Using emojis to stimulate consumer engagement: A content analysis of emoji-use in 10,222 firm-generated tweets.

29  Download (0)

Full text


Using emojis to stimulate consumer engagement: A content analysis of emoji-use in 10,222 firm-generated tweets.

Name: Toos de Zeeuw Student number: 13403419 Date: 25/06/2021

Thesis supervisor: Gijs Overgoor

MSc. in Business Administration – Consumer Marketing University of Amsterdam

EBEC approval: 20210304120348


Table of contents

Acknowledgements 3

Abstract 3

Statement of originality 3

Introduction 4

Literature review 7

Social media marketing and owned social media 7

Consumer engagement in firm-generated content 7

Emojis in firm-generated content 8

Emojis as peripheral and central cues 9

Emojis and sentiment in firm-generated content 12

Research methods 13

Sample 13

Data collection 14

Variable operationalization 14

Data analysis 15

Results 16

Discussion and conclusion 18

References 21

Appendix 26

I. Top 50 Fortune 500 companies (2020) with Twitter accounts 26

II. Syntax used to mine data 27



First of all, and in specific, I would like to thank my thesis supervisor Mr. G. (Gijs) Overgoor.

Without your help and guidance, I would have stayed in my comfort zone and chosen a research method that I was familiar with. Thanks to your enthusiasm and continuous

motivation, I found a new interest in programming and data science. In addition, I would like to thank Dr. J. (Jonne) Guyt for the helpful course of social media research. Thanks to these classes, I was able to quickly learn the basics of the R language and explore all its



The use of emojis is growing rapidly and emojis are now increasingly used in social media marketing. Prior literature illustrated the advantages of emojis in online communication, but little is known about the effects of emojis on user engagement. Therefore, this research studies emoji-use in firm-generated content on Twitter and its effect on the number of favorites and retweets. Using R language, 10,222 tweets from the top 50 Fortune 500

companies were analyzed. The results show no significant effect for the presence of emojis in firm-generated content and the number of favorites. However, the results demonstrated that the use of emojis leads to a lower number of retweets. In addition, it was found that the congruency of the emoji does not influence user engagement. Finally, the findings show that stronger sentiment in tweets containing emojis does not lead to more user engagement.

Nonetheless, the results did demonstrate that strong text sentiment negatively influences user engagement, and that strong emoji sentiment positively influences user engagement. This study contributes to the current literature about the use of emojis in online communication as well as to the social media research field, as this study shows the wide possibilities of using data mining for research purposes.

Statement of originality

This document is written by Toos de Zeeuw [13403419] who declares to take full

responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is

responsible solely for the supervision of completion of the work, not for the contents.



Consider the written utterance: ‘We need to talk’. Are you able to recognize the emotion of the writer? Now examine the difference between ‘We need to talk 😃’ and ‘We need to talk

😡’. Taking into account that in face-to-face situations there exist verbal cues, this example shows that written language can be more difficult to interpret when social cues are lacking.

However, through the use of emojis in written text, the sender is able to clearly display the underlying meaning and emotions which contributes to the understandability of the message (Daniel & Camp, 2018).

Nowadays, communicating does not necessarily require face-to-face situations. Online communication through social media is rapidly growing. In fact, there were 4.14 billion active social media users worldwide in 2020 (Statista, 2021). Not only user-generated content can be found on these platforms, but also organizations have increasingly been making use of social media over the past decade. In 2019, 53% of the organizations in Europe were active on at least one social media channel (Eurostat, 2020). Of this percentage, 86% used their social media account as a platform for brand communication and marketing purposes. However, with a change from offline to online communication also comes a change in language use as speakers have to find new ways to express themselves (McCulloch, 2019). Seeing that visual communication offers alignment in the conversation between the speaker and listener using non-verbal cues, digital communication with a lack of non-verbal cues can be more difficult in terms of understanding. Non-verbal cues are fundamental for message understanding as they help the interlocutors to avoid uncertainty and interruption and contribute to more interaction and conversation effectiveness (Rutter & Stephenson, 1977). Taking into account that digital communication is often asynchronous, there is a need for non-verbal cues

(Alshenqeeti, 2016).

Such non-verbal cues have developed by means of emojis. Where at first an emotion in written language had to be spelled out entirely in order for a reader to understand, one emoji can also be sufficient. In other words, speakers can use emojis as a substitute to

physical gestures (McCulloch, 2019). Thus, it is not surprising that the use of emojis amongst individuals and organizations is rapidly growing. Where in August 2018 14.92% of the tweets contained emojis, in April 2020 this number was 19.04% (Broni, 2020). In addition, another analysis revealed that of a sample of 34,121 Instagram business profiles, 47.7% used emojis in their Instagram posts (Statista, 2019).


Brand communication through social media is a way for organizations to drive sales, build their image, and develop relationships with their customers (Kumar et al., 2016).

Through firm-generated content, organizations can engage with consumers. By enhancing customer engagement using social media, organizations can develop a sustainable competitive advantage (Lin, 2020). This social media engagement is manifested by consumers by means of ‘liking’, ‘sharing’, and ‘commenting’ (Sherriff, Zulkifli, & Othman, 2018).

In order to enhance customer engagement, good copywriting is of great importance in firm-generated content (Sherriff, Zulkifli, & Othman, 2018). This is necessary to capture the attention of the audience, to persuade and to communicate the intended meaning. However, text is not always sufficient to capture the attention of social media users. Using eye-tracking, Vraga et al. (2016) found that social media posts with only text received significantly less attention compared to content containing visuals or links. In addition, words are not always adequate to convey the intended meaning and to explain abstract concepts (Stoner, 2009).

Besides the fact that the text needs to be adequate and well-written, organizations furthermore need to compete for the consumers’ attention as only 1.85% of the followers will see the social media post and solely 0.1% of the audience will engage (Sullivan, 2014). Taking into account that humans have an average attention span of 8 seconds (Microsoft, 2015), it is extremely important to create content that grabs the attention.

According to various business sources, emojis are an effective way to attract attention and stimulate engagement (Bullis, 2018; Patel, n.d.). In addition, several studies found that emojis contribute to a better understanding of the message (Aldunate & González-Ibáñez, 2016; Das et al., 2019; Daniel & Camp, 2018). Furthermore, emojis provide organizations with the opportunity to include emotions and display underlying meanings in their social media posts (Mathews & Lee, 2018). They found that the main objective for organizations to use emojis as a marketing tool was to increase customer engagement. However, little research has investigated the impact of emojis on consumer engagement.

Following previous literature (Aldunate & González-Ibáñez, 2016; Das et al., 2019;

Daniel & Camp, 2018), it is clear that emojis are extremely helpful for message understanding and are an increasingly popular way to replace non-verbal cues. However, as emojis in

marketing are a fairly new phenomenon, little research has been done to investigate how they should be used. This is important, as there are various ways in which emojis are currently used. For example, emojis could be added in addition to the text or they can be used to replace a word or even a sentence. Next to that, users can add only one emoji or a whole string of emojis (Moussa, 2019). Furthermore, they can be distributed over different sections of the


content such as the body copy and headline (Karlson, 2017) and the emojis can be congruent or incongruent (Daniel & Camp, 2018).

In short, there are numerous possibilities for organizations to implement emojis in their social media posts. Emojis are still rather new within the field of firm-generated content as well as in marketing research. Therefore, we build on a theory within online brand

communication to investigate the role of emojis in social media to attract attention and stimulate consumer engagement, and how they are currently used in firm-generated content.

This is important for organizations, as the use of emojis also comes at a risk of

misinterpretation (Mathews & Lee, 2018) and requires careful thought. Incorporating emojis in corporate social media posts demands an understanding of the possible meanings and interpretations each emoji could convey among different consumers. In addition, the creator has to possess some creativity in what ways emojis can be incorporated into firm-generated content seeing that there are currently over 3,000 emojis (Unicode, 2020).

To date, there is no clear understanding whether the use of emojis in social media posts indeed influences consumer engagement and what factors of emoji-use contribute to this effect. For this reason, the primary goal of this research is to shed a light on the influence of emoji-use on consumer engagement of firm-generated social media content. Moreover, this research will focus on the factors that influence the relationship between emoji-use and user engagement. Thus, the following research question has been formulated: ‘To what extent does the use of emojis in firm-generated content on Twitter influence online consumer


This research will contribute insights to the marketing research domain about the use of emojis in firm-generated content. Through data mining and text analysis, this study will provide an overview of how emojis are currently used by real organizations on Twitter and how the use of emojis affects consumer engagement. This research technique is unfamiliar and underexplored in the field of social media marketing research despite its potential. Thus, this study will contribute to innovative methods of examining consumer behavior on social media. In addition, considering emojis can be used as a global language and as a way to incorporate nonverbal communication in written language, more knowledge is required to adopt this effectively. With the results of this study, organizations can incorporate emojis in their firm-generated content more successfully in order to stimulate engagement.


Literature review

Social media marketing and owned social media

The arrival of information and communication technologies and digital media has caused an evolution in consumer behavior (Rana et al., 2020). In particular social media provided consumers with the opportunity to connect to others and empowered them to share their opinions (Sharma & Verma, 2018). As a matter of fact, it was found that one in five tweets posted by consumers consist of a mention about a brand (Jansen et al., 2009). Considering the impact of digital media on consumer behavior, social media marketing is a growing field within marketing. Where in 2009 only 3.5% of the marketing budget was spend on social media, this number was 23.2% in 2020 (CMO Survey, 2021).

Not only did this shift in consumer behavior come with challenges for organizations, but it also provided new business opportunities. Through owned social media, organizations are able to connect with their customers and display their products using their own brand’s social media channels (Colicev et al., 2018). In fact, research has shown that owned social media contributes positively to brand purchases (Thornhill, Xia & Lee, 2017). Consequently, social media marketing through owned media is an important aspect of an organization’s marketing strategy.

Consumer engagement in firm-generated content

The use of an organization’s social media channel for marketing communication purposes can be described as firm-generated content (Kumar et al., 2016). In their study, they have shown that firm-generated content is a method to strengthen the relationship with customers and that organizations can benefit from this content by means of higher revenues and profits. They discovered that firm-generated content enhances both online and offline purchases and that it provides synergy across various marketing communication channels, such as e-mail

marketing and television marketing. In addition to improving revenues and building customer relationships, firm-generated content is also beneficial for the organization’s performance in a way that it contributes to value creation and reputation management (Bai & Yan, 2020).

When creating firm-generated content, one of the main aims is to generate customer engagement (Lee, Hosanagar & Nair, 2014). That is, the extent to which customers on social media like, share, and comment on the posts generated by an organization (Sherriff et al., 2018). Social media engagement is an important measure for organizations as it helps to stimulate interest and reach even more users (Hoffman, 2010). Moreover, consumer

engagement affects the retention, acquisition, and termination of customers (Malthouse et al,


2013). Social media engagement generates commitment amongst consumers and improves their loyalty (Hoffman, 2010). Hoffman (2010) stresses that even though in social media marketing much emphasis is placed on monetary returns, organizations should also focus on other returns such as consumer engagement. Even if a consumer does not convert after the social media post, the consumer can spread the content when engaging, which might trigger actions of other consumers. Furthermore, the consumer could convert in the future due to the long-term relationship that was established with the organization.

Nevertheless, nowadays it is extremely difficult to reach the firm’s audience without paid content (Kumar et al., 2016). Considering only a very small number of followers will organically see an organization’s social media post, the design is of great importance to achieve engagement and a larger reach (Lee et al., 2014). Besides, the content that manages to reach the consumer’s timeline does not provide a guarantee for organizations that their

content will be seen by the audience (Berger, 2013). As there is now an abundance of messages and updates on social media, some content goes unnoticed. Therefore, it is fundamental for organizations to create engaging content.

Emojis in firm-generated content

In order to attract attention and increase consumer engagement on social media, organizations are increasingly using emojis in their firm-generated content. Literally translated from

Japanese, emojis are picture (‘e’) characters (‘moji’). Whereas emoticons can only display emotions using the characters of a keyboard (e.g. < 3), emojis provide more ways to express an emotion or object due to the different shapes and colors, such as 💕 or 💛 (Grannan, 2016).

The popularity of these picture characters has grown exponentially: Whereas there only existed 76 emojis in 1995, there currently exist 3,292 emojis in version 13.0 (Unicode, 2020).

It is estimated that over 5 billion emojis are sent each day on Facebook and in Facebook Messenger (Statista, 2020). Considering the great popularity and universal understanding from all over the globe, emojis are often perceived as the language of the future (Das et al., 2017).

The use of emojis offers several contributions to online brand communication. Firstly, emojis can provide additional information about emotions and underlying meanings, such as when expressing sarcasm. In addition, they contribute to a larger comprehension of the message (Aldunate & González-Ibáńez, 2016). Furthermore, Huang et al. (2008) found that the use of emoticons in instant messaging improved communication and helped users to demonstrate their feelings, especially when expressing sarcasm. Moreover, users perceived


the use of emoticons as more fun, more interactive, and more aesthetically pleasant.

Additionally, emoticons had a positive effect on the communication process. Participants that were using emoticons responded faster, leading to a higher interaction. Even though this research focused on emoticons in instant messaging, these finding could indicate that social media users would also perceive social media posts containing emojis as more enjoyable and engaging. The positive effect of emoticon use in communication is supported by an fMRI study by Yuasa et al. (2011). This research demonstrated that emoticons activated the same parts of the brain as nonverbal communication, indicating that emoticons are able to replace nonverbal communication.

Secondly, emojis contribute to the effectiveness of the message. For example, Yakin and Eru (2015) studied the effectiveness of emoji-use in social marketing advertisements.

They focused on nonprofit organizations and discovered that using emojis in social

advertising campaigns are effective, as the participants perceived advertisements containing emojis as more interesting, creative and innovative. In addition, participants perceived the advertisement containing emojis as easier to understand compared to the advertisement without. Furthermore, Das et al. (2019) investigated the effect of emoji-use in advertisements.

They found that using emojis in advertising leads to positive affect and higher purchase intentions. Moreover, they found that using emojis is more effective for hedonic products compared to utilitarian products. Despite that this research contributed interesting findings in the domain of using emojis in advertising, the study only focused on using emojis as an image in advertising rather than as complementor for text-based communication. Additionally, non- academic articles mention the effectiveness of using emojis in firm-generated content. As an illustration, WordStream (2018) compared two tweets, of which one contained emojis and the other did not. The tweet containing emojis received a 25.4% higher engagement compared to the tweet without. Though this is not scientifically proven, it does raise the question to what extent emojis influence consumer engagement.

Emojis as peripheral and central cues

The idea that emojis can both attract attention and provide deeper understanding of the message could be explained by means of the Elaboration Likelihood Model (Petty &

Cacioppo, 1981). The Elaboration Likelihood Model is a theory focusing on attitude change that tries to understand the processes that influence the effectiveness of persuasive messages (Petty & Cacioppo, 1986). Following the Elaboration Likelihood Model, there are two ways of persuasion; central and peripheral (Petty & Cacioppo, 1981). These routes are determined


by the person’s motivation and ability to process a certain message. When a person is highly motivated and able to interpret a message, this person will look carefully and thoughtfully at the message. In this case, the person is interested to process the information deeply and focuses on the quality of the arguments. However, when a person is less motivated to process the message, the peripheral route is used. As the person is less inclined or able to process the message, this person will favor simple cues, such as visual aids or celebrity endorsement (Shin et al., 2020). These cues require less cognitive effort and information processing.

As digital technologies allow every user to publish content online, the amount of available information has increased, and the credibility of sources became more difficult to assess. This makes it more difficult for users to process information deliberately (Sundar, 2009) and makes users pay less attention (Shin et al, 2020). In this case, considering the research by Huang et al. (2008), emojis can help attract attention and facilitate the consumer to process the message more quickly, more fun and more easily using the peripheral route. In fact, Ray & Merle (2020) investigated if the use of emojis in restaurant inspection reports would influence the feelings of the reader. Based on the Elaboration Likelihood Model (Petty

& Cacioppo, 1981), they expected that emojis could influence participants that were less involved. They compared a disgust emoji to a smiling emoji and found that the disgust emoji led to increased perceptions of risk and avoidance tendencies. The smiling emoji improved the idea that the restaurant had high cleaning standards. The results demonstrated that

participants who were less involved were more influenced by the emoji and perceived this as a peripheral cue.

In addition, Shin et al. (2020) studied the effects of content features of social media posts on consumer engagement. They argued that due to the information overload on social media, visual peripheral cues could enhance the persuasiveness of the message, as peripheral cues can attract attention and help process the message more easily. In contrast, visual cues can also enhance the central route, as this contributes to the processing of the information within the content. When comparing simple visual peripheral cues to a high message complexity, it was found that visual peripheral cues contribute more positively to consumer engagement. Even though the study by Shin et al. (2020) did not include emojis as visual peripheral cues, this is an interesting starting point to study the effects of emojis as peripheral and central cues. On the one hand, emojis can attract attention as a peripheral cue. However, on the other hand, they can help the reader to understand the underlying and additional meaning and helps them to process the message more elaborately. Therefore, the first hypothesis is formulated as:


• H1: Tweets containing emojis lead to higher user engagement compared to tweets without emojis.

Focusing on the understandability, believability, and shareability of tweets containing emojis, Daniel & Camp (2018) studied the effects of emojis on the processing fluency of consumers. They created three conditions, using congruent, incongruent, and neutral emojis.

The findings of this study demonstrated that congruent emojis were perceived as more understandable and more believable compared to when no emoji was used. In addition, incongruent emojis were perceived to be less understandable or believable than the tweets without an emoji. Moreover, congruent emojis were found to cause a higher intention to share the message compared to using incongruent or no emojis. Another study by Hogenboom et al.

(2013) found that emoticons help people to demonstrate and emphasize the sentiment of a text, making the contents of a message more accurate. Additionally, the findings of Cramer, De Juan & Tetreault (2016) demonstrate that the use of emojis can provide complementary situational and emotional information. To illustrate this, ‘I am on my way 🚗’ gives additional contextual information that the sender is on his way by car rather than by bike. Reflecting back on the Elaboration Likelihood Model, these findings could suggest that emojis can help the consumer process the message more deliberately when using the central route.

Taking into account that emojis enhance message understanding by making underlying meanings and emotions more visible (Huang et al., 2008) and congruent emojis are perceived as more understandable (Daniel & Camp, 2018), I suggest that emojis can contribute to the user engagement through the central route. Therefore, the following hypothesis has been formulated as:

• H2: The effect of emoji-use in firm-generated content is moderated by the congruency of the emojis such that the effect for congruent emojis is stronger compared to

incongruent emojis.


Emojis and sentiment

Not only do emojis attract attention and contribute to the understandability, but they can also communicate the emotions of the message (Thompson & Foulger, 1996). When designing firm-generated content, organizations can generally choose between two, not mutually exclusive, types of appeals: Emotional appeals or informative appeals (Wagner, Baccarella, Voigt, 2017). The first is centralized around evoking consumers’ emotions and the latter is focusing on communicating factual information about the brand or product (Bagozzi, Gopinath, & Nyer, 1999).

Rietveld et al. (2020) studied the effect of message content on consumer engagement in Instagram posts. In their study, it was found that emotional appeals influence consumer engagement more positively compared to informative appeals. In addition, Lee et al. (2018) focused on the effects of advertising content on consumer engagement on Facebook. Similar to Rietveld et al. (2020), they demonstrated that informative content leads to lower


When focusing on the emotions of the messages, various studies show that the sentiment of a tweet has an effect on user engagement. Firstly, Gruzd, Dorion & Mai (2011) studied the sentiment and sharing behavior of 46,000 tweets about the 2010 Winter Olympics. They found that positive tweets were retweeted more often compared to negative tweets. In addition, Stieglitz & Dang-Xuan (2013) discovered that of the 165,000 political tweets, the messages containing positive and negative emotions were more often and faster shared compared to neutral tweets. To continue, Tsugawa & Ohsaki (2017) built on the theory of the relationship between the sentiment of the tweets and the number of retweets. They found that negative tweets were spread faster and with a larger volume. Even though these studies have resulted in differing findings for negative and positive emotions, they do show that a stronger sentiment generates more engagement compared to neutral tweets. Taking into account that emojis can help display emotions and contribute to the sentiment of the message, the third hypothesis is defined as:

• H3: The effect of emoji-use on consumer engagement is moderated by the overall sentiment of the tweet such that the effect of emoji-use is stronger for firm-generated content containing a positive or negative sentiment compared to neutral tweets.


Figure 1: Conceptual model demonstrating the effect of emoji-use on engagement with moderating variables

Research methods

The aim of this study is to explore the effect of emoji-use on consumer engagement in firm- generated content. An innovative and effective way to study large amounts of online data is text mining (Berezina et al., 2016). Text mining can provide meaningful insights for vast amounts of unstructured data in a short period of time (Salloum et al., 2017). Considering the wide variety in the use of emojis in firm-generated content and various aspects contributing to user engagement, text mining is considered an effective and insightful approach to the

research aim of this study. Using text mining, it is possible to study a large sample and gather various variables that could be of influence on consumer engagement.


In total, 71,342 tweets from the top 50 of the Fortune 500 companies (2020) were retrieved.

The tweets were collected based on a fixed time frame from the 17th of June 2020 until the 17th of June 2021. This time frame has been chosen as the use of social media and emojis changes rapidly. Emoji-use in older tweets could vary from more recent tweets and could thus influence the results. In addition to the fixed timeframe, tweets with less than 20 retweets and 20 favorites were filtered out. Considering Fortune 500 organizations have a large number of followers, tweets with less than 20 retweets and 20 favorites were be perceived as a message that was not designed for the full audience. After applying this filter and removing tweets with missing values, a sample of 10,222 tweets was left.

Twitter has been chosen as social media channel since Twitter allows us to subtract data containing solely text. In addition, it is easier to gather Twitter data compared to

Instagram and Facebook, due to the Twitter API that is quite easy and quick to obtain. Despite that Instagram and Facebook are popular and interesting social media platforms, the content

Emoji-use Consumer engagement


Sentiment H1 (+)

H2 (+) H3 (+)


published on these channels is often focused around or accompanied by visuals. It has been demonstrated that visuals influence consumer engagement (e.g Li & Xie, 2020) and therefore the results for this study are more reliable when studying emoji-use in tweets without

accompanying visuals.

According to Center for Marketing Research (2020), 95% of the Fortune 500 companies owned at least one social media account in 2019. Fortune 500 companies are interesting to study, as they consist of organizations in various industries (Center for Marketing Research, 2020). Moreover, these organizations are influential and often have established social media accounts with a sufficient number of followers and user engagement.

Data collection

The data has been collected from Twitter by means of text mining using the Twitter API and R language. The complete syntax used to retrieve the data can be found in Appendix II. First, the tweets were gathered and prepared for further study by filtering out the retweets and replies such that only the organic firm-generated tweets were left. Moreover, the data was preprocessed and cleaned such that unrelated variables for this study, such as hashtags and URLs, were erased. To be able to study the emojis and indicate their specific characteristics, the emoji dictionary developed by Peterka-Bonetta (2015) and Lyons (2017) was

incorporated. Furthermore, to detect the sentiment of each emoji, the emoji sentiment data base from the sentiment analysis by Kralj Novak et al. (2015) was used. These two

preexisting files were uploaded to the R script and were used for the data gathering of the variables. Additionally, in order to establish the sentiment of the text in the tweets, a sentiment analysis was carried out following the lexicon by Jockers (2017). At last, the different variables were merged together into one data frame in RStudio and were then exported.

Variable operationalization

Firstly, this study will investigate the independent variable of emoji-use (present, not present) in tweets. This variable has been operationalized by means of the function ‘ji_detect’ from the

‘rtweet’ package by Kearney (2019) that indicates if there is an emoji present in the text.

Furthermore, the dependent variable of user engagement has been operationalized by retrieving two variables for the number of favorites and the number of retweets using the

‘get_timeline’ function following the package by Kearney (2019). This study solely focuses


on the favorites and retweets as the Twitter API did not allow to gather the number of replies.

Due to a limited time frame, the factor of replies has been left out for this research.

Moreover, the moderating variable of congruency (congruent, incongruent) was established by calculating the polarity score of the sentiment of the emoji with the range of - 1.0 (negative) to 1.0 (positive) following the emoji sentiment analysis by Kralj Novak et al.

(2015). As the emoji sentiment dictionary was not up to date to the current version of Unicode emojis (v13.1) and was not able to indicate every emoji, the emoji sentiment score was

adjusted. The emoji sentiment score was calculated based on the average sentiment score of the emojis that were possible to indicate. This score provided the opportunity to calculate the emoji sentiment without having to delete a high number of tweets due to missing values. The emoji polarity score was compared to the polarity score of the overall sentiment of the text in the tweet with the range of -1.0 (negative) to 1.0 (positive) using the ‘sentimentr’ function based on the lexicon by Jockers (2017). When the emoji polarity score was situated in the same range (-1 to -0.3 negative; -0.3 to 0.3 neutral; 0.3 to 1 positive) as the polarity score of the text, the emoji was considered congruent.

In addition, the moderating variable of sentiment was measured using the outcomes of the polarity scores from the sentiment analysis as well (Jockers, 2017). The tweet was rated from -1.0 to 1.0, in which -1 to -0.3 was considered negative, -0.3 to 0.3 neutral, and 0.3 to 1 positive.

Moreover, the moderating variable of the amount of emojis was measured by calculating the number of emojis present in the tweet using the ‘ji_count’ function of the

‘rtweet’ package (Kearney, 2019). This returned the number of emojis in the given tweet.

Lastly, to assure that the effect on consumer engagement was caused by the presence of emojis, the control variable of follower count was assessed. This was calculated using the

‘rtweet’ package and provided a variable with the number of followers of the firm’s Twitter account (Kearney, 2019).

Data analysis

In IBM SPSS 27, the data was checked and was given the correct labels and values. The data was tested on normality which demonstrated that the data for each variable was

overdispersed. A log transformation to adjust the non-normal distribution was used to correct the distribution of the data for the dependent variables. The hypotheses were tested by means of a multiple linear regression in one model for the number of favorites and one model for the number of retweets.



Table 1: Descriptives of the variables

Mean SD Min. Max.

Favorite count 1,053.82 3,143.62 21 132,411

Retweet count 197.61 713.96 21 35,455

Total engagement 1251.43 3806.21 42 167,866

Followers 3,195,905.85 3,663,469.41 5309 23,096,329

Emoji presence 0.10 0.30 0 1

Congruency 0.94 0.23 0 1

Text sentiment 1.48 1.69 -0.81 0.99

Emoji sentiment 0.04 0.14 -0.49 0.78

Overall sentiment 0.09 0.11 -0.40 0.73

Emoji count 0.18 0.984 0 72

Of the 10,222 firm-generated tweets, 10.0% of the tweets contained at least one emoji. The majority of the tweets containing emojis had one emoji present (63.1%). Furthermore, 19.6%

of the tweets contained two emojis and 9.60% of the tweets contained three emojis. In

addition, it was found that 64.6% of the tweets used incongruent emojis and 35.4% congruent emojis. The overall sentiment of the emojis was positive (73.8%) and neutral (25.8%). Only a few emojis were rated negative (0.5%). The majority of the sentiment of the text in the tweets was neutral (82.6%). Furthermore, 17.0% of the tweets were positive, and 0.5% were negative in text sentiment.

Table 2: Correlation matrix

1 2 3 4 5 6 7

Favorites Retweets

1) Favorite | Retweets -

2) Followers .612 .463 -

3) Emoji presence .140 .078 .165 -

4) Emoji count .055 .025 .052 .548 -

5) Congruency -.007 .029 -.016 -.342 -.190 -

6) Text sentiment -.089 -.111 -.087 .015 .012 -.528 -

7) Emoji sentiment .132 .078 .144 .887 .491 -.371 .024 -


A multiple linear regression was conducted to test the effects of emoji use on user

engagement in firm-generated content. The results indicated that the model was a significant predictor for the number of retweets (F(7, 10214) = 410.93, p <.001, R2 = .220) and favorites (F(7, 10214) = 410.93, p <.001, R2 = .220) in firm-generated content. The results of the multiple linear regression are summarized in table 3.

Table 3: Summary of linear regression results of the variables predicting user engagement

Adjusted R2 B SE B β p

Retweet Favorite Retweet Favorite Retweet Favorite Retweet Favorite Retweet Favorite

Follower count Emoji presence Congruency Text sentiment Emoji sentiment

Emoji presence * sentiment Emoji count

.219 .378

1.40 -.15 .02 -.46 .43 -.01 .00

2.56 .00 .01 -.32 .55 -.05 .00

.00 .08 .03 .07 .20 .08 .00

.00 .09 .04 .09 .24 .10 .01

.46 -.04 .01 -.07 .05 .00 .03

.60 .00 .00 -.04 .05 -.01 .00

<.001 .047 .611

<.001 .031 .91 .985

<.001 .995 .863

<.001 .019 .615 .826

Firstly, the multiple linear regression demonstrated there is a small significant negative effect of the use of emojis on the number retweets (β = -.04, p = .047) but not on the number of favorites (β = .00, p = .995). The use of emojis leads to lower retweets. Thus, H1 is not supported.

Furthermore, the multiple linear regression did not show a significant result for the effect of congruency for retweets (β = .01, p = .611) and favorites (β = .00, p = .863). A congruent emoji did not lead to higher user engagement compared to an incongruent emoji.

Hence, H2 is rejected.

Thirdly, the interaction effect of emoji presence and positive and negative sentiment of the tweet compared to neutral tweets showed a non-significant effect for both retweets (β = .00, p = .91) and favorites (β = -.01, p = .62). Positive and negative sentiment did not lead to more user engagement compared to neutral tweets. Therefore, H3 is rejected. However, the multiple linear regression showed a significant negative effect for text sentiment on the number of retweets (β = -.07, p < .001) and favorites (β = -.04, p < .001). A higher text sentiment score causes a decrease in user engagement. Moreover, the results show a


significant result for the sentiment of the emoji on user engagement for both retweets (β = .05, p = .031) and favorites (β = .05, p = .019). Emojis with a stronger sentiment cause an increase in user engagement.

An additional analysis was calculated to test the effect of the number of emojis on user engagement. However, it was found that there is a non-significant effect for the number of emojis on both retweets (β = .03, p = .985) and favorites (β = .00, p = .826). Thus, more emojis in a tweet does not lead to a higher user engagement.

The effects of the predictors were controlled for the number of followers and showed a significant effect for retweets (β = .46, p <.001) and favorites (β = .60, p <.001). The user engagement was higher for firms with more followers.

Discussion and conclusion

In this study, the effect of emoji-use on consumer engagement in firm-generated content was examined. The results suggest that there is no positive significant main effect of the presence of emojis on user engagement, rejecting H1. However, the results did demonstrate a small significant negative effect of emoji use on the number of retweets, indicating that the use of emojis leads to slightly less retweets compared to not using emojis. Moreover, the results showed that there is no significant effect for the congruency of the emoji on user engagement.

A congruent emoji does not lead to more user engagement compared to an incongruent emoji, leaving us to reject H2. Furthermore, no interaction effect was found for emoji presence and a positive and negative sentiment compared to neutral sentiment of the tweet. For tweets

containing emojis, positive and negative sentiment do not lead to more user engagement as compared to neutral sentiment. Therefore, H3 was rejected. However, it was found that the sentiment of the text leads to lower user engagement. In contrast, the sentiment of the emoji leads to higher consumer engagement. Lastly, an additional analysis discovered that the number of emojis does not have an effect on user engagement.

The result demonstrating that emoji-use does not influence the number of favorites and negatively influences the number of retweets was contradictory to the expectations and results of experiments in previous business articles (WordStream, 2018). It was expected that, based on the Elaboration Likelihood Model (Petty & Cacioppo, 1981), emojis would function as a visual peripheral cue to grab attention as well as to provide the reader with a deeper

understanding when using the central route. A possible reason for this result is that the emoji dictionary used in this research was not up to date to the current version of emojis and was not able to signal every emoji. This caused missing values and these emojis were therefore not


used for the analyses. Another reason can be found in business literature. Hiebert (2016) found that 58% of the consumers aged between 18 and 34 years old believe that firms are overusing emojis. Thus, the effects of emoji-use could be influenced by differences in consumer groups. Moreover, the number of impressions and replies were not measured and included in this study. This could have affected the results with regards to the outcomes of user engagement, as a retweet requires a higher level of user engagement compared to a view or favorite (Aldous, An & Jansen, 2019). Other forms of engagement, such as impressions, could provide different insights about the influence of emoji-use on user engagement.

The finding that emoji congruency does not affect user engagement in firm-generated content is in contrast with previous findings (Huang et al., 2008; Daniel & Camp, 2018). It was expected that congruent emojis would function as a central cue (Petty & Cacioppo, 1981), as they provide the reader with additional information about the context and

underlying meanings (Cramer et al., 2016). As the congruency in this research was calculated based on the average emoji sentiment and not each emoji individually, this could have caused different results.

In contrast to our expectations, no significant interaction was found for the use of emojis and sentiment of the tweet. However, a significant effect was found that stronger text sentiment leads to a lower number of retweets. This finding was contradictory to previous literature (Stieglitz & Dang-Xuan, 2013; Gruzd, Dorion & Mai, 2011; Tsugawa & Ohsaki, 2017) that demonstrated that stronger sentiment contributes to more user engagement

compared to neutral sentiment. A possible reason for the contradictory results could be due to high- and low-arousal appeals in the firm-generated content. Rietveld et al. (2020) found that negative high-arousal and positive low-arousal appeals lead to lower engagement rates. This study did not measure high- and low-arousal, which might have influenced the results.

Another reason for conflicting results could be the exceptionally low number of negative tweets. The number of tweets with negative sentiment was too low to make assumptions.

Nevertheless, the results did demonstrate that the sentiment of the emoji positively influences the number of favorites and retweets. This finding is in line with previous research explaining that the use of emotional appeals and stronger sentiment leads to more user engagement (e.g.

Rietveld et al., 2020; Gruzd, Dorion & Mai, 2011; Tsugawa & Ohsaki, 2017). As emojis contribute to the visualization of emotions, they can help display stronger sentiment and emotional appeals.


Lastly, the result that the use of more emojis does not lead to more engagement can be explained by the business study by Hiebert (2016). In the questionnaire, the majority of the consumers responded that they felt that organizations were overusing emojis in their firm- generated content. This could imply that consumers do not prefer content with many emojis present and that firms should be careful when using an abundance of emojis.

The findings of this study contribute to the literature about the use of emojis and sentiment in owned social media for organizations. In addition, the results demonstrate the importance of sentiment in firm-generated content for user engagement. Furthermore, this research has shown that it is possible to study a large amount of unstructured data by implementing text mining as a research method. This can be useful for future studies as text mining can be used for numerous research topics in online marketing. From a practical perspective, this study provides insights for organizations on how they can use emojis in their firm-generated content. The findings of this research show that emojis with a high sentiment help increase user engagement.

Despite the findings of this study, some limitations should be noted. First of all, this research was not able to incorporate every contributor to consumer engagement. In the current study, solely retweets and favorites were incorporated in the engagement rate, whereas in comments and impressions are also standard measures for user engagement. Secondly, the emoji dictionary by Bonetta (2015) and Lyons (2017) and the emoji sentiment list by Kralj Novak et al. (2015) were not up to date to the current list of Unicode emojis (v13.1). For this reason, not every emoji in the firm-generated content could be indicated and scored based on its sentiment.

Following the limitations, future research could focus on the effects of emoji-use on the total user engagement rate and could implement an updated emoji sentiment list such that more emojis and their effects on consumer engagement can be explored. In addition, future research could dive in deeper in the use of emojis in firm-generated content by studying the effects of different emoji-types on consumer engagement. Furthermore, different

organizations and/or products could be studied to see if there is a difference amongst the various sectors and firms. For example, future studies could look into the difference in effects for emoji-use for utilitarian and hedonic products.



Aldous, K., An, J. & Jansen, B. (2019). View, like, comment, post: Analyzing user engagement by topic at 4 levels across 5 social media platforms for 53 news organizations. In Proceedings of the Thirteenth International AAAI Conference on Web and Social Media (ICWSM).

Bagozzi, R., Gopinath, M., & Nyer, P. (1999). The role of emotions in marketing. Journal of the Academy of Marketing Science, 27(2), 184-206.

Bai, L. & Yan, X. (2020). Impact of firm-generated content on firm performance and consumer engagement: Evidence from social media in China. Journal of Electronic Commerce Research, 21(1), 56-74.

Belch, G. & Belch, M. (2001). Advertising & promotion: An integrated communications perspective, 5th ed. Boton: McGraw-Hill Irwin.

Berezina, K., Bilgihan, A., Cobanoglu, C. & Okumus, F. (2016). Understanding satisfied and dissatisfied hotel customers: Text mining of online hotel reviews. Journal of

Hospitality Marketing & Management, 25(1), 1-24.

Berger, J. (2013). Contagious: Why things catch on. New York: Simon & Schuster Paperbacks.

Broni, K. (2020, May 1). Emoji Use in the New Normal. Emojipedia. Retrieved January 29, 2021, from

Buchholz, K. (2020, 25 september). In 2021, Global Emoji Count Will Grow to 3,353.

Statista. Retrieved June 2, 2021, from of-emojis-from-1995-bis-2019/

Bullis, J. (2018, December 17). How to grab people’s attention on LinkedIn with emojis.

LinkedIn. Retrieved June 1, 2021, from peoples-attention-linkedin-emojis-james-bullis/

CMO Survey. (2021). The transformation of marketing: Emerging digital, social, and political trends. The Highlights and Insights Report 2021. Retrieved June 1, 2021, from Highlights_and_Insights_Report-February-2021.pdf

Cramer, H., de Juan, P. & Tetreault, J. (2016). Sender-intended functions of emojis in US messaging. In Proceedings of the 18th International Conference on Human-Computer Interaction with Mobile Devices and Services, 504-509.

Daniel, T. & Camp, A. (2018). Emojis affect processing fluency on social media. Psychology of Popular Media Culture, 9(2), 1-6.


Das, S., Jha, M., Sarkar, S., Sundaram, K., & Gayathri, N. (2017). Emojis: The language of the future? – A literature review. International Journal of Humanities and Social Sciences, 7(2), 135-139.

Eurostat. (2020). Social media – statistics on the use by enterprises. Retrieved January 14, 2020, from _statistics_on_the_use_by_enterprises#Social_media.2C_a_business_paradigm_shift from

Ge, J. & Gretzel, U. (2018). Emoji rhetoric: a social media influencer perspective. Journal of Marketing Management, 34, 1272-1295.

Grannan, C. (2016). What’s the difference between emoticons and emojis?. Brittanica.

Retrieved June 22, 2021, from between-emoji-and-emoticons

Gruzd, A., Doiron, S. & Mai, P. (2011). Is happiness contagious online? A case of Twitter and the 2010 Winter Olympics. In: Proceedings of the 44th Hawaii International Conference on System Sciences (HICSS ’22), 1-9.

Guesalaga, R. (2016). The use of social media in sales: Individual and organizational antecedents, and the role of customer engagement in social media. Industrial Marketing Mangagement, 54, 71-79.

Hiebert, P. (2016, November 17). Brands beware: Don’t overdo it with the emojis. YouGov.

Retrieved June 24, 2021, from reports/2016/11/17/consumers-tired-of-emojis

Hogenboom, A., Bal, D., Frasincar, F., Bal, M., De Jong, F. & Kaymak, U. (2013). Exploiting emoticons in sentiment analysis. In Proceedings of the 28th Annual ACM Symposium on Applied Computing, SAC 2013 (pp. 703-710). Association for Computing

Machinery (ACM).

Huang, A., Yen, D. & Zhang, X. (2008). Exploring the potential effects of emoticons.

Information & Management, 45(7), 466-473.

Jockers, M. L. (2017). Syuzhet: Extract sentiment and plot arcs from text. Retrieved June 1, 2021, from

Karlson, K. (2017, 30 May). How to use emojis in Facebook ads – Complete guide.

Adspresso by Hootsuite. Retrieved Februari 1, 2021, from

Kearney, M. (2019). Rtweet: Collecting and analyzing Twitter data. Journal of Open Source Software, 4(42), 1829. R package version 0.7.0.


Kim, L. (2018, 11 april). The stupid-simple secret ingredient to better engagement on Twitter.

WordStream. Retrieved January 21, 2021, from


Kralj Novak P., Smailović J., Sluban B. & Mozetič, I. (2015). Sentiment of Emojis. PLOS ONE, 10(12).

Kumar, A., Bezawada, R., Rishika, R., Janakiraman, R., & Kannan, P. K. (2016). From Social to Sale: The Effects of Firm-Generated Content in Social Media on Customer

Behavior. Journal of Marketing, 80(1), 7–25.

Lee, D., Hosanagar, K., & Nair, H. S. (2014). The effect of social media marketing content on consumer engagement: Evidence from Facebook (pp. 1-51). Stanford, CA: Stanford Graduate School of Business.

Li, F., Larimo, J. & Leonidou, L. (2021). Social media marketing strategy: definition, conceptualization, taxonomy, validation, and future agenda. Journal of the Academy of Marketing Science, 49, 51-70.

Li, Y. & Xie, Y. (2020). Is a picture worth a thousand words? An empirical study of image content and social media engagement. Journal of Marketing Research, 57(1), 1-19.

Lyons, K. (2017). Emoji dictionary. Retrieved May 17th, 2021, from

Malthouse, E., Haenlein, M., Skiera, B., Wege, E. & Zhang, M. (2013). Managing customer relationships in the social media era: Introducing the social CRM house. Journal of Interactive Marketing, 27(4), 270-280.

Mathews, S. & Lee, S. (2018). Use of emoji as a marketing tool: an exploratory content analysis. Fashion, Industry and Education, 16(1), 46-55.

McCulloch, G. (2019). Because internet: Understanding the new rules of language. New York, NY: Riverhead Books.

McQuarrie, E. F., & Phillips, B. J. (2008). Go figure! New directions in advertising rhetoric.

Armonk, NY: M.E. Sharpe.

Microsoft Canada (2015). Attention spans. Retrieved March 30, 2021, from

Moussa, S. (2019). An emoji-based metric for monitoring consumers’ emotions toward brands on social media. Marketing Intelligence & Planning, 37(2), 211-225.

Murphy, J., Hofacker, C. & Mizerski, R. (2006). Primacy and recency effects on clicking behavior. Journal of Computer-Mediated Communication, 11, 522-535.

Ogilvy, D. (1985). Confessions of an Advertising Man. New York: Atheneum.


Ordenes, G., Grewal, D., Ludwig, S., De Ruyter, K., Mahr, D., Wetzels, M., Kopalle, P., Morwitz, V. (2019). Cutting through Content Clutter: How Speech and Image Acts Drive Consumer Sharing of Social Media Brand Messages. The Journal of Consumer Research, 45(5), 988–1012.

Patel, D. (n.d.). How to use emojis to boost your social media posts. Social Pilot.

Retrieved 30 March 2021, from boost-your-social-media-posts

Peterka-Bonetta, J. (2015). Emoticons decoder for social media sentiment analysis. Opiate for the mass. Retrieved May 17th, 2021, from


Petty, R. & Cacioppo, J. (1986). The elaboration likelihood model of persuasion. Advances in Experimental Social Pyschology, 19, 123-192.

Petty, R., Cacioppo, J. & Goldman, R. (1981). Personal involvement as a determinant of argument-based persuasion. Journal of personality and social psychology, 41(5), 847.

Rana, N. P., Slade, E. L., Sahu, G. P., Kizgin, H., Singh, N., Dey, B., … Dwivedi, Y. K.

(2020). Digital and Social Media Marketing Emerging Applications and Theoretical Development (1st ed. 2020.). Cham: Springer International Publishing.

Ray, E. & Merle, P. (2020). Disgusting face, disease-ridden place?: Emoji influence on the interpretation of restaurant inspection reports. Health Communication. Retrieved February 1, 2021 from


Rietveld, R., Van Dolen, W., Mazloom, M. & Worring, M. (2020). What you feel, is what you like: Influence of message appeals on customer engagement on Twitter. Journal of Interactive Marketing, 49, 20-53.

Rutter, D. & Stephenson, G. (1977). The role of visual communication in synchronising conversation. European Journal of Social Psychology, 7(1), 29-37.

Salloum, S., Al-Emran, M., Monem, A. & Shalaan, K. (2017). A survey of text mining in social media: Facebook and Twitter perspectives. Advances in Science, Technology and Engineering Systems Journal, 2(1), 127-133.

Sehl, K. (2019). All the different ways to calculate engagement rate. Hootsuite. Retrieved June 6, 2021, from

Sharma, S. & Verma, H. (2018). Social Media Marketing: Evolution and change. In Heggde, G. & Shainesh, G. (2018). Social Media Marketing: Emerging concepts and

applications (pp. 19-36). Singapore: Palgrave Macmillan.


Sheriff, N. M., Zulkifli, A. S., & Othman, W. N. W. (2018). Accentuating Customer Engagement, Visual Presentation and Copywriting for Effective Social Media Marketing: A Case Study. International Journal of Academic Research in Business and Social Sciences, 8(12), 1619–1628.

Shin, D., He, S., Lee, G., Whinston, A., Cetintas, S., & Lee, K. (2020). Enhancing Social Media Analysis with Visual Data Analytics: A Deep Learning Approach. MIS Quarterly, 44(4), 1459-1492.

Smith, C. (2013, 18 september). Facebook Users Are Uploading 350 Million New Photos Each Day. Business Insider. Retrieved January 14, 2021, from 9?international=true&r=US&IR=T

Statista (2020, October). Number of daily active Facebook users worldwide as of 3rd quarter 2020. Retrieved January 14, 2021, from

Statista. (2021, 27 januari). Worldwide digital population as of October 2020. Retrieved January 29, 2021, from worldwide/

Stieglitz, S. & Dang-Xuan (2013). Emotions and information diffusion in social media:

Sentiment of microblogs and sharing behavior. Journal of Management Information Systems, 29(4), 217-248.

Stoner, M. (2009). Effectively communicating with visual aids. Retrieved January 14, 2021, Sullivan, D. (2014). Just like Facebook, Twitter’s new impression stats suggest few

followers see what’s Tweeted. Retrieved January 29, 2021,


Sundar, S. (2008). The main model: A heuristic approach to understanding technology effects on credibility. Digital media, Youth, and Credibility. Cambridge, MA: The MIT Press.


Thompson, P. & Foulger, D. (1996). Effects of pictographs and quoting on flaming in electronic mail. Computers in Human Behavior, 12, 225-243.

Thornhill, M., Xie, K. & Lee, Y. (2017). Social media advertising in a competitive market:

The effects of earned and owned exposures on brand purchase. Journal of Hospitality and Tourism Technology, 8(1), 87-100.

Tsugawa, S. & Ohsaki, H. (2017). On the relation between message sentiment and its virality on social media. Social Network Analysis and Mining, 7(19), 1-14.


Unicode, Inc. (2020). Emoji Counts, v13.0. Retrieved January 15, 2021, from

Vraga, E., Bode, L. & Troller-Renfree, S. (2016). Beyond self-reports: Using eye tracking to measure topic and style differences in attention to social media content.

Communication Methods and Measures, 10(2-3), 149-164.

Wagner, T., Baccarella, C. & Voigt, K. (2017). Framing social media communication:

Investigating the effects of brand post appeals on user interaction. European Management Journal, 35(5), 606-616.

Yuasa, M., Saito, K. & Mukawa, N. (2011). Brain activity when reading sentences and emoticons: an fMRI study of verbal and nonverbal communication. Electronics and Communications in Japan, 94(5), 17-24.


I. Top 50 Fortune 500 companies (2020) with Twitter accounts

Company Twitter account

1. Walmart @Walmart

2. Amazon @Amazon

3. Exxon Mobil @ExxonMobil

4. Apple @Apple

5. CVS Health @CVSHealth

6. Berkshire Hathaway -

7. UnitedHealth Group @UnitedHealthGrp

8. McKesson @McKesson

9. AT&T @ATT

10. AmerisouceBergen @Healthcare_ABC

11. Alphabet @Google

12. Ford Motor @Ford

13. Cigna @Cigna

14. Costco Wholesale @Costco

15. Chevron @Chevron

16. Cardinal Health @cardinalhealth

17. JPMorgan Chase @jpmorgan

18. General Motors @GM

19. Walgreens Boots Alliance @WBA_Global 20. Verizon Communications @Verizon

21. Microsoft @Microsoft

22. Marathon Petroleum @MarathonPetroCo

23. Kroger @kroger

24. Fannie Mae @FannieMae

25. Bank of America @BankofAmerica


26. Home Depot @HomeDepot

27. Phillips 66 @Phillips66Co

28. Comcast @Comcast

29. Anthem @AnthemInc

30. Wells Fargo @Wellsfargo

31. Citigroup @Citi

32. Valero Energy @ValeroEnergy

33. General Electric @generalelectric

34. Dell Technologies @DellTech

35. Johnson & Johnson @JNJNews

36. State Farm Insurance @StateFarm

37. Target @Target

38. IBM @IBM

39. Raytheon Technologies @RaytheonTech

40. Boeing @Boeing

41. Freddie Mac @FreddieMac

42. Centene @Centene

43. UPS @UPS

44. Lowe’s @Lowes

45. Intel @Intel

46. Facebook @Facebook

47. FedEx @FedEx

48. MetLife @MetLife

49. Walt Disney @Disney

50. Procter & Gamble @ProcterGamble

II. Syntax used to mine the tweets

#load packages library(twitteR) library(rtweet) library(dplyr) library(tidyverse) library(rvest) library(Unicode) library(NLP) library(tm) library(emo) library(tidyr)

# Twitter access

twitter_token <- create_token(

app = "Thesis_Toos",

consumer_key = "KkhhduggVvATwfzFv7MojDfip", consumer_secret =


access_token = "1363471306403041283-MMrmxOmLbHGTmfs4tCpohstS8Y6dvk", access_secret = "hqrwVTZ1U73ANF9JjD7izwyw0Leen5viOEs9rtB0xDVUi")

# gather general tweets for Amazon without replies and retweets Tweets_df <- get_timeline(c("@Walmart", "@Amazon","@ExxonMobil",

"@Apple", "@CVSHealth", "@UnitedHealthGrp", "@McKesson" , "@ATT",

"@GM", "@Ford", "@Healthcare_ABC", "@Chevro", "@cardinalhealth",


"@Costco", "@Verizon", "@kroger", "@generalelectric", "@WBA_Global",

"@jpmorgan", "@FannieMae", "@Google", "@HomeDepot", "@BankofAmerica",

"@ExpressScripts", "@WellsFargo", "@Boeing", "@Phillips66Co",

“@Comcast” "@AnthemInc", "@Microsoft", “@MarathonPetroCo”,

"@WellsFargo", "@Citi", "@ValeroEnergy", "@generalelectric",

"@DellTech", "@JNJNews", "@StateFarm", "@Target", "@IBM",

"@RaytheonTech", "@Boeing", "@FreddieMac", "@Centene", "@UPS",

"@Lowes", "@Intel", "@Facebook", "@FedEx", "@MetLife", "@Disney",

"@ProcterGamble"), n=20000, fromDate = 202006170000, toDate = 202106170000,is_retweet = FALSE)

Tweets_df <- subset(Tweets_df,$reply_to_status_id)) Tweets_df = Tweets_df[c("user_id", "status_id", "screen_name", "text",

"display_text_width", "is_quote", "is_retweet", "favorite_count",

"retweet_count", "followers_count", "friends_count", "listed_count",

"statuses_count", "favourites_count")]


#load emoji dictionary (see file)

emoji_dictionary <- read_csv("~/Dropbox/001 - School/01- MBA/Thesis/R analysis/emoji_dictionary.csv")

#Emoji_presence (TRUE / FALSE)

Tweets_df$Emoji_presence <- ji_detect(Tweets_df$text)

#Emoji_count (Amount)

Tweets_df$Emoji_count <- ji_count(Tweets_df$text)


Tweets_df$Emoji_extract <- ji_extract_all(Tweets_df$text)

#emoji sentiment library(readxl)

EMOJI_SENTIMENT <- read_excel("~/Dropbox/001 - School/01- MBA/Thesis/R analysis/EMOJI_SENTIMENT.xlsx",

skip = 1)

# this creates a matrix and a vector. The matrix is the sentiment score for each emoji in the tweet. Obviously it is 0 for no emojis.

# num_emojis is a vector that simply counts the number of emojis per tweet. So if that number is 5, then that means there are 5 emojis in the tweet. The matrix then shows 5 numbers and 275 zeros for that row.

(I have used 280 as maximum number of emojis because that is how many characters a tweet has, just in case). Of course most rows will have almost everything zero. A row with 1 emoji will have a score in the first column, and then 279 zeros. If a tweet has 50, then it has scores in the first 50 columns and zeros in the rest.

emoji_sentiment_scores=matrix(0,length(Tweets_df$Emoji_extract), 280) num_emojis = rep(0,length(Tweets_df$Emoji_extract))

for(i in 1:71468) {

if (Tweets_df[i,"Emoji_presence"]==TRUE){

num_emojis[i] = length(Tweets_df$Emoji_extract[[i]]) for(j in 1:num_emojis[i]){

tmp =

which(Tweets_df$Emoji_extract[[i]][j]==EMOJI_SENTIMENT$Char) if (length(tmp)!=0){

emoji_sentiment_scores[i,j] = EMOJI_SENTIMENT$SentimentScore[tmp]

} else {emoji_sentiment_scores[i,j] = -10}



} }

# this second for loop calculates the average sentiment score for all emojis. I have made it that it will take the average of each row, only for the number of emojis it has so that it doesn't count all the other zeros.

avg_emoji_sent = rep(0,length(Tweets_df$Emoji_extract))

avg_emoji_sent_interpolated =rep(0,length(Tweets_df$Emoji_extract)) emoji_not_found =rep(0,length(Tweets_df$Emoji_extract))

for (i in 1:71342) {

tmp = as.numeric(emoji_sentiment_scores[i,1:num_emojis[i]]) tmps = which(tmp==-10)

if (length(tmps)!=0) { emoji_not_found[i]=1

avg_emoji_sent_interpolated[i] = mean(tmp[-tmps]) } else {

avg_emoji_sent_interpolated[i]=mean(tmp) avg_emoji_sent[i] = mean(tmp)}


Tweets_df$avg_emoji_sentiment = avg_emoji_sent

Tweets_df$avg_emoji_sent_interpolated = avg_emoji_sent_interpolated Tweets_df$emoji_not_found = emoji_not_found

#Sentiment of the tweet: polarity score (sentiment analysis) install.packages("sentimentr")


average_sentiment = rep(0,length(Tweets_df$text)) std_sentiment = rep(0,length(Tweets_df$text)) word_count = rep(0,length(Tweets_df$text)) for (i in 1:71342) {

tmp = sentiment_by(Tweets_df$text[i], polarity_dt = lexicon::hash_sentiment_jockers_rinker)

average_sentiment[i] = tmp$ave_sentiment std_sentiment[i] = tmp$sd

word_count[i] = tmp$word_count}

Tweets_df$average_sentiment = average_sentiment Tweets_df$std_sentiment = std_sentiment

Tweets_df$word_count = word_count

#remove emoji_extract from data frame library(dplyr)

Tweets_df_final <- select(Tweets_df, -Emoji_extract, -text)

#convert logical numbers to numeric library(magrittr)

Tweets_df_final %<>% mutate_if(is.logical, as.numeric)

#export dataframe

write.csv(Tweets_df_final, "Tweets_df_final_2206_CORRECT.csv") write.csv(Tweets_df_final, file="~/Dropbox/001 - School/01- MBA/Thesis/R analysis/tweets_df_final2.csv", quote = FALSE)




Related subjects :