1
Twitter, football, and the fans:
An emotion diffusion analysis of tweets associated with the European Super League
Pavlos Ferlachidis 13304852
Master’s Thesis
Graduate School of Communication Master’s in Communication Science
Supervisor: Dr. C.W. (Christel) van Eck Date of completion: 11 February 2022 Number of words: 7163
2
Abstract
The creation of the European Super League (ESL) caused an immediate backlash from various stakeholders in the football world and beyond. A significant part of these reactions was expressed on social media. After collecting a sample of approximately 450000 tweets, this study investigated the emotional language of tweets associated with the ESL, as well as the dissemination of these tweets through retweeting. The results indicated that positive emotional language and positive emotions such as anticipation and trust appear more than negative emotional language, and distinct negative emotions such as anger and sadness in the dataset. This is also true for the timeframe after the suspension and during the days where a “spike” in tweet creation was noted.
Furthermore, for the dissemination part, positive sentiment and emotions have a bigger effect both on retweetability and the number of shares. The results are contrary to the majority of the studies that were reviewed during this study. As a result, more research is suggested for this topic, including parameters such as influencers/super-diffusers of content, potentially using a different methodological tool.
Keywords: ESL, The Super League, Twitter, football, emotion analysis, diffusion analysis
3
Introduction
On April 18, 2021, the creation of a new, 20-team closed-league championship by the name of European Super League (ESL) (formally “The Super League”), was
announced. ESL was founded by 15 of the most popular, rich, and successful clubs, aiming to replace Europe’s top football competition (Reynolds, 2021). Multiple football stakeholders, including fans, coaches, and players, along with politicians around Europe, opposed the project immediately (Liew, 2021;). The overwhelming opposition, as documented in a snap poll with 79% of negative opinions (Ibbetson, 2021), led to the announcement of the temporary suspension of the league by the ESL (Millar, 2021).
Despite the severe backlash and the withdrawal of the majority of founding clubs from the league (Marcotti, 2021), the project was not officially abandoned. The presidents of key teams like Real Madrid, Juventus, and recently Barcelona, are still backing the breakaway league (Millar, 2021). In October, the ESL announced a different league format, where the qualification of every team will be based on national league placement (and not on the criterium of permanence for founding clubs, as before), plus the creation of a second-tier competition, with teams qualifying based on the same criteria (Wettach, 2021). As alleged, the economic toll of the COVID-19 pandemic, along with the loss of interest of younger generations for football as a product, in general, were the underlying reasons for the creation of the ESL (Marcotti, 2021).
The contemporary reality of the “economy of attention” (p. 286) requires sports organizations to maintain a multifocal approach, taking under consideration the
4 opinions of multiple stakeholders (Garcia, 2011). This means that a balance should be stricken between producing an eye-pleasing spectacle and winning titles on the one hand, and sustaining financial health on the other, while at the same time harnessing a lasting relationship with the fans, and promoting their social responsibility role
(Abosag et al., 2010; Rouvrais-Charron & Durand, 2009). Undoubtedly, football has been long associated with societal and political events and forces, maintaining deep roots in communities and ideologies (Power et al., 2020). The exclusivity of a league to a few powerful clubs, where participation is not earned meritocratically,
eliminating the risks of athletic or financial failure, was perceived to disregard the norms and ethics upon which European football was built (Liew, 2021).
The domination of commercialization and Murdochization1 (Lawrence and Crawford, 2018) in football, had already alienated a considerable segment of the (European) football fanbase (Merkel, 2012). Equally important, another fan segment remains engaged but dissatisfied, choosing football activism (Zaimakis, 2016) as the way to preserve its authenticity (Edensor, 2015) and showcase its contradiction with the new direction of the sport. Hence, fans tend to express themselves regarding developments they perceive as unfair. Besides the physically present protestations, reactions on the Web exist as well.
In the contemporary, digital environment, the primacy of mass media as the main communication medium between fans and football clubs in social media (SM) has
1Murdochization refers to the effects caused by the changes in the English football in the early ‘90s, when Rupert Murdoch’s media conglomerate bought the broadcasting rights for the English Premier League, and turned football from a formerly free to watch spectacle, into a fee-paying, globalized sports product.
5 been disrupted (Guerra et al., 2014). The relationship between fans and sports
organizations has changed, with fans’ opinions now considered to be at least as visible as the clubs’ (Anagnostopoulos et al., 2018). The shift has been recorded in the
production and the consumption processes of football, moving digital media in the epicenter of these developments, and leading to the consolidation of a post-broadcast model (Lawrence and Crawford, 2018). The tribal nature of sports fans thrives in the SM ecosystem, amplifying their fan identity (McCarthy et al., 2014). Thereby, due to the information dissemination architecture of SM (Manoli, 2019), the diffusion of news and information in SM is rapid (Gruber et al., 2015), while the discussion around sports can become highly polarized (Morgan & Wilk, 2021; Rouvrais-Charron
& Durand, 2009). Thus, the importance for a sports organization to control the narrative on SM, and protect its reputation, is imperative (Coombs, 2007).
Twitter, is considered the most disruptive of SM due to its extensive adoption by sports stakeholders, and its simple and fast use (Gruber et al., 2015). Described as a
“thermometer of social problems and events” (Jiménez-Zafra et al., 2021, p. 1), the main function of Twitter as a microblogging platform is the creation of tweets. The 280-character posts contain information concerning personal matters, social issues, or current news, as well as the expression of emotions (Wunderlich & Memmert, 2020).
On a secondary level, users can reply (using the @ symbol and adding the name of the user) or retweet (share with their followers) a user’s text (Tirdad et al., 2021; Zhang et al., 2021). Twitter attracts football discussions (Aloufi & Saddik, 2018), offering the capability to express strong emotional language (Lucas et al., 2017), and to react real- time to various events and news updates (Nazir et al., 2019), rendering it fitting to study the online content produced about these events (Yu & Wang, 2015).
6 Emotional expression is considered an important factor that influences fan behavior (Shakina et al., 2020), during the value and experience co-creation process between clubs and fans (Tinson et al., 2021). It can lead to positive results or a user-generated crisis for a football organization (Manoli, 2019), depending on fans’ reactions. To look into the reactions of fans after the announcement of the creation of ESL, an emotion diffusion analysis of tweets was chosen as the most appropriate method of studying the emotional responses and stances of fans on Twitter.
This study deals with a topic both relevant and ongoing. Until now, no other piece of scientific literature was identified, looking into the brisk failure of the ESL, a
potentially game-changing development in the world of football, and sports in general. The analysis will focus on whether emotion-laden language is more positive than negative, what are the distinct emotions that users express, on the days when the highest amount of tweets are posted (Nazir et al., 2019), during the crisis, and after the suspension of the project (Bhatia et al., 2019; Morgan & Wilk, 2021). A second, but no less important focus of the research, will be on the dissemination of tweets in the Twittersphere, through retweeting (Steiner-Correa et al., 2018). The identified gap in academic literature concerns the lack of studies employing emotion analysis to look into specific emotions expressed by SM users about a sports issue, and how these online reactions can potentially influence the decisions taken by sports organizations and club executives (Gong et al., 2021). Most of the studies identified in the literature review process were focused on classifying the expressed sentiment in their data, as positive, negative, or neutral, and only a small minority studied distinct emotions as well. The goal of this study is to advance the academic literature on the online
7 emotion expression process and the diffusion of content that includes emotion-laden language, by looking into the emotional language associated with the ESL as
expressed on Twitter.
Taking into consideration the abovementioned points, the following exploratory Research Question (RQ) is proposed:
RQ: To what extent can the use of emotional language in tweets associated with the ESL explain the swift failure of its project?
Theoretical framework The use of emotion in the contemporary age
Humans express their emotions in multiple direct or indirect ways, through speech, facial expressions, or writing (Sailunaz et al., 2018). After analyzing millions of books, it was concluded that in the last couple of decades, the use of emotion in language has surpassed the use of rationality (Scheffer et al., 2021). First, a distinction has to be made between sentiment and emotions. Sentiment -also referred to as affect in various studies-, concerns a quickly processed, general feeling or opinion classified as positive, negative, or neutral, while emotions are considered more distinct,
complex, less subtly expressed, associated with personal mood or social relationships (Sailunaz et al., 2018; Smith & Leiserowitz, 2014).
Several emotional models have been used in emotion classification studies, with three of them being common. Ekman’s model is based on six basic emotions, basically anger, fear, joy, sadness, surprise, and disgust. Plutchik’s model includes Ekman’s six emotions plus anticipation and trust, organized in bipolar pairs. Finally, Parrot’s
8 model overlaps with Ekman’s and Plutchik’s and is organized in a three-level
hierarchical structure to represent different levels of emotional strength (Bandhakavi et al., 2021). In this study, Plutchik’s eight basic, paired emotions (anger-fear, anticipation-surprise, joy-sadness, trust-disgust) will be considered as the basis for the analysis of the emotional language of tweets.
Studying emotional language in the social media context
Social media produce big amounts of data in terms of volume, velocity, and variety (Hartmann et al., 2016), requiring an automatization approach in their analysis (Ligthart et al., 2021). To this end, computational methods of text classification (Minaee et al., 2020; Yarch et al., 2021) are employed to study the emotional
language of SM, in contexts such as societal issues or commercial products (Ligthart et al., 2021; Yu & Wang, 2015). More specifically, natural language processing is used to identify the changes in textual data, looking into opinions and emotion (Minaee et al., 2020; Nandwani and Verma, 2021) or the opinions towards an entity or commodity (Jimenez-Zafra et al., 2021; Philander & Zhong, 2016; Yu & Wang, 2015).
Dissemination of content on Twitter
The use of eWOM about sports in SM. The emotion in digital platforms spreads through informal electronic word of mouth (eWOM) (Kim et al., 2020), a concept highly relevant in describing fan experiences in SM, like communicating with other fans and creating and posting content (Philander & Zhong, 2016). With eWOM, fans reach people beyond their circle of friends and acquaintances (Araujo et al., 2015), since a tweet can be seen by an average of 1000 users if retweeted (Kwak et al.,
9 2010). Additionally, the use of emotions in tweets that include more hedonic elements is shared more than messages that do not include emotion (Araujo et al., 2015).
The role of retweeting in content dissemination on SM. The two main ways of disseminating information on Twitter are diffusion (number of users that see the content, associated with elements such as a user’s followers) and virality (based on the effect of the dissemination of the message, all other external factors being neutralized) (Jimenez-Zafra et al., 2021). Retweeting is considered the main
mechanism for information dissemination (Hansen et al., 2011). Twitter users tend to disseminate information more through retweeting when the content triggers emotional reactions and is combined with informational cues, when the content is considered useful or valuable, or when using functions such as hashtags (Araujo et al., 2015), and/or URLs (Naveed et al., 2011; Suh et al., 2010).
Influence of positive and negative sentiment/emotions in retweeting content.
Whether negative or positive sentiment/emotions have a stronger effect on the dissemination of content through retweeting, differs from topic to topic. In general, a negativity bias exists in SM, since negative emotions usually induce stronger
responses and tend to go more viral compared to positive ones (Lucas et al., 2017;
Wang and Lee, 2020). In this direction, studies from Jimenez-Zafra et al. (2021), Zhu et al. (2020), and Tsugawa & Ohsaki (2017), about the Catalan referendum of 2017 and CDC tweets respectively, reported stronger effects of negative emotional
language on retweets, as well as virality and diffusion, highlighting the opposite effect for the inclusion of positive languageBurnap et al. (2014) analyzed Twitter data
10 related to a terrorist attack, concluding that the negative content did not propagate through retweeting, while positive tweets spread for longer.
Ferrara and Yang (2015) concluded that tweets that include negative emotions spread faster, while in terms of diffusion, tweets with positive emotions have more retweets.
They added that unexpected events are followed by the expression of negative sentiment. Hansen et al. (2011) used multiple datasets and their results indicated that both positive and negative sentiment can boost retweeting, depending on whether a tweet was categorized in the non-news or news category, respectively. Finally, Stieglitz & Dang-Xuan (2013) analyzed two datasets finding in one of them that tweets with positive content have a positive effect on the number of retweets, and in the second one that negative sentiment tweets are retweeted in a greater quantity.
To examine the content dissemination process on Twitter, and the importance of particular emotional language in tweets associated with the ESL, the following sub- questions (SQ) are proposed:
SQ1: Which are the most prevalent emotions in the language of the tweets associated with the ESL?
SQ2: How does emotional language influences the dissemination of tweets associated with the ESL through retweeting?
Based on the abovementioned literature but also taking into consideration the fact that the ESL constitutes a big event (Thelwall et al., 2012), introduced unexpectedly to the public (Bhatia et al., 2019), the conclusion is considered as negative for the
11 organization, the following hypotheses are posed about the relationship of users’
expressed emotions and retweets:
H1: Overall, negative emotional language is present significantly more than positive emotional language in tweets.
H2a: Tweets that contain negative emotional language are retweeted significantly more than tweets that contain positive emotional language.
H2b: The use of negative emotional language leads to a higher number of retweets compared to positive language.
Content dissemination and distinct emotions on SM. In terms of specific emotions associated with virality, according to Berger & Milkman (2012) high-arousal
emotions (eg. awe, anger, anxiety) tend to disseminate more than low-arousal emotions (eg. sadness). Guerra et al. (2014) mention that extreme emotions such as anger, anxiety, awe, and excitement are associated with arousal and preparation for taking on further action. They translate this preparation for action in the context of SM, as the expression of intense emotions by SM users. Pröllochs et al. (2021a) found that online rumors associated with aggressiveness tend to go viral one standard
deviation more than other emotions. Pröllochs et al. (2021b) concluded that false rumors on Twitter go more viral when they include discrete emotions such as trust, anticipation, and anger. Wang and Lee (2020) also identified negativity and anger as better predictors of diffusion than sadness or fear. Finally, Wang and Wei (2020) found in their study of cancer-related tweets, that joy was the emotion shared the most between Twitter users, but anger along with joy were the emotions that led to the most retweets.
12 Taking into consideration the abovementioned research about the expression of
emotional language in tweets, the following hypotheses have been formulated:
H3: Anger emotional language appears significantly more than any other emotional language (eg anticipation, surprise, joy, trust).
H4a: Tweets that contain anger emotional language, are retweeted more than tweets that contain any other emotional language (eg anticipation, surprise, joy, trust).
H4b: The use of anger emotional language leads to a higher number of retweets compared to other emotional language (eg anticipation, surprise, joy, trust).
Expression of emotions for an organization after a crisis
It has been argued before that a change in sentiment can have an immediate impact on a brand, and subsequently on the financial success of its product (Bonfim and
Furtado, 2017). Bonfim and Furtado’s (2017) study of football fans’ sentiment, looked into the gradual change of individual sentiment, depending on the events that affect them, concluding that a sentiment measure is complete when conducted in multiple different points. Morgan and Wilk’s (2021) sentiment analysis study of a 2018 sports crisis, revealed that in the occasion of an online crisis for an organization, a “crisis identity building process” (p. 10) is initiated by SM users, essentially
labeling and setting the boundaries of the unveiling situation on SM. They concluded, that if another crisis is not initiated following the initial, a shift to positive emotion is recorded after the initial crisis is considered over.
13 A study by Bhatia et al. (2019), examined the sentiment change on Twitter around a US Senate election and the NFL, on the principle that “expectations influence effect”
(p. 2), meaning that the combination of beliefs and outcomes affect the emotional reactions of humans. They found that especially regarding negative surprises, their emotive influence is higher compared to positive ones, indicating that fans do not necessarily change their opinion about a person or organization, even when the crisis is considered over. Conversely, according to Morgan and Wilk (2021
Based on the conflicting available data, and since the ESL affair came to a swift and satisfactory conclusion for the fans, the following hypotheses, based on Plutchik’s eight basic emotions are posed, to look into the Twitter users’ emotions after the suspension of the ESL:
H5a: Emotional language associated with anticipation is present significantly more than surprise after the suspension of the ESL.
H5b: Emotional language associated with joy is present significantly more than sadness after the suspension of the ESL.
H5c: Emotional language associated with trust is present significantly more than disgust after the suspension of the ESL.
Spikes in the creation of tweets
According to Lucas et al. (2017), excitement has been associated with stronger negative emotion, highlighting that the results are in contrast with the common knowledge in sports economics. Moving further, the analysis of their study
established a connection between a higher quantity of tweets per (football) World Cup game and a higher percentage of negative tweets. Thelwall et al. (2012) showed that a
14 spike in interest implicating an important event on Twitter is primarily associated with negative sentiment and negative emotional reactions, even when the context of the discussion is positive. Furthermore, Tsugawa & Ohsaki (2017) added that the virality of negative sentiment dissemination leads to flaming (spread of negatively charged rumors or fake news). Based on previous research associated with increase of user activity on Twitter, the days where at least 1% of the tweets was posted were identified, formulating the following, sixth hypothesis:
H6: On the four days when the highest number of tweets associated with the ESL are posted, the emotional language is significantly more negative than positive.
Methodology Data Collection
The data were scraped using ‘twarc2’, a command-line tool that uses Python programming language, and have been extracted through Academic Twitter API (Application Programming Interface). Twitter API is the intermediary that links Twitter with the party requesting the data and safeguards the process against data leaks. Using the Academic level of license, a researcher is offered access to “Twitter’s real-time and historical data, to download as 10 million tweets per month” (Twitter Developer Platform, n.d.), limited though to tweets whose users have set their privacy settings of their posts to ‘public”.
The data were collected on December 3, 2021. The sample includes tweets from April 18, 2021 (the date the creation of ESL was announced), and until December 3, 2021
15 (the last day of data collection). In total, 538252, tweets were collected using these hashtags and the keyword. The keywords and hashtags selected to examine the sentiment of Twitter users about the study’s topic were ‘europeansuperleague’,
‘#notoeuropeansuperleague’, ‘#yestoeuropeansuperleague’,
‘#stopeuropeansuperleague’, ‘#supporteuropeansuperleague’. The hashtags were identified using the Twitter ‘explore’ function and were selected to maintain a balance between tweets with a negative and positive inclination towards the ESL, along with the keyword which is considered neutral. Even though the league is widely known as European Super League or ESL for shorter, the hashtag ‘#esl’ was not used because it was observed to contain a lot of tweets unrelated to the topic, namely about other sports competitions that use the same abbreviation.
Based on the principles of GDPR2, the Trustworthy AI guidelines set by the European Commission (lawful, ethical, and robust) were followed in the study (European Commission, 2019). During the whole period of this study the principles of data minimization (working with necessary data only), availability (open to be contacted by Twitter users included in the sample), integrity (maintenance of the data to remain
“intact, complete, and up-to-date”) (p. 4), confidentiality (data accessed only by the researcher), unlikability (use of data only in the context of the specific research purpose it was initially intended), transparency (openness about what data were used and with what methods), and intervenability (data subjects reserve the right to ask for a change in the status of their data, at any time) (Sanchez et al., 2021).
2 General Data Protection Regulation, issued by the European Union as a law regulating privacy and data protection in the European Economic Area
16 Data cleaning
First, a check was conducted to make sure all the tweets were in English dropping all the cases that were spotted to contain text in another language. Then, the data were preprocessed in Python following a series of steps. The tweets’ text letters were turned to lowercase, URLs and HTML were removed, as well as punctuations, stopwords, and the 10 most frequent and rare words. Next, a spell check was
conducted and the emojis included in the text were removed. Finally, the tweets’ text was lemmatized. Beautiful Soup and NLTK packages were used during these steps.
Data Analysis
After the data cleaning process was completed, the dataset included 455089 tweets, of which 200538 were retweets. A decision was made to keep the retweets in the sample, to study the dissemination of tweets based on whether a tweet is retweeted or not, and its number of shares. The Twitter data used in sentiment/emotion analyses contain user-generated texts, mainly of subjective and informal character (Paltoglou and Thelwall, 2012), as well as hashtags and references. Emotion analysis was conducted using the NRC Word-Emotion Association Lexicon (Mohammad and Turney, 2013).
This lexicon uses Macquarie Thesaurus to classify words as positive or negative, and assign a distinct emotion or sentiment to each word. The lexicon contains 2312 words for positive sentiment, 3324 words for negative sentiment, 839 words for anticipation, 1058 words for disgust, 1476 words for fear, 689 words for joy, 1191 words for sadness, 534 words for surprise, and 1231 words for trust. The use of this lexicon was to decided to analyze the emotions based on Plutchik’s emotion model of eight basic and prototypical emotions in four pairs (Yu & Wang, 2015), namely, joy–sadness, trust–disgust, fear–anger, and surprise–anticipation. The eight basic emotions are
17 paired, based on the physiological reaction of each, and their opposite polarity
(Plutchik, 1980) The analyses of the data were conducted using IBM SPSS Statistics 27 and Python 3.8.8 in Jupyter Lab.
Results
For H1, an independent samples t-test was conducted on SPSS to test if negative sentiment is more prevalent than positive in the dataset. To run the test, first, the data were transformed from a wide to a long format, essentially doubling the dataset cases.
This change was carried out to compare the two independent variables on the DV volume, which was created in this process. As shown on Table 3 The result positive sentiment is present significantly more (M = .19, SD = .24) compared to negative sentiment (M = .11, SD = .18). The effect is considered weak to moderate, t (909844)
= -192.72, p < .001, 95% CI [-0.08, -0.09], d = .38. Thus, H1 that negative sentiment appears more than positive sentiment is rejected.
A logistic regression model was used to test H2a thatTweets that contain negative emotional language are retweeted significantly more than tweets that contain positive emotional language, with the dependent variable (DV) being whether a tweet is retweeted or not. According to Nagelkerke’s (pseudo-)R2, 26% of the variance around the DV is explained by the IVs. The dataset includes human behavior data, which are expected to explain less of the variance, so the pseudo-R2 figure suggests no reason for concern. The result of the test indicated that holding the negative sentiment variable constant, the odds of retweeting a tweet increased by 129% for one unit increase in positive sentiment. Holding the positive variable constant, the odds of retweeting a tweet decreased by 34.5% for a one-unit increase in negative sentiment.
18 A preliminary analysis showed that the assumption of multicollinearity was met (VIF
< 10, tolerance > .2). This hypothesis is rejected as a result.
Table 1
A logistic regression model including positive and negative emotions as IVs and whether a tweet is retweeted or not as the DV
Independent Variables b* SE p Odds
Constant .75 .01 < .001 2.12
Positive sentiment 1.29 .02 < .001 3.62
Negative sentiment
-.35 .02 < .001 .71
Model χ2 = 8199.13 p < .001
Nagelkerke (pseudo-)R2 .26
n = 455089
H2b hypothesizing that the use of negative emotional language leads to a higher number of retweets compared to positive language was tested using multiple linear regression. The model is significant F (2, 455086) = 3856.33, p < .001, and included retweet count as the DV and positive and negative sentiment as the IVs. Thus, the model can predict the number of times a tweet is retweeted, but with very low strength, since only 2% of the variation in retweet count can be predicted based on
19 positive and negative sentiment (R2 =.02). Positive sentiment, b* = .13, t = 89.50, p <
.001, 95% CI [1811.63, 1892.76] has a significant, moderately strong association with retweet count, whereas negative sentiment, b* = -.06, t = -50.30, p < .001, 95% CI [- 1441.65, -1333.51] has a weak association with retweet count. Summing up, for every additional point in the scale of positive sentiment, a tweet is retweeted 1852.19 times more, while for every additional point in the scale of negative sentiment a tweet is retweeted 1387.58 times less, keeping the other IVs constant each time. Accordingly, this hypothesis is rejected as well.
Table 2
Multiple regression model predicting retweet count with positive and negative sentiment as IVs
Retweet count Constant .68
Positive -.03 Negative .12
R2 0.02
F(2, 455086) 3856.33 n = 455089, p <.001
To test H3, a series of independent samples t-tests were conducted, between every emotion and anger, after transforming the data from wide format to long, with the count of the variables as the DV. All the t-tests conducted between anger and the other emotions were significant.
20 As can be seen in Table 3 and according to the results, trust (M=.12, SD= .18) is the most prevalent emotion in the dataset, followed by anticipation (M=.09, SD= .17), while the other emotions scored much lower. Based on these results, hypothesis H3 is rejected.
Table 3
Barchart depicting the NRC Lexicon scores of distinct emotions and sentiment in the dataset
Table 4
Means, standard deviations, and sample size of the emotions and sentiment expressed in tweets
Emotion
Mean
Standard Deviation
n
Anger .04 .09 455089
Anticipation .09 .17 455089
Disgust .03 .07 455089
Fear .05 .11 455089
21
Joy .05 .09 455089
Sadness .04 .08 455089
Surprise .03 .08 455089
Trust .12 .18 455089
Positive .19 .24 455089
Negative .11 .18 455089
Logistic regression was used to test H4a. Whether a tweet was retweeted or not was used as DV and the emotions anger, anticipation, disgust, fear, joy, sadness, surprise, trust as IVs. The odds of retweeting a tweet increased by 75% for one unit increase in anger, 24% for one unit increase in anticipation, 24% for one unit increase in fear, 54% for one unit increase in joy, and 186% for one unit increase in trust. The odds of retweeting a tweet decreased by 156% for one unit increase in disgust, 46% for one unit increase in sadness, 65% for one unit increase in surprise. The results were generated holding the other predictor variables constant. A preliminary analysis showed that the assumption of multicollinearity was met (VIF < 10, tolerance > .2).
H4a that tweets that contain anger emotional language, are retweeted more than tweets that contain any other emotional language (eg anticipation, surprise, joy, trust). is rejected.
Table 5
Logistic regression model including anger, anticipation, disgust, fear, joy, sadness, surprise, and trust emotions as IVs and whether a tweet is retweeted or not as DV
22
Independent Variables b* SE p Odds
Constant .72 .01 < .001 2.06
Anger 1.29 .02 < .001 2.11
Anticipation -.35 .02 < .001 1.27
Disgust .24 .05 < .001 .21
Fear -1.56 .03 < .001 1.71
Joy .24 .04 < .001 1.27
Sadness -.46 .04 < .001 .63
Surprise -.65 .04 < .001 .52
Trust 1.86 .02 < .001 6.40
Model χ2 = 10673.18 p < .001
Nagelkerke (pseudo-)R2 .03
n = 455089
Table 6
How much a tweet is retweeted or not depending on the existence of a particular sentiment or emotion
23 Multiple linear regression was used to test whether the use of anger emotional
language leads to a higher number of retweets compared to the use of other emotional language, for H4b. The model is significant F (8, 455088) = 1270.15, p < .001, and can predict the number of times a tweet is retweeted based on the use of each of the eight emotions, with very low strength, since only 2.2% of the variation in retweet count can be predicted based on anger, anticipation, disgust, fear, joy, sadness, surprise and trust (R2 =.02). This figure is considered worryingly low to negligible and demotes the predictive value of the model.
Anger, b* = .03, t = 18.69, p < .001, 95% CI [0.15, 0.18], anticipation, b* = .02, t = 13.75, p < .001, 95% CI [0.05, 0.06], disgust, b* = -.05, t = -31.07, p < .001, 95% CI [-0.36, -0.32], fear, b* = .03, t = 17.70, p < .001, 95% CI [0.10, 0.12], joy, b* = .02, t
= 10.65, p < .001, 95% CI [0.07, 0.10], fear, b* = .03, t = 17.70, p < .001, 95% CI [0.10, 0.12], sadness, b* = -.02, t = -10.90, p < .001, 95% CI -[0.11, -0.08], surprise, b* = -.03, t = -17.20, p < .001, 95% CI [-0.16, -0.12], and trust, b* = -.13, t = 82.54, p
< .001, 95% CI [0.31, 0.32], have a significant, weak association with retweet count.
24 Keeping the other variables constant, for every additional point in the scale of anger emotional language, a tweet is retweeted .16 times more, for anticipation emotional language, .05 times more, for fear emotional language .11 times more, for joy
emotional language .08 times more, and trust emotional language, .31 times more. For every additional point in the scale of disgust emotional language, a tweet is retweeted .34 times less, for sadness emotional language .09 times less, and for surprise
emotional language, .14 times less. For every one of these predictors' effects, the other IVs are held each time constant. In conclusion, this hypothesis has to be rejected as well.
Table 7
Regression model predicting retweet count with anger, anticipation, disgust, fear, joy, sadness, surprise, trust as IVs
Retweet count
Constant .67
Anger .03
Anticipation .02
Disgust -.05
Fear .03
Joy .02
Sadness -.02
Surprise -.03
Trust .13
R2 0.02
25 F(8, 455080) 1270.15
n = 455089, p <.001
To test H5a, b, and c, a dataset containing only the tweets posted after the
announcement of the suspension of ESL on 21/04/2021, (n=27409) was used. The data were transformed from wide to long format resulting in a dataset of n = 219272.
Three independent samples t-tests were used to test whether emotional language associated with positive emotions in Plutchik’s wheel of emotions is more prevalent than their negative pair. Anticipation (M = .10, SD = .18) is present significantly more than surprise (M = .03, SD = .09) and a medium effect exists, t (54816) = 56.640, p <
.001, 95% CI [0.07, 0.07], d = .49. Thus, hypothesis H5a was supported by the data.
Regarding H5b, joy (M = .05, SD = .09) is not present significantly more than sadness (M = .03, SD = .10). The effect is zero, t (54816) = -.021, p = .98, 95% CI [- 0.00, 0.00], d = 0. Therefore, this hypothesis is rejected. For H5c, trust (M = .11, SD = .17) is present significantly more than disgust (M = .03, SD = .08). The effect is medium, t (54816) = 76.41, p = <. 001, 95% CI [0.08, 0.09], d = 0.602168. H5c is also supported by the data.
Table 6
Means, standard deviations, and sample size of the emotions expressed in tweets in the dataset after the suspension of the ESL
Emotion Mean Standard
Deviation
n
26
Anger .05 .08 219272
Anticipation .10 .17 219272
Disgust .03 .08 219272
Fear .06 .11 219272
Joy .05 .09 219272
Sadness .05 .10 219272
Surprise .03 .09 219272
Trust .11 .17 219272
Table 7
Barchart depicting the NRC Lexicon scores of distinct emotions and sentiment in the dataset after the suspension of the ESL
For H6, a new dataset was created, containing only the tweets of the four days when the vast majority of the tweets’ initial dataset, are posted. The dataset (n = 439985) was transformed from wide to long format resulting to a dataset of n = 879970.
27 Positive sentiment (M = .19, SD = .24) appears significantly more in the dataset than negative sentiment (M = .11, SD = .18). The reported effect is low to medium, t (879968) = 139.50, p = < .001, 95% CI [0.09, 0.09], d = 0.38. Therefore H6 is rejected.
Discussion
The ESL project failed spectacularly, in just three days, despite its huge economic backing. In the present study, the online aspect of this failure was examined, by looking into the emotional language expressed by Twitter users, and the way it contributed to the dissemination of the tweets through retweets. An extensive
literature review was conducted, looking into the topics of emotion expression in SM, the use of big data in SM research, eWOM, and the expression of sentiment/emotions on SM. Furthermore, the role of emotional language in the diffusion and the virality of content on Twitter and other SM were studied, along with the online expression of emotions after a crisis, and the spikes in the creation of content.
This study attempts to extend previous research like the ones of Yu and Wang (2015) about the use of emotional language during the 2014 World Cup, Lucas et al. (2017) about users’ emotional reactions over time on Twitter, and Morgan and Wilk (2021) who studied the expression of sentiment in a user-generated crisis on Twitter.
Previous studies conducted regarding the intersection of sports, Twitter, and the analysis of sentiment/emotion, refer mainly to reactions of users to sports events or incidents. In this case, the focus of the study was a failed attempt to fundamentally change the architecture of European football, and not a sporting event reaction per se,
28 connotating societal and political associations besides the athletic part, as explained in the initial parts of the study.
For the data analysis part, a decision was made to retain the cases that are retweets.
First of all, retweets constitute nearly half of the cases. The study aimed to analyze the data and the opinions of Twitter users as unfiltered as possible since a retweet is also an important part of user expression on Twitter. Finally, retweets are considered valuable in indicating the dissemination of content in the platform, while one single retweet can potentially reach on average a thousand Twitter users (Kwak et al., 2010).
Furthermore, answering the call from a previous study (Vermeer and Araujo, 2020), retweets were chosen to be studied as the main instrument that contributes to
information diffusion, disregarding mentions and replies.
Regarding the results part of the study, it can be argued that the analysis of the data led to odd conclusions compared to the relevant academic research, as well as towards the public sentiment, and the conclusion of the ESL case. Out of the ten hypotheses proposed, eight are rejected. Positive emotional language is present significantly more than negative, predicts retweetability of a tweet and retweet count in the period under consideration. In terms of emotional language which includes distinct emotions, anger is evidently not the emotion that appears more than any other in the data, nor the emotion that leads to higher retweetability and/or higher number of retweet count, as hypothesized. In like manner, in the four days that the overwhelming majority of the tweets are posted, positive emotional language, dominates in the data. Two out of the three hypotheses that emotional language containing positive distinct emotions more than their negative counterparts, were supported by the data. In this case, though, the
29 emotional language in the period after the suspension did not change from negative (before the suspension) to positive (after the suspension), as was suggested by past research (Morgan and Wilk, 2021).
To answer the research questions and subquestions posed at the beginning of this study, the use of emotional language in this dataset, cannot explain the swift failure of the ESL, based on the online reactions of the users. If anything, the results of this study, suggest that the emotional language expressed online about ESL was highly positive. Trust and anticipation appear as the most prevalent distinct emotions, as well as in their effect in the dissemination of tweets through retweeting. The results are - partially- in line with the results of Yu and Wang (2015), who studied a relevant topic (2014 FIFA World Cup tweets), finding that joy and anticipation were more prevalent than negative emotional language. Still, the context of this study differs from the present, in the sense that the creation of the ESL was considered to be an incident perceived in a unilaterally negative way. Another study by Lucas et al. (2017) that looked into the same topic, found that a high percentage of tweets was associated with negative emotional language.
If the results of the present analysis are accurate, this would mean that at least for Twitter users, the creation of the ESL was a positive development. celebrated by the majority of the fans in Twittersphere. Likewise, the tweets that positively referred to ESL are shared more than tweets that include more negative emotional language, both in terms of retweetability and total number of shares, reaching a larger number of people.
30 The reason for the inconsistency between the current data and earlier analyses
possibly lies in the use of the NRCLexicon. Even though the lexicon consists of more negative than positive words (so can recognize more negative words than positive words), descriptives showed that 60% of the tweets were found to have zero negative polarity, while only 35% of the tweets were found with zero positive polarity. This big difference indicates that a bias may exist, regarding negative emotional language.
A qualitative examination of certain tweets confirmed it. An example is a tweet where a Twitter user expresses his/her frustration with the president of Arsenal FC, by writing “Kroenke out”. This tweet is not classified as negative, because the lexicon cannot recognize its language as such.
Implications
In terms of implications, this study contributes to the field of corporate understanding of fans’ emotions towards a sports entity; in this case, football fans and sports
organizations. The methods of big data analytics offer a cost- and time-effective method of analyzing thousands or millions of data, using automated algorithms (Mulholland et al., 2015; Nandwani and Verma, 2021; Philander & Zhong, 2016).
This can be an extremely useful way, to understand what the consumers feel towards, for example, a brand, and identify any sudden changes in the online-expressed
sentiment. Besides, the results of the current study, suggest that the use of positive emotional language is a more effective way to have your content shared more, and reach a higher number of people.
Likewise, for the methodological part, this is one of the few identified studies examining emotional language that includes distinct emotion use, in the context of
31 sports in general. Emotion diffusion analysis constitutes an intriguing choice for organizations in various industries, to understand the sentiment of their clients at any given time, and to examine how the use of particular emotional language impacts their posts. More research should be devoted to this direction for academic and
professional purposes since it can be still considered a promising area that offers the capability to study many different topics and platforms.
Limitations and Future Research
Although this study contributes to the largely unexplored field of SM users’ emotion expression classification, and message dissemination around football, certain
limitations exist as well. First of all, the study is subject to commonly applied
limitations of other studies. SM users and online discussions are not representative of the offline world citizens and dialogue (Yarch et al., 2021). Thus, the results should not be generalized beyond the context of SM. Finally, the results are based on information that may be changing constantly, so the analysis results represent the opinions of Twitter users during the period under study (Gong et al., 2021; Philander
& Zhong, 2016).
Furthermore, as mentioned above, the analysis of emotions is not based on a sports lexicon. Tweets may contain slang or words that represent more than one emotion (eg.
through sarcasm) (Sailunaz et al., 2018). Sentiment/emotion analysis is considered an extremely domain-dependent method (Thelwall et al., 2012), and the use of a corpus designed specifically for football discussions in SM is suggested for future studies.
Possible solutions to the drawbacks of the present lexicon include either the use of a football-specific lexicon or more advanced packages that use machine learning to
32 combine “inductive pattern extraction with deductive classification” (Yarch et al., 2021, p. 115). Still, despite the issues of task and genre/domain specificity mentioned across the literature (González-Bailón & Paltoglou, 2015; Thelwall et al., 2012), lexicon-based tools have been found to classify the polarity of tweets with very high accuracy (Wunderlich & Memmert, 2020), that may even exceed the effectiveness of supervised models (Paltoglou & Thelwall, 2012).
In addition, a journalist investigation revealed that executives of the clubs backing the ESL project, bought thousands of bots to generate positive messages and ameliorate the online perception of public opinion towards the project (Reuters, 2021). This would mean that a distorted image of reality was recorded, and the emotion on Twitter was more negative than what was depicted in this study. Filtering out bot accounts from a dataset is considered a difficult task, and a potential solution is deleting all the tweets two standard deviations away from the mean of the dataset’s users (Coskun & Ozturan, 2018). In this study, the presence of bots is not considered necessarily a limitation of the study, since bots can be regarded as part of the online discourse. However, their existence cannot be overlooked. A future study could focus on the identification of bot accounts and tweets, associated with sports-related
trending topics. In conclusion, even though the dissemination of tweets was one of the subjects looked into in this study, the role of influencers and super-diffusers of
content was not analyzed. Future studies could examine the association of these users with emotional language, and the impact they have in the dissemination of content and even in opinion-making.
33 Despite its limitations, this study still offers useful insights into a virtually unexplored issue such as the ESL, while it also contributes to the exploration of sports-related, emotional expression by SM users, and the effect of emotional language on the virality and diffusion of tweets. The study builds on previous knowledge, and further advances it, by reporting its set of results and highlighting its caveats, informing future researchers about potential existing pitfalls. New studies are required to further advance the current knowledge, without disregarding interesting, unexplored areas of research.
34
Acknowledgments
First of all, I would like to express my gratitude to my supervisor, dr. Christel van Eck. You have been helpful and considerate during the -often- frustrating process that writing this thesis has been. Your valuable feedback helped me write a much better thesis than I thought I could have, and improved me much as a researcher, overall. To dr. Theo Araujo, a big thanks for sparking my interest to use data analytics in my study, and for your help and feedback on the most technical parts. Concluding, to my family and friends, both in the Netherlands and in Greece, thank you for being there in numerous ways, during this 1.5 year of studying at the University of Amsterdam.
Your support meant the world to me.
35
References
Abosag, I., Roper, S., & Hind, D. (2012). Examining the relationship between brand emotion and brand extension among supporters of professional football clubs.
European Journal of Marketing, 46(9), 1233-1251.
http://dx.doi.org/10.1108/03090561211247810
Aloufi, S., & Saddik, A. E. (2018). Sentiment Identification in Football-Specific Tweets. IEEE Access, 6, 78609–78621.
https://doi.org/10.1109/ACCESS.2018.2885117
Anagnostopoulos, C., Parganas, P., Chadwick, S., & Fenton, A. (2018). Branding in pictures: using Instagram as a brand management tool in professional team sport organisations. European Sport Management Quarterly, 18(4), 413–438.
https://doi.org/10.1080/16184742.2017.1410202
Araujo, T., Neijens, P., & Vliegenthart, R. (2015). What motivates consumers to re- tweet brand content?: The impact of information, emotion, and traceability on pass-along behavior. Journal of Advertising Research, 55(3), 284–295.
https://doi.org/10.2501/JAR-2015-009
Bandhakavi, A., Wiratunga, N., Massie, S., & P, D. (2021). Emotion‐aware polarity lexicons for Twitter sentiment analysis. Expert Systems, 38(7).
https://doi.org/10.1111/exsy.12332
Berger, J., & Milkman, K. L. (2012). What Makes Online Content Viral? Journal of Marketing Research, 49(2), 192–205. https://doi.org/10.1509/jmr.10.0353 Bhatia, S., Mellers, B., & Walasek, L. (2019). Affective responses to uncertain real-
world outcomes: Sentiment change on Twitter. PloS One, 14(2), e0212489–
e0212489. https://doi.org/10.1371/journal.pone.0212489
36 Bomfim, R., & Furtado, V. (2017). Learning the sentiment of soccer fans from data
on bets and social nets. Web Intelligence, 15(2), 101–113.
https://doi.org/10.3233/WEB-170355
Coombs, W. T. (2007). Protecting Organization Reputations During a Crisis: The Development and Application of Situational Crisis Communication Theory.
Corporate Reputation Review, 10(3), 163–176.
https://doi.org/10.1057/palgrave.crr.1550049
Coşkun, M., & Ozturan, M. (2018). #europehappinessmap: A Framework for Multi- Lingual Sentiment Analysis via Social Media Big Data (A Twitter Case Study). Information, 9(5). https://doi.org/10.3390/info9050102
Edensor. T. (2015). Producing atmospheres at the match: Fan cultures,
commercialisation and mood management in English football. Emotion, Space and Society, 15, 82–89. https://doi.org/10.1016/j.emospa.2013.12.010
European Commission’s Guidelines for Trustworthy AI.
https://ec.europa.eu/newsroom/dae/document.cfm?doc_id=60419
Ferrara, E. & Yang, Z. (2015). Quantifying the effect of sentiment on information diffusion in social media. PeerJ. Computer Science, 1, e26–.
https://doi.org/10.7717/peerj-cs.26
Gong, H., Watanabe, N. M., Soebbing, B. P., Brown, M. T., & Nagel, M. S. (2021).
Do Consumer Perceptions of Tanking Impact Attendance at National
Basketball Association Games? A Sentiment Analysis Approach. Journal of Sport Management, 35(3), 254-265.
https://journals.humankinetics.com/view/journals/jsm/35/3/article-p254.xml Gruber, D., Smerek, R. E., Thomas-Hunt, M. C., & James, E. H. (2015). The real-
time power of Twitter: Crisis management and leadership in an age of social
37 media. Business Horizons, 58(2), 163–172.
https://doi.org/10.1016/j.bushor.2014.10.006
Guerra, P. C., Meira, W., & Cardie, C. (2014). Sentiment Analysis on Evolving Social Streams: How Self-Report Imbalances Can Help. Proceedings of the 7th ACM International Conference on Web Search and Data Mining, 443–452.
https://doi.org/10.1145/2556195.2556261
Hansen, L. K., Arvidsson, A., Nielsen, F. A., Colleoni, E., & Etter, M. (2011). Good Friends, Bad News—Affect and Virality in Twitter. In J. J. Park, L. T. Yang,
& C. Lee (Eds.), Future Information Technology (pp. 34–43). Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-642-22309-9_5
Hartmann, P.M., Zaki, M., Feldmann, N., & Neely, A. (2016). Capturing value from big data – a taxonomy of data-driven business models used by start-up firms.
International Journal of Operations & Production Management, 36(10), 1382–1406. https://doi.org/10.1108/IJOPM-02-2014-0098
Ibbetson, C. (2021, April 19). Snap poll: football fans overwhelmingly reject European Super League. YouGov.
https://docs.cdn.yougov.com/i9qr8cjrlg/YouGov%20-
%20European%20Super%20League.pdf
Jiménez-Zafra, S.M., Sáez-Castillo, A. J., Conde-Sánchez, A., & Martín-Valdivia, M.
T. (2021). How do sentiments affect virality on Twitter? Royal Society Open Science, 8(4), 201756–201756. https://doi.org/10.1098/rsos.201756
Kim, J., Overton, H., Bhalla, N., & Li, J. (2020). Nike, Colin Kaepernick, and the politicization of sports: Examining perceived organizational motives and public responses. Public Relations Review, 46(2), 101856–.
https://doi.org/10.1016/j.pubrev.2019.101856
38 Kwak, H., Lee, C., Park, H., & Moon, S. (2010). What is Twitter, a social network or
a news media? Proceedings of the 19th International Conference on World Wide Web, 591–600. https://doi.org/10.1145/1772690.1772751
Lawrence, S., & Crawford, G. (2018). The hyperdigitalization of football cultures (pp.
1–16). https://doi.org/10.4324/9781351118903-1
Liew, J. (2021, April 23). The furore over the proposed European Super League was about one question: who runs our lives? New Statesman, 150(5616), 40.
https://link.gale.com/apps/doc/A661498154/AONE?u=amst&sid=bookmark- AONE&xid=bf9c4c1b
Ligthart, A., Catal, C., & Tekinerdogan, B. (2021). Systematic reviews in sentiment analysis: a tertiary study. The Artificial Intelligence Review, 54(7), 4997–
5053. https://doi.org/10.1007/s10462-021-09973-3
Lucas, G. M., Gratch, J., Malandrakis, N., Szablowski, E., Fessler, E., & Nichols, J.
(2017). GOAALLL!: Using sentiment in the world cup to explore theories of emotion. Multimodal Sentiment Analysis and Mining in the Wild Image and Vision Computing, 65, 58–65. https://doi.org/10.1016/j.imavis.2017.01.006 Marcotti, G. (2021, April 21). Super League suspended: Why English clubs pulled
out, and what’s next for them and UEFA. ESPN.
https://www.espn.com/soccer/blog-marcottis-musings/story/4365465/super- league-suspended-why-english-clubs-pulled-outwhats-next-for-them-and-uefa McCarthy, J., Rowley, J., Jane Ashworth, C., & Pioch, E. (2014). Managing brand
presence through social media: the case of UK football clubs. Internet Research, 24(2), 181–204. https://doi.org/10.1108/IntR-08-2012-0154 Mendoza, S.A., Lane, S. P., & Amodio, D. M. (2014). For Members Only: Ingroup
Punishment of Fairness Norm Violations in the Ultimatum Game. Social
39 Psychological & Personality Science, 5(6), 662–670.
https://doi.org/10.1177/1948550614527115
Merkel, U. (2012). Football fans and clubs in Germany: conflicts, crises and compromises. Soccer and Society, 13(3), 359–376.
https://doi.org/10.1080/14660970.2012.655505
Millar, C. (2021, October 19). Super League chiefs change five competition rules after huge backlash first time round. Mirror.
https://www.mirror.co.uk/sport/football/news/european-super-league-football- changes-25253034
Minaee, S., Kalchbrenner, N., Cambria, E., Nikzad, N., Chenaghlu, M., & Gao, J.
(2021). Deep Learning–based Text Classification: A Comprehensive Review.
ACM Computing Surveys, 54(3), 1–40. https://doi.org/10.1145/3439726 Mohammad, S. M. & Turney, P. D. (2013). CROWDSOURCING A WORD-
EMOTION ASSOCIATION LEXICON. Computational Intelligence, 29(3), 436–465. https://doi.org/10.1111/j.1467-8640.2012.00460.x
Morgan, A., & Wilk, V. (2021). Social media users’ crisis response: A lexical exploration of social media content in an international sport crisis. Public Relations Review, 47(4), 102057.
https://doi.org/10.1016/j.pubrev.2021.102057
Naveed, S., Gottron, T., Kunegis, J., & Alhadi, A. (2011). Bad news travel fast: a content-based analysis of interestingness on Twitter. Proceedings of the 3rd International Web Science Conference, 1–7.
https://doi.org/10.1145/2527031.2527052
Nazir, F., Ghazanfar, M. A., Maqsood, M., Aadil, F., Rho, S., & Mehmood, I. (2018).
Social media signal detection using tweets volume, hashtag, and sentiment
40 analysis. Multimedia Tools and Applications, 78(3), 3553–3586.
https://doi.org/10.1007/s11042-018-6437-z
Paltoglou, G. & Thelwall, M. (2012). Twitter, MySpace, Digg: Unsupervised
Sentiment Analysis in Social Media. ACM Transactions on Intelligent Systems and Technology, 3(4), 1–19. https://doi.org/10.1145/2337542.2337551
Philander, K., & Zhong, Y. (2016). Twitter sentiment analysis: Capturing sentiment from integrated resort tweets. International Journal of Hospitality
Management, 55, 16–24. https://doi.org/10.1016/j.ijhm.2016.02.001 Plutchik. R. (1980). Emotion : a psychoevolutionary synthesis. Harper and Row.
Power, M. J., Widdop, P., Parnell, D., Carr, J., & Millar, S. R. (2020). Football and politics: The politics of football. Managing Sport and Leisure, 25(1–2), 1–5.
https://doi.org/10.1080/23750472.2020.1723437
Pozzi, F., Fersini, E., Messina, E., & Liu, B. (2017). Chapter 1 – Challenges of Sentiment Analysis in Social Networks: An Overview. In Sentiment Analysis in Social Networks, pp. 1–11. Elsevier Inc. https://doi.org/10.1016/B978-0-12- 804412-4.00001-2
Pröllochs, N., Bär, D., & Feuerriegel, S. (2021b). Emotions explain differences in the diffusion of true vs. false social media rumors. Scientific Reports.
https://doi.org/10.3929/ethz-b-000513362
Pröllochs, N., Bär, D., & Feuerriegel, S. (2021a). Emotions in online rumor diffusion.
EPJ Data Science, 10(1), 1–17. https://doi.org/10.1140/epjds/s13688-021- 00307-5
Reuters. (2021, November 11). Study: Bots Used to Manipulate Social Media in Favor of European Super League. VOA. Retrieved January 27, 2022, from
41 https://www.voanews.com/a/study-bots-used-to-manipulate-social-media-in- favor-of-european-super-league-/6309046.html
Reynolds, M. (2021, April 20). JPMorgan-backed breakaway group would disrupt soccer’s Champions League. SNL Bank and Thrift Daily. https://advance- lexiscom.proxy.uba.uva.nl/api/document?collection=news&id=urn:contentIte m:62H0-CPK1-JDVW-N50N-00000-00&context=1516831.
Rouvrais-Charron, C., & Durand, C. (2009). European football under close scrutiny.
International Journal of Sports Marketing & Sponsorship, 10(3), 230–243.
https://doi.org/10.1108/IJSMS-10-03-2009-B005
Sailunaz, K., Dhaliwal, M., Rokne, J., & Alhajj, R. (2018). Emotion detection from text and speech: A survey. Social Network Analysis and Mining, 8(1), 28.
https://doi.org/10.1007/s13278-018-0505-2
Sanchez, D., Viejo, A., & Batet, M. (2021). Automatic Assessment of Privacy Policies under the GDPR. Applied Sciences, 11(4), 1762–.
https://doi.org/10.3390/app11041762
Scheffer, M., van de Leemput, I., Weinans, E., & Bollen, J. (2021). The rise and fall of rationality in language. Proceedings of the National Academy of Sciences - PNAS, 118(51), 1–. https://doi.org/10.1073/pnas.2107848118
Shakina, E., Gasparetto, T., & Barajas, A. (2020). Football Fans’ Emotions:
Uncertainty Against Brand Perception. Frontiers in Psychology, 11, 659–659.
https://doi.org/10.3389/fpsyg.2020.00659
Smith, N., & Leiserowitz, A. (2014). The Role of Emotion in Global Warming Policy Support and Opposition. Risk Analysis, 34(5), 937–948.
https://doi.org/10.1111/risa.12140
42 Stieglitz, S., & Dang-Xuan, L. (2013). Emotions and Information Diffusion in Social
Media-Sentiment of Microblogs and Sharing Behavior. Journal of Management Information Systems, 29(4), 217–248.
https://doi.org/10.2753/MIS0742-1222290408
Thelwall, M., Buckley, K., & Paltoglou, G. (2012). Sentiment strength detection for the social web. Journal of the American Society for Information Science and Technology, 63(1), 163–173. https://doi.org/10.1002/asi.21662
Tinson, J., Sinclair, G., & Gordon, R. (2021). How value is disrupted in football fandom, and how fans respond. European Sport Management Quarterly, 1–18.
https://doi.org/10.1080/16184742.2020.1871394
Tirdad, K., Dela Cruz, A., Sadeghian, A., & Cusimano, M. (2021). A deep neural network approach for sentiment analysis of medically related texts: an analysis of tweets related to concussions in sports. Brain Informatics, 8(1), 12–12.
https://doi.org/10.1186/s40708-021-00134-4
Tsugawa, S., & Ohsaki, H. (2017). On the relation between message sentiment and its virality on social media. Social Network Analysis and Mining, 7(1), 1–14.
https://doi.org/10.1007/s13278-017-0439-0
Twitter Developer Platform. (n.d.). Twitter API for Academic Research | Products.
https://developer.twitter.com/en/products/twitter-api/academic-research Vermeer, S. & Araujo, T. (2020). Keep the ball rolling: Information diffusion within
large sports-related networks through social mediators. Communication and Sport, 8(6), 803–824. https://doi.org/10.1177/2167479519841868
Wang, J., & Wei, L. (2020). Fear and Hope, Bitter and Sweet: Emotion Sharing of Cancer Community on Twitter. Social Media + Society, 6(1),
2056305119897319. https://doi.org/10.1177/2056305119897319
43 Wang, X., & Lee, E. W. J. (2021). Negative emotions shape the diffusion of cancer
tweets: Toward an integrated social network–text analytics approach. Internet Research, 31(2), 401–418. https://doi.org/10.1108/INTR-04-2020-0181 Wettach, S. (2021, October 15). Super League: Neustart ohne ständige Mitglieder
geplant. WirtschaftsWoche.
https://www.wiwo.de/unternehmen/dienstleister/offenes-format-ohne- staendige-mitglieder-super-league-plant-neustart/27705960.html
Wunderlich F., & Memmert, D. (2020). Innovative Approaches in Sports Science- Lexicon-Based Sentiment Analysis as a Tool to Analyze Sports-Related Twitter Communication. Applied Sciences, 10(2), 431–.
https://doi.org/10.3390/app10020431
Yarchi, M., Baden, C., & Kligler-Vilenchik, N. (2021). Political Polarization on the Digital Sphere: A Cross-platform, Over-time Analysis of Interactional, Positional, and Affective Polarization on Social Media. Political Communication, 38(1–2), 98–139.
https://doi.org/10.1080/10584609.2020.1785067
Yu, Y., & Wang, X. (2015). World Cup 2014 in the Twitter World: A big data analysis of sentiments in U.S. sports fans’ tweets. Computers in Human Behavior, 48, 392–400. https://doi.org/10.1016/j.chb.2015.01.075
Zhang, J., Wang, Y., Shi, M., & Wang, X. (2021). Factors Driving the Popularity and Virality of COVID-19 Vaccine Discourse on Twitter: Text Mining and Data Visualization Study. JMIR Public Health and Surveillance, 7(12), e32814–
e32814. https://doi.org/10.2196/32814
Zhu, X., Kim, Y., & Park, H. (2020). Do messages spread widely also diffuse fast?
Examining the effects of message characteristics on information diffusion.
44 Computers in Human Behavior, 103, 37–47.
https://doi.org/10.1016/j.chb.2019.09.006
Appendix 1
The following script was used to extract the tweets:
twarc2 search 'europeansuperleague OR #yestoeuropeansuperleague OR
#notoeuropeansuperleague OR #stopeuropeansuperleague OR
#supporteuropeansuperleague lang:en' --start-time 2021-4-18 --end-time 2021-12-03 - -archive eursupleague.jsonl
