame elites as the enemy of the people they represent. Engesser et al righteously point out that in traditional media, journalists as well as social barriers would prevent populists from articulating themselves in this way. The network media logic of social media on the other hand encourages harsher language, as it may draw more attention and possibly result in virality (Engesser, Ernst, et al., 2017).
Jacobs & Spierings consider how populist political actors and social media further serve each other. They theorize that the online balkanization and echo chamber effects complement the ideas of populist parties (Jacobs & Spierings, 2018). Echo chambers on Twitter are people-centered, allowing real, likeminded people to spread and hear ideas that are ignored by traditional media. Echo chambers also fit the populist focus on the populist party leader, who the echo chamber can be centered around (Jacobs & Spierings, 2018). A strong echo chamber will keep a distance from those who do not share populists’ opinion, and may contribute to messages going viral (Jacobs & Spierings, 2018). This viral potential should not be underrated as a tool in the range of contemporary political communication strategies, allowing parties to reach secondary audiences (Ernst et al., 2017). Viral social media posts often become news in traditional media itself, widely expanding the audience reached. For example, Tweets by politicians like Donald Trump are sometimes presented as independent news items in traditional media. Consequently, when taking populist ideology and group sharing behavior in consideration, scholars consider populist parties to have incentive to use harsh language differentiating in-group from out-group.
10
2.1.1 Social media and populism in the Netherlands
So far, research on populist parties in the Netherlands has centered on the PVV (Partij voor de Vrijheid) and SP (Socialist Party). Jacobs & Spierings researched the Twitter use of these parties. Looking at the adoption and use of Twitter by the PVV and SP Jacobs & Spierings conclude that populists use Twitter later and less than other Dutch politicians. Specifically, party leaders are first to adopt Twitter, and MPs follow slowly (Jacobs & Spierings, 2018). They explain this through the party structure of the parties: populist parties are highly centralized and often feature a strong charismatic leader. Backbencher politicians of populist parties risk taking attention away from the party leader through their Tweets, or even exposing within-party conflict (Jacobs & Spierings, 2018). While party leaders use Twitter, the parties are wary of public dissent. In interviews, SP’s campaign leader suggested that the party does not want individual MPs to start airing their individual opinions and criticizing each other on social media (Jacobs & Spierings, 2018). Former MPs of the PVV suggest that “Wilders fears a loss of control stemming from conflicts”, which is in line with populists’ tendency of a centralized party structure (Jacobs & Spierings, 2018).
FVD’s online presence appears to be a different creature altogether. While this political party partially resembles the PVV and SP in its populist critique of the established elite, it seems to differ significantly in its use of the internet in its campaigns. From its beginning, it has had a very pro-active and strong online presence. With a website that provides members the opportunity to start and support grassroot initiatives, FVD’s presence resembles post-materialist parties like GroenLinks more than its fellow populists (Forum voor Democratie, 2017; Jacobs & Spierings, 2018). When asked about the party’s online campaigning strategy, party leader Thierry Baudet replied “we’re doing it just like Trump” (FTM, 2017). According to him, FVD performs daily benchmarks on hundreds of thousands of social media users (FTM, 2017). In 2017, Baudet also claimed FVD used the services of data company Cambridge Analytica, a company pivotal in Trump’s electoral success (FTM, 2017). This company specializes in “election management strategies” using personality profiles of social media users. Baudet did not specify what services of the company FVD used, and later, after Cambridge Analytica was hit with public controversy, his statement was withdrawn (FTM, 2017). Whether or not Cambridge Analytica was involved, it is clear that FVD’s social media presence is a focal point for the party, regularly Tweeting data
11
showing them as the fastest growing, most active or most engaged political party online (FTM, 2017).
Jacobs & Spierings found that the communication style of populist parties such as the PVV and SP involves every day, simple language, which translates well to the character limits of a Tweet (Jacobs & Spierings, 2018). Baudet, on the other hand, has been known to favor more complexity in his communication style, at one occasion even addressing parliament in Latin (Telegraaf, 2017). In this paper, I will not look at the complexity of words, but at meaning that is universal: that of emotions. The next section will center around how looking at emotional language can help us understand how networks around populist parties are shaped. A further analysis of FVD will follow in section 4.1.1.
12
2.2 Emotions and ideology on social media
Recent research has drawn attention to the importance of emotions in our processing of political and social information. In this section, I will review the relation between emotions and social media.
The relation between emotions and social media has been studied extensively. Generally speaking, the consensus is that emotionally charged tweets tend to be retweeted more often and more quickly compared to neutral ones (Stieglitz & Dang-Xuan, 2013). Consequently, tweets with emotional content go viral more often than neutral ones. Emotional stimuli such as emotional words may cause cognitive processes such as attention, as well as raised levels of physiological arousal (Stieglitz & Dang-Xuan, 2013). This increased cognitive involvement is translated into a higher likelihood of information sharing. Moreover, it also stimulates emotional contagion, which is “the spread of mood and affect through populations by simple exposure” (Stieglitz & Dang-Xuan, 2013). Emotional contagion in turn may impact individual and group-level behavior in terms of sharing content.
Not all emotions impact sharing behavior the same. Emotional expressions that evokes high-arousal responses can be positive, such as joy and hopefulness, or negative, like anger or fear. They are linked to virality. On the other hand, low-arousal emotions like sadness result in deactivation and are less viral (Brady et al., 2017). Between positive and negative emotions, using positive emotions reinforces a sense of community and encourages continued participation within the community, while negative expressions result in quicker and more emotional and hostile interactions (Stieglitz & Dang-Xuan, 2013). Studies of social media have shown that posts expressing negative sentiment produce more reactions than those with positive sentiment. The negativity in the original message is echoed in the responses. The negativity in negative expressions is generally considered more negative than the positivity of positive content. This is called negativity bias (Stieglitz & Dang-Xuan, 2013). In short, both positive and negative emotional expressions are linked with activation and virality. Positive emotional content is more likely to be shared and strengthens a sense of community, and negative emotional content breeds more emotional and hostile responses.
Importantly, the messages we encounter online may not just influence our emotions, but also our views. Emotional appeals, research shows, are effective persuasive tools. They influence “what we notice, what we learn, what we remember, and ultimately the kind of judgments and
13
decisions we make” (Stieglitz & Dang-Xuan, 2013). Moreover, emotions may influence our morality. Our morality guides our evaluation of societal norms, and therefore our ideology. Morality is mostly shaped by the social world, and the contemporary social world increasingly cannot be seen apart from the online world (Brady et al., 2017).
In their 2017 publication, Brady et al research how social contagion over social media works. Social contagion here is the phenomenon of developing similar ideas and intuitions as those we are socially connected to (Brady et al., 2017). In this way, morality is socially transmitted. According to Brady et al, emotions amplify and may even moralize actions and ideas. Therefore, emotions play a key role in online social contagion (Brady et al., 2017). This contagion may be amplified through fragmentation and echo chamber effects such as described in section 2.3 of this paper. If this is the case, negative emotional expressions may cause fragmentation, by a contagious disapproval of the out-group. Brady et al find this is the case in their research on moral-emotional language use in social media. They conclude that moral-emotional language and thus social contagion spreads more widely within the in-group compared to the out-group, and that this effect is significantly stronger for conservatives than for liberals (Brady et al., 2017). Non-moral emotional language, emotional language that is not tied specifically to morality and politics, shows a similar, but slightly weaker effect.
14
2.3 Fragmentation and echo chambers on social media
Looking only at the uses of social media for political actors leaves out an important aspect: the negative consequences that may arise from the use of social media as a news source. The development and widespread usage of social media in the last decade has reinvigorated a debate sparked by the initial rise of the Internet decades ago. At its emergence, the internet was lauded for providing open access to a large quantity and variety of information, and for allowing people to communicate across geographical, ideological and social barriers. Similarly, social media platforms were thought to possess democratic potential, offering an unprecedentedly free and open debate on current events accessible to all, improving and pluralizing the public debate. However, cyber-pessimists have long emphasized the adverse effects online political communication have on society (Jacobs & Spierings, 2016). Numerous studies have found evidence that online communication on public forums as well as social media exhibits a degree of fragmentation (Bright, 2018; Colleoni, Rozza, & Arvidsson, 2014; Del Vicario et al., 2016; Dyagilev & Yom-Tov, 2014; Flaxman, Goel, & Rao, 2016; Garimella et al., 2018; Sunstein, 2007). Fragmentation here entails the idea that “online conversations about politics are divided into a variety of groups, and that this division takes place along ideological lines, resulting in people only talking to those who are ideologically similar” (Bright, 2018). This is especially concerning, as exposure to and tolerance for other ideas are considered essential to good democratic practice.
While there is no consensus on an exact definition of echo chamber, it can generally be understood as a high degree of this fragmentation. There are two main recurring aspects prevalent in definitions given by scholars: homophily and ideological isolation (Flaxman et al., 2016). Homophily refers to a low distance between ideological positions. Messages within the chamber have a high level of sameness, only conforming to each other (Dubois & Blank, 2018). Ideological isolation refers to the preference of those in an echo chamber to connect only with each other and cut ties to those outside the chamber. There is no engagement with alternative ideas (Bastos, Mercea, & Baronchelli, 2017). Visually, therefore, the echo chamber therefore appears as a closed group as opposed to showing an overlap in publics. But why and how do these echo chambers form?
There are two main mechanisms behind the echo chamber: selective exposure and algorithmic interference. Selective exposure here refers to how greater access to information does not necessarily result in individuals taking in a wide range of views (Barberá, Jost, Nagler, Tucker,
15
& Bonneau, 2015). This is because people have the tendency to expose themselves to information that reinforces their existing views. Selective exposure goes hand in hand with psychological mechanisms like selective retention and selective perception, through which people are more likely to recall and perceive information aligned with their views. These psychological mechanisms are often mediated by emotional responses eliciting emotional and social contagion (Brady et al., 2017). Algorithmic interference refers to social media platforms’ use of algorithms to filter content. This interference strengthens the effect of selective exposure, by pre-selecting the information individuals are shown according to their personal preferences (Krasodomski-Jones, 2016). In a sense, social media filters mitigate or amplify personal bias.
The concern with echo chambers is that as more people rely more on social media platforms as a (primary) news source, as alternative views are filtered out, polarization and fragmentation are stimulated. This does not seem far-fetched: a 2016 research shows that American adults under age 49 generally trust political information shared in their social media network more than news from other sources (Barthel et al., 2016). And while social media can be linked to greater exposure to opposing perspectives, it is less diverse than for example a web-search (Flaxman et al., 2016).
More research supports the idea that fragmentation is prevalent on social media platforms and has predominantly negative effects. For example, a 2017 research on the ramifications of the transition from low to high-choice political information environment on the character and quality of our democracies finds that “the emergence of social media results in a decreasing diversity in political information, lower news quality, growing inequality in political knowledge and increasing fragmentation and polarization” (Van Aelst et al., 2017).
Social media as a source of political information and debate are that online coverage of controversial events is more partial and unbalanced (Nikolov, Oliveira, Flammini, & Menczer, 2015). Most evidence for this effect concerns political discourse. News presenting a strong political perspective are disproportionately shared amongst those who share those perspectives (Krasodomski-Jones, 2016). This effect is found most strongly in partisan networks, with those in the network often oblivious to, or openly hostile to alternative perspectives (Nourbakhsh, Liu, Li, & Shah, 2018).
While much of the research on the topic depicts echo chambers as an underrated and quite prevalent threat to democratic values, some evidence suggests that the findings showing an increase in polarization on especially Twitter should be taken with a grain of salt. Most
16
importantly, valid criticisms were outed on how fragmentation has been measured. Dubois and Blank point out that even if one social medium is polarized, people still use other sources. In order to measure whether someone is in an echo chamber, their full offline and online “media diet” has to be taken into account (Dubois & Blank, 2018). The larger and more varied your media diet, the less likely you are in an echo chamber. Therefore, people at risk to end up in an echo chamber are largely those that lack political interest, for example using only 1 news medium to get their information (Dubois & Blank, 2018). Research into political activity reinforces this idea. Those that are most politically active are in the most balanced environment, which is in line with the disagreement theory: political activity is invigorated by disagreement with peers (Dyagilev & Yom-Tov, 2014). Therefore, the most active users of social media platforms carry their opinions to people with differing views. Still, the majority of social media users is nowhere near this level of activity. Assuming that the most active Twitter users Tweet and read Twitter daily, this automatically eliminates at least 52 percent of Twitter’s monthly user base (Recode, 2017).
In sum, research points toward the negative effects of social media platforms on good democratic practice. High levels of fragmentation such as echo chambers are encouraged by reinforcing personal bias and algorithms invisible to social media users. This is especially problematic when it concerns political information or news with strong political perspectives. However, measuring the impact of such fragmentation on people still provides challenges.
17
3. Research Questions and Hypotheses
The aim of this paper is to map to what extent and effect FVD uses emotional language on Twitter. This can be split into two research questions, the first being:
(RQ1:) To what extent is emotional language used by FVD on Twitter?
As members of a political party, it is expected that Tweets by FVD members regularly touch upon political topics. As described in section 2.1, these topics are associated with more emotional expressions. There is also an incentive for FVD members to Tweet more emotionally: Tweets that get most exposure among the in-group are those were emotions are present. FVD’s best tactic to get exposure with their in-group is through expressing emotions in their Tweets, rather than tweeting predominantly neutrally. This leads to the following hypothesis:
H1: FVD more often uses emotional expressions in their Tweets than not.
Lacking the social barriers and gatekeepers that would apply offline and in traditional media, social media amplify the use of harsh and extreme language. This is especially rewarding for populist actors, as they can encourage negative views in their network toward populist enemies such as political, economic, legal and media elites. Therefore, the following hypothesis can be deduced:
H2: Tweets by FVD more often contain negative than positive expressions of emotion.
The second research question I propose is:
(RQ2:) To what effect is emotional language used by FVD on Twitter?
Emotional Tweets should be more popular than neutral ones. Both negative and positive sentiment should influence sharing behavior positively. However, negativity was shown to have a larger gradient than positivity, with negativity being perceived as more negative than positive statements were considered positive due to so-called negativity bias. Research has also stressed the opportunity social media provide populist parties in harshly attacking their perceived enemies. The FVD Twitter accounts should be connected mostly to their primary audience: those they claim to represent. This in-group should respond strongly to negative sentiment expressed about populist enemies. I expect this negativity to be emotionally contagious. Therefore, the following hypothesis is presented:
18
4. Research Design
4.1 Case
4.1.1 Forum voor Democratie
FVD is a relatively new political party. Emerging first as a thinktank in 2015, it transitioned into politics in September 2016. The success of FVD could simply be dismissed as an ideological gap in the Dutch landscape being filled. The party quickly created a lot of buzz with criticisms on the political establishment. A series of scandals surrounding leader Thierry Baudet and other members ensured constant exposure in the media. However, FVD’s self-attributes their success to their online presence. This presence was also a focal point of their campaign. This may show in their constituency, in which men are overrepresented: 76% of FVD voters are male (I&O Research, 2017). FVD has also been very successful in reaching a group considered hard to reach in campaigns: youths. This group is considered one of the hardest groups to get involved in politics, and relatively active online. Social media have been an important channel to this end, with Baudet retweeting “memes” that he is often at the center of. The FVD youth organization (JFVD) reached a thousand members within six hours of its announcement on social media platforms, and has since then reached six thousand members, making it the second largest political youth organization of all parties (JFVD, 2017). Of all political parties, they also have the largest following on Instagram, which is mainly focused on youths. Research marks that the voter retention of FVD is very high, with 88% (I&O Research, 2017). While this may be due to the novelty and age of the party, it nonetheless indicates that those that vote FVD are invested in the FVD brand or ideology.
FVD combines populism with a nativist, right-wing ideology. This is shown in their main points on their website. One of the main points in their political program is the protection and promotion of Dutch national values that are under pressure by (Islamic) immigrants. Their self-proclaimed main focus, however, is to “break open the cartel of political parties” (Forum voor Democratie, 2017). FVD emphasizes the corruptness and nepotism among established political parties and the elite. The interests of these parties oppose those of the Dutch people. In sum, FVD claims to represent the true interest of the people and to oppose the self-interested established elite in typical populist fashion. FVD also fits the populist characteristic of a highly centralized party structure, with some party members even resigning due to a lack of internal democracy in favor of Thierry Baudet and Theo Hiddema’s authoritarian leadership (Het Parool, 2018).
19
Being new, their entire existence is easily retrieved, as it was documented. It is possible to look at how the party positioned itself on the online political landscape overtime. FVD is known to have a large online support, have a very active online presence, which includes an online platform through which grassroot movements and petitions can be started and supported (JFVD, 2017).
4.1.2 Twitter
Twitter is a social media platform on which people communicate using short messages with a maximum of 140 characters, called “Tweets”. The standard news feed of a user consists of messages of users that have been “followed”, messages tagged with “hashtags” (topics) relevant to the user, as well as “mentions”, in which the user is tagged.
Twitter has proved a fruitful resource for research. This has three reasons: data availability, usability and its popularity among politicians (Jacobs & Spierings, 2016). The Twitter application programming interface (API) allows the extracting of tweets up until seven days after the initial tweet when searching for hashtags or mentions. The API gives access to a larger collection of Tweets when going on a “per user” basis, allowing the retrieval of their latest tweets up to a maximum of around 3000 tweets. Facebook, for example, is less forthcoming about giving out data of their users, especially after having been criticized for bad privacy practices. Data collection on their users is nigh impossible.
Twitter’s data is also highly usable. Many previous researches have tackled what can and cannot be extracted from tweets (Barberá et al., 2015; Jungherr, 2016). Follows, retweets and mentions are indications of the ideological position of Tweeters. Retweets and mentions show who are interacting with who, and retweets show sharing behavior. Follows give information on ideological similarity: people are more likely to follow those that are ideologically similar to themselves (Barberá et al., 2015). Retweets are highly modular and segregated into more homogenous communities (Conover, Ratkiewicz, & Francisco, 2011). Mentions on the other hand are less homogeneous, as hashtags are used to argue with ideologically opposed users, and result in exposure to more diverse content (Barberá et al., 2015).
Twitter is interesting to check for the use of emotional language of populists, as it has been shown to be more polarized than other platforms (Nikolov et al., 2015). It is often used to discuss controversial events that are highly polarizing. Its format in some ways complements populist ideology (Engesser, Ernst, et al., 2017; Jacobs & Spierings, 2018). Without any editing or
20
censoring through gatekeepers or social barriers, and through the allure of virality, the short microblogging format invites harsher language and through this higher emotions (Engesser, Ernst, et al., 2017). Web sources are also generally more partial and unbalanced. Only a small fraction of readers visits more than two different web sources (Gil de Zúñiga, Weeks, & Ardèvol-Abreu, 2017). In other words, of all social media, it is most likely to find an echo chamber effect on Twitter. If no evidence is found on Twitter for polarized political communication with high emotions, it is less likely that such evidence would be found on other, less polarized social media platforms.
21
4.2 Dataset
In this section, I will describe the dataset used for this research.
To find out how FVD tweets, first a representative sample of tweets had to be gathered. Using the R package “rtweet”, a collection of tweets was downloaded from the 5 verified FVD Twitter accounts. These accounts were: the official FVD Twitter account, party leader and member of parliament (MP) Thierry Baudet, MP Theo Hiddema, and Annabel Nanninga and Anton Schijndel, two FVD representatives recently elected to the Amsterdam municipality. Limiting the FVD politicians included in this research to those that have been elected will ensure that they are known to a broader audience and are acceptable as representation of FVD. For each account, the maximum number of recent tweets were retrieved as allowed by the Twitter Application Programming Interface. The limitations on the retrieval of these tweets were their age and amount. The data collection resulted in a total of 9852 tweets between February 27th of 2017 and May 23rd of 2018 and is shown in Table 1.
Accounts Twitter handle Number of tweets (N) Oldest tweet
Forum voor Democratie @fvdemocratie 3192 February 27th, 2017
Thierry Baudet @thierrybaudet 3186 May 25th, 2017
Theo Hiddema @thiddema 262 January 24th, 2017
Annabel Nanninga @ananninga 3197 August 9th, 2017
Anton Schijndel @schijndelanton 15 October 23rd, 2017
Table 1: Overview of data gathered from Forum voor Democratie Twitter accounts.
The N for tweets from the official FVD account, Nanninga and Baudet are about the same. The Twitter account of Schijndel is an outlier, as it has a comparably low N. While this low N may disallow us to draw valid conclusions on an individual level, his tweets will be used in the combined FVD sample. Each tweet is a short message used to converse on Twitter. Besides the content of the message in the tweet, further information on it is also stored. This includes the Twitter account that tweeted it, the exact date it was posted and the number of times the tweet has been favorited or retweeted.
22
4.3 Method & Analysis
To find out what emotions FVD express online, sentiment analysis was performed on the retrieved Tweets. Sentiment analysis is a computational method to identify opinions, sentiments and subjectivity in unstructured texts (Bae & Lee, 2012). Its two main uses are to classify statements as subjective or objective, or as expressing positive or negative sentiment (Deitrick, Valyou, Jones, Timian, & Hu, 2013). For this research, I will focus on the latter use. The sentiment analysis technique used here is semantic orientation. This entails the creation of a dictionary of subjectively meaningful words, and using it to score the content of data (Deitrick et al., 2013).
I chose sentiment analysis over other types of content analysis because it is particularly suitable for quickly capturing emotional expressions in large quantities of text. However, it is important to stress some restrictions of the method. First and foremost, sentiment analysis does not allow the capturing of non-textual content (Nakov, 2017). This is problematic, because social media platforms are commonly used to share videos, hyperlinks and images. However, there does not yet exist a comprehensive way to analyze both text and such visual data (Nakov, 2017). While leaving out visual content and links may paint a skewed picture, written text should still give a good indication of FVD’s emotional expressions, especially on political topics. A second issue is that sentiment analysis does not capture latent content (Nakov, 2017). For example, dog whistling, sarcasm or context-related emoting is not captured. A solution to this would be to go with a machine learning sentiment analysis technique, which is trained to distinguish between different sentiment classes, or to use a dictionary which for example includes popular dog-whistling terms (Deitrick et al., 2013). The problem with these solutions is that they are not as reliable, as these types of text are open to (mis)interpretation even to insiders. Therefore, within the existing limitations, sentiment analysis is the best option.
To prepare the data for sentiment analysis, it was cleaned, removing non-word components from the tweets like symbols and hyperlinks. The dictionary used was the Dutch translation of the NRC Emoticon Lexicon. This word-emotion association lexicon is a list of words that are indicative of each emotion. It has entries for more than 10,000 word-sense pairs. The lexicon was created through crowdsourcing and tested thoroughly to ensure its reliability (Mohammad & Turney, 2013). From the dictionary, the subsets of positive and negative emotions were used, as well as four of the basic emotions: joy, disgust, fear and anger respectively.
23
Applying the dictionary to the dataset resulted in a count of words expressing emotions for each tweet. Two separate analyses were done to interpret the results. In the first one, the focus was on the word counts for positive and negative. Tweets with a word count of 1 or higher for positive or negative were scored as such, tweets with both positive and negative scored were scored as “mixed” or inconclusive, and tweets without any emotions expressed were scored as “neutral”. As the dictionary based on the four basic emotions is not binary, it cannot be analyzed the same way. The emotions overlap with each other, as well as with the positive and negative scores. For example, the word “verlating” or “abandonment” scores on negative, fear and anger. To check the dictionary’s accuracy in marking the Tweets, I will hand check a random sample of five Tweets in section 4.3.1. This will give insight into how sentiment analysis works, and what its reliability is when faced with the challenges described above such as sarcastic or ambiguous language.
To find out whether there is a relationship between certain expressed emotions in Tweets and their popularity, I will look at the retweet count for every Tweet and perform a correlation test between this count and the present emotions.
4.3.1 Sentiment analysis check
To check how accurately the used word-emotion association lexicon deals with the dataset of FVD Tweets, I have hand checked three Tweets. These Tweets were selected by using the first three numbers generated by a random number generator configured to select a number between 1 and the number of Tweets in the dataset. In this test I will focus on the text that was analyzed. The words that matched the lexicon are bolded out in each Tweet. The complete original Tweets and English translations of the texts can be found in the appendix at the end of this paper.
4.3.1.1 Tweet 1
[@fvdemocratie] Draghi: "Elk land dat de Euro wil verlaten moet eerst afrekenen!" Dat is dan €102 miljard Euro graag voor NL. #Nexit #Draghi
#Target2 #FvD Score: 1 negative, 1 anger, 1 fear.
The sentiment analysis marked 1 point in each negative, anger, and fear for this Tweet, making it count overall as expressing negative sentiment. This score is based on the marked word “verlaten”, which can be translated as “leaving” or “abandoning”, and is classified as corresponding with negative sentiment, anger and fear in the NRC Emotion Lexicon. Interestingly, this word is part of a quote by European Central Bank President Mario Draghi that FVD is
24
responding negatively to. The comment on the quote made by FVD does not lend itself well for sentiment analysis: the tone of such a quip gets lost, and it lacks emotional words that might allow a better categorization of the Tweet. Based on the tone of this quip, I would score this Tweet on negative and anger, but not on fear. While this Tweet illustrates how sentiment analysis may fail to grasp the underlying tone in any given text, it does correctly classify it as negative. It does not mistake the quip for positive or grateful sentiment which may sometimes be the case with ambiguous language.
4.3.1.2 Tweet 2
[@thierrybaudet] Geweldig klassieke muziek festival Wonderfeel: https://www.facebook.com/WonderfeelFestival/ … top! #renaissance Score: 3 positive, 2 joy.
The analysis easily picks up on the very positive sentiment expressed in this Tweet, in which FVD party leader Baudet shows off his personal interests and life outside of politics. “Top”, “geweldig”, and “festival” (“great”, “splendid” and “festival”) are the words that made for this marking. The lexicon correctly associates these words with positivity and joy, and the Tweet is correctly classified here as overwhelmingly positive.
4.3.1.3 Tweet 3
[@thierrybaudet] Ziet nou niemand het grotere plaatje? Eerst creëert de EU chaos, oorlog, ellende; en daarna roept de EU op tot centrale machtsvorming,
tot een Europees leger, tot één groot rijk. Precies zoals Bismarck het deed in de 19de eeuw.
Score: 3 negative, 2 disgust, 2 anger, 3 fear.
The lexicon correctly identifies the negative sentiments of anger, disgust, and fear in this Tweet. Baudet’s Tweet has a tone of despair to it: “is everyone blind to the absolutely awful things happening in front of our eyes?”. Chaos, “oorlog” and “ellende” (“war” and “misery”) are all strong indicators of very negative sentiment, and the sentiment analysis does not fail to pick up on this.
As anticipated, the main trouble with sentiment analysis is properly scoring ambiguous or mixed content. This is illustrated by Tweet 1. But while the three Tweets coincidentally did not include a mixed score that included negative and positive sentiment, many Tweets do. These ambiguous scores are harder to analyze. Some scholars approach such mixed content by scoring
25
words with negative sentiment negatively (Bae & Lee, 2012). For example, a Tweet with two positive and 3 negative words will be scored +2 -3 resulting in a negative -1 score. As a result, this Tweet would be categorized as a slightly negative Tweet in their scale. Using this “scale” to categorize Tweets, a Tweet that contains one negative and one positive sentiment counts as neutral is simply misleading. I have chosen not to use this way of scoring, because I find it does not acknowledge the limits of sentiment analysis in recognizing content with ambiguous or mixed content. The analysis in this research will focus on the convincingly positive, negative and neutral Tweets, because mixed Tweets simply leave too much room for mistakes in classifying the texts.
26
5. Results
5.1 Sentiment analysis
First, a sentiment analysis was performed on the cleaned-up dataset of FVD Tweets. The results of the initial sentiment analysis are shown in Table 2. The table shows the average amount of words in each tweet that are emotional. Here it can be seen how many words on average with a certain emotion are expressed. In general, it can be said that when positivity is expressed, the most words are used to do so, while joy and disgust are significantly more uncommon.
Nanninga Hiddema FVD Baudet
Anger 0.232 0.297 0.270 0.246 Disgust 0.134 0.009 0.009 0.009 Fear 0.240 0.282 0.240 0.230 Joy 0.161 0.114 0.143 0.154 Negative 0.393 0.317 0.330 0.334 Positive 0.620 0.637 0.706 0.631
Table 2: Average words per Tweet expressing emotions for each Twitter account.
The FVD official Twitter and the accounts of the individual politicians express roughly the same kinds and amounts of emotions. Notably, Nanninga expressed disgust and negativity the most. Relatively, when emotions are expressed on Thierry Baudet’s Twitter account, it is more likely that words are used associated with disgust, and slightly less likely with anger – with about 1%.
Graph 1: The percentage of NRC Emotion Lexicon dictionary words expressed in the tweets of all FVD accounts.
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Nanninga Hiddema FVD Baudet
Pe rc e n tag e o f d ic ti o n ar y wo rd s i n Twe e ts Mixed Neutral Negative Positive
27
Graph 1 shows how many Tweets are positive, negative, neutral or a mix of positive and negative. All four Twitter accounts show similar proportions in their use of sentiment. Theo Hiddema posts most neutrally, with a score of 113 out of 262 Tweets, accounting for half his Tweets.
5.2 Correlation test
Second, a scatterplot was plotted out of all the available data to indicate if there was a possibility of a linear relationship between the variables. As this was the case, a Pearson Correlation test was then performed to see if there is a correlation between the retweet count of Tweets and the emotions expressed in them. The results are shown in Table 3. The correlation coefficient shows the linear relationship between the variables, with negative coefficients indicating a negative relationship, and positive coefficients a positive
relationship. Pearson coefficients closer to 0 are more likely to show negligible or no correlation, while coefficients closer to 1 and -1 show stronger relationships. In this case, Table 3 clearly shows that the relationship between retweet count and negative sentiment, disgust, anger and fear are negligible in the used data.
The main finding of the Pearson Correlation test concerns positive sentiment. Table 3 shows the negative relationship between Tweets with positive sentiment and retweet count as significant for a confidence level of 0.95.
Emotion Coefficient P value
Positive -0.028 0.01* Negative 0.009 0.33* Disgust -0.003 0.74* Anger -0.001 0.94* Joy -0.017 0.09* Fear 0.002 0.84*
Table 3: Results for the correlation test
between retweet count and listed emotion. *Significant at 5%.
28
6. Discussion
The results show that except in the case of Hiddema, the majority of FVD Tweets contain positive, negative or mixed sentiment. Therefore, H1 can be accepted: Tweets with emotional content outnumber neutral Tweets. However, Hiddema’s deviance from the other Twitter accounts in this regard does raise questions. What explains Hiddema’s preference for neutral terms? One possible explanation is that Hiddema does not discuss political topics as much as other FVD party members. As has been argued before, political topics are more controversial than other topics, and thus involve more sentiment. Another explanation is that it is part of his personality or political communication style. More personal and behind-the-scenes Tweets of the more eccentric Baudet and Nanninga may have skewed the results toward them showing more sentiment. Consider, for example, Baudet’s rejoicing over a classical musical festival shown in the randomly highlighted Tweet in section 4.3.1.3. In Hiddema’s entire Tweet history, he has only twice referred to a personal preference, namely when he Tweeted about attending a Helene Fischer concert. Accepting this seemingly plausible explanation consequentially means that the aspect of personalization possibilities which were thought to make Twitter so attractive for populists may in the end not be as important as theorized.
H2 encapsulated the expectation that negative sentiment in Tweets is more common that positive sentiment. Table 1 however shows that FVD overwhelmingly uses more positive than negative or mixed expressions of emotion in Tweets. Therefore, H2 cannot be accepted. When it comes to the amount of words used to express sentiment, the results in Table 2 may in fact illustrate the negativity bias theory. When positivity is expressed, more words with positive sentiment are used than when negativity is expressed. This is in line with the assumption that negativity has a steeper scale than positivity. To express the same level of positivity as an equally negative event, more words are necessary.
Based on this analysis, RQ1 can be answered as follows: in their online communication style on Twitter, Tweets by FVD mostly contain positive or neutral sentiments. Both Tweets expressing positive emotional language as well as neutral Tweets were more common than negative, making negative sentiment the least common. The extent to which FVD uses emotional language is thus significant, but as opposed to my expectations the expressed sentiment was for the majority positive. While Nanninga, Baudet and the official FVD Twitter account show similar
29
proportions in their use of emotional language, Hiddema is more neutral. This challenges the idea of a campaigning party strategy that encourages using sentiment.
The second part of the analysis was a Pearson Correlation test between retweet count and sentiment. This test showed that the relation between positive sentiment and Tweet popularity was significant. The coefficient in this correlation was -0.028, meaning that a one unit increase in positive expression is expected to trigger 0.028 or approximately 3 percent less Tweets. Interestingly, this does not support the theory that positive Tweets are shared more. Instead, the presence of words expressing positive sentiment correlates with a lower popularity, where negative Tweets did not show any significant effect.
While the other sentiments lacked significant relations, they did show an interesting difference in the coefficients. Note that Tweets in which negative emotions or fear are expressed correlate positively to the number of retweets, as opposed to the negative correlation of positive sentiment and the number of retweets. As the number of Tweets with negative sentiment was lower than the number of positive Tweets, these negligible may be a result of a too small dataset, which could be solved by increasing the number of Tweets analyzed.
Using these results, H3, which states that Tweets with negative sentiment are retweeted more often than those with positive sentiment, cannot be rejected nor accepted. The significant negative correlation between positive sentiment and popularity is not enough to prove H3, especially considering no other relationship between sentiment and retweet count was found. The relation between negative sentiment and popularity could be more random or mediated by other variables such as the topic of the Tweet. More research is needed to properly map this relation.
Based on this analysis, RQ2 can be answered as follows: the effect of FVD’s use of emotional language is not clear-cut. FVD’s use of positive sentiment negatively impacts their Tweets’ popularity, yet this sentiment is most commonly used in their Tweets. This implies that generally speaking, negative sentiment is favored among FVD followers. If so, negative Tweets may contribute to a polarizing echo chamber effect. However, since the number of Tweets with negative sentiment is relatively low, the effect is debatable. Further research with a higher N and preferably more political parties as a benchmark is needed to establish this relation.
30
7. Conclusion
In this paper, I have researched to what extent and effect the Dutch populist political party Forum voor Democratie (FVD) uses emotional language on Twitter. Using research on populists’ social media use and the effects of sentiment on morality and virality, I theorized that populists are more likely to express negative sentiment in Tweets in order to express their hostility toward populist enemies, captivate and mobilize an ideologically homophilic online in-group, while at the same time increasing the odds of going viral. Such a campaigning strategy, I have argued, at its worst may encourage increased political polarization and fragmentation through the echo chamber effect, which is particularly prominent online.
Through sentiment analysis, this research has shown that Tweets by FVD most often contain positive sentiment. They are a bit less likely to be neutral, and a lot less likely to contain mixed or negative emotional language. While in the majority of Tweets some sentiment was expressed, the data did not match the expectations of a populist preference for negative messages. In an interesting turn of events, positive sentiment in Tweets was found to be correlated to a lower popularity. This finding opposes emotional sharing theories that presuppose positive messages are more likely to be spread. It also challenges the idea that populist parties have a centralized online political campaigning strategy aimed specifically at stimulating an echo chamber effect or going viral. Instead, it presents us with a lot of questions and grounds for further research.
The most important question is: can these results be generalized to all political and populist parties on the Dutch landscape? A comparative study on sentiment in Tweets of the Netherlands’ largest political parties is the logical next step in this research. A higher number of Tweets improves the likelihood of finding relations between sentiment and popularity. Most above all, a benchmark is needed to compare the findings in this study to. To understand FVD’s use of Twitter, we must understand the context in which this usage takes place. Therefore, future research should focus on analyzing Tweets of all big Dutch political parties. In this way, populists’ online strategies can be compared to “regular” political parties, and FVD can be compared to both in a significant way.
Another route that future research might take, is to look at specific Tweets in the positive or negative categories to analyze whether they are specifically tailored to fit populist ideas. For example, are all Tweets that express negative sentiment anti-establishment Tweets, and are Tweets
31
with positive sentiment focused on glorifying the people the populists claim to represent? One way to do this is to combine sentiment analysis with automatic topic extraction.
Finally, to assess the potential negative consequences of emotional language in social media, sentiment analysis should be combined with network mapping. Doing this allows us to map whether emotional Tweets are only spread among the in-group as hypothesized in echo chamber theories, or also reach the out-group. It is not possible to answer a question concerning negative effects of populists’ use of emotional language online without taking this factor into account.
32
8. Bibliography
Bae, Y., & Lee, H. (2012). Sentiment analysis of twitter audiences: Measuring the positive or negative influence of popular twitterers. Journal of the Association for Information Sciences and Technology, 63(12), 2521–2535. https://doi.org/10.1002/asi
Barberá, P., Jost, J. T., Nagler, J., Tucker, J. A., & Bonneau, R. (2015). Tweeting From Left to Right: Is Online Political Communication More Than an Echo Chamber? Psychological Science, 26(10), 1531–1542. https://doi.org/10.1177/0956797615594620
Barthel, M., Gottfried, J., Mitchell, A., & Shearer, E. (2016). The 2016 Presidential Campaign - a News Event That’s Hard to Miss. PewResearchCenter. Retrieved from
http://www.journalism.org/2016/02/04/the-2016-presidential-campaign-a-news-event-thats-hard-to-miss/
Bastos, M. T., Mercea, D., & Baronchelli, A. (2017). The Spatial Dimension of Online Echo Chambers. CoRR, 10(4). Retrieved from https://arxiv.org/pdf/1709.05233.pdf
Brady, W. J., Wills, J. A., Jost, J. T., Tucker, J. A., & Van Bavel, J. J. (2017). Emotion shapes the
diffusion of moralized content in social networks. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.1618923114
Bright, J. (2018). Explaining the Emergence of Political Fragmentation on Social Media: The Role of Ideology and Extremism. Journal of Computer-Mediated Communication, 23(1), 17–33. https://doi.org/10.1093/jcmc/zmx002
Colleoni, E., Rozza, A., & Arvidsson, A. (2014). Echo Chamber or Public Sphere? Predicting Political Orientation and Measuring Political Homophily in Twitter Using Big Data. Journal of
Communication, 64(2), 317–332. https://doi.org/10.1111/jcom.12084
Conover, M., Ratkiewicz, J., & Francisco, M. (2011). Political polarization on twitter. Icwsm, 133(26), 89–96. https://doi.org/10.1021/ja202932e
de Boer, N., Puts, J., Livestro, B., & van Brussel, T. (2014). Twitter en de Tweede Kamer. Retrieved from http://webershandwick.nl/wp-content/uploads/2014/03/Twitter-en-de-Tweede-Kamer.-Weber-Shandwick.pdf
de Hond, M. (2018). Peil.nl Peiling 6 mei 2018. Retrieved from https://home.noties.nl/peil/politieke-voorkeur/
33
Deitrick, W., Valyou, B., Jones, W., Timian, J., & Hu, W. (2013). Enhancing Sentiment Analysis on Twitter Using Community Detection. Communications and Network, 05(03), 192–197.
https://doi.org/10.4236/cn.2013.53022
Del Vicario, M., Vivaldo, G., Bessi, A., Zollo, F., Scala, A., Caldarelli, G., & Quattrociocchi, W. (2016). Echo Chambers: Emotional Contagion and Group Polarization on Facebook. Scientific Reports. https://doi.org/10.1038/srep37825
Dubois, E., & Blank, G. (2018). The echo chamber is overstated: the moderating effect of political interest and diverse media. Information Communication and Society, 21(5), 1–17.
https://doi.org/10.1080/1369118X.2018.1428656
Dyagilev, K., & Yom-Tov, E. (2014). Echo chamber amplification and disagreement effects in the political activity of Twitter users. Retrieved from http://arxiv.org/abs/1403.7102
Engesser, S., Ernst, N., Esser, F., & Büchel, F. (2017). Populism and social media: how politicians spread a fragmented ideology. Information Communication and Society, 20(8), 1109–1126.
https://doi.org/10.1080/1369118X.2016.1207697
Engesser, S., Fawzi, N., & Larsson, A. O. (2017). Populist online communication: introduction to the special issue. Information, Communication & Society, 20(9), 1279–1292.
https://doi.org/10.1080/1369118X.2017.1328525
Ernst, N., Engesser, S., Büchel, F., Blassnig, S., & Esser, F. (2017). Extreme parties and populism: an analysis of Facebook and Twitter across six countries. Information Communication and Society, 20(9), 1347–1364. https://doi.org/10.1080/1369118X.2017.1329333
Flaxman, S., Goel, S., & Rao, J. M. (2016). Filter bubbles, echo chambers, and online news consumption. Public Opinion Quarterly, 80(Specialissue1), 298–320. https://doi.org/10.1093/poq/nfw006
Forum voor Democratie. (2017). Standpunten - Forum voor Democratie. Retrieved from https://forumvoordemocratie.nl/standpunten
FTM. (2017). De digitale strategie van Thierry Baudet: “net als Trump.” Retrieved from https://www.ftm.nl/artikelen/thierry-baudet-net-als-trump
Garimella, K., Morales, G. D. F., Gionis, A., & Mathioudakis, M. (2018). Political Discourse on Social Media: Echo Chambers, Gatekeepers, and the Price of Bipartisanship. CEUR Workshop
