#iamsohappy: A perspective on emotional content on Twitter and value creation

(1)

#iamsohappy: A perspective on emotional content on

Twitter and value creation

Xeniya Kondrat 11105720

Supervisor: Niels van Doorn 2nd reader: Natalia Sanchez Querubín

24-06-2016

Media Studies: New Media and Digital Culture University of Amsterdam

(2)

(3)

3.1 Big Data and datafication __________________________________________________________ 16 3.2 Predictive analytics _______________________________________________________________ 20 3.3 Sentiment analysis ________________________________________________________________ 21 4. TWITTER ___________________________________________________________________ 23 4.1 What is Twitter? _________________________________________________________________ 24 4.2 APIs ___________________________________________________________________________ 30 5. METHODOLOGY ____________________________________________________________ 32 5.1 Literature analysis _______________________________________________________________ 34 5.2 Close reading of the case study _____________________________________________________ 35 5.3 Interviews_______________________________________________________________________ 38

5.3.1 Skype/phone interviews _________________________________________________________________ 41 Interviewee 1 - Scott Kildall ________________________________________________________________ 42 Interviewee 2 - Svetlana Kiritchenko _________________________________________________________ 42 Interviewee 3 - Ibo van de Poel ______________________________________________________________ 43 Interviewee 4 - Sabine Roeser _______________________________________________________________ 44 Interviewee 5 - Adam Arvidsson _____________________________________________________________ 44 5.3.2 Email interview ________________________________________________________________________ 45 5.3.3 Limitations ___________________________________________________________________________ 45 6. DISCUSSION________________________________________________________________ 47 7. CONCLUSION_______________________________________________________________ 60 8. BIBLIOGRAPHY_____________________________________________________________ 62 9. APPENDIX _________________________________________________________________ 83 List 1- Stock market shares owned by EquityBot ________________________________________________ 83 List 2 – EquityBot’s keywords ______________________________________________________________ 83 Interview with Scott Kildall ________________________________________________________________ 87 Interview with Prof. Adam Arvidsson _________________________________________________________ 96 Interview with Prof. dr.ir. I.R. Ibo van de Poel _________________________________________________ 101 Interview with Prof. dr. Sabine Roeser _______________________________________________________ 106 Interview with Svetlana Kiritchenko _________________________________________________________ 111

(4)

4

1. INTRODUCTION

Most of us are using at least one social media platform today. Some just want to connect with their friends, others wish to socialize and find new friends or perhaps to promote themselves. There are many reasons, but they are irrelevant because we still all share something on these platforms. Creating a profile on these platforms requires giving them at least a modicum of information, valid or otherwise. We update our status to let our friends know how we are doing, posting videos and pictures that we like or share stories about our experiences. All of these become a part of large datasets that are collected, stored and sold as consumer behavior information to large corporations and even to stock markets. Our information is analyzed and stored, so it can be used to better predict future purchases and advertising online. We are used to hearing and reading about personal data being published or shared with a third-parties, where we imagine something like social security numbers, telephone numbers and bank details being given away, but we do not really think that our emotions could be a part of this data set. We tend to perceive our emotions as irrational and unstable, so then how can they be a part of something rational and serious and attract the attention of large corporations or the stock market? Well, a long time ago, Gabriel Tarde mentioned that stock market exchange numbers actually presented “changes in public desires and interests” as well as society’s attitudes towards the success of certain corporations (Social Laws 22). However, his statements were barely noticed until recently. The continuous and rapid developments of new technologies, which have broadened data gathering possibilities, bring his ideas to life with the growth of behavioral economics and finance. Emotional content became as important and valuable as other data today, because we started to learn how to measure, quantify and datafy these complex psychological states of human beings, which are expressed online. Nevertheless, there are several issues with our emotional content being scraped, measured and datafied such as privacy issues, stereotyping, generalization of the public and making incorrect predictive assumptions.

This thesis aims to analyze how our online emotional content, specifically on Twitter, are consumed and treated as a valuable trading asset in our modern world. This will be done by looking at a literary analysis of several theories and hypotheses, which were suggested by various

researchers and philosophers in the past. This will include a discussion about value creation and the role of classified and measured emotions, as well as their datafication through predictive analytics and sentiment analysis, which are all a part of the valuation process of these emotions. Later, Twitter will be discussed as a social media platform that allows all the parts of the valuation process to be

(5)

5

done within one platform. To support the theoretical discussion, a case study will be used, which is EquityBot - a stock trading algorithm that trades based on the correlation between the frequency of emotions on Twitter and fluctuations in the stock market. This case study, which is an artistic

project by nature, is particularly interesting because it correlates random emotions with stock shares. This will be one of the key points of the discussion, where I would argue, that these types of

correlations are more assumptions and artifacts of the desired outcomes of those performing the predictive analytics. Therefore, the research question of this thesis is:

In what way is emotional content on Twitter transformed into valuable data?

This thesis is important because there is a growing tendency in academic and corporate circles of trying to correlate various emotions, which are predominantly collected from Twitter, to different events. These are then made to attempt to predict things such as Oscar nominees and winners, the popularity of movies or its box office returns, urban crime prediction, presidential elections, stock market fluctuations and others. However, most of this research is not comprehensive and explanatory, and some of it cannot even explain why there is a correlation. Therefore, there is a need to look critically at this tendency and analyze the past and current situation, as well as see in what way we are conceptualizing value in relation to emotions, since it can now be measured and evaluated.

This thesis consists of three chapters which are parts of the theoretical framework. The first chapter is the Value of Emotions, which will discuss the role of emotions in economics, the

classification and measurement of emotions and finally their value online. The second chapter, the Datafication of emotions, will focus its discussion on Big Data and Datafication. Later, predictive analytics and sentiment analysis will be evaluated as separate subchapters. Finally, Twitter’s appearance, development and current state will be discussed together with methods of scraping its data. Afterwards, the Methodology chapter will discuss methods, literary analysis, the close reading of the case study and interviews with professionals, which were used for acquiring knowledge necessary for this thesis. The sixth chapter is the Discussion, where the results of this thesis will be discussed and analyzed together with the literature from the theoretical framework, the interviews and the case study. Finally, the conclusion will discuss the outcome of this thesis and suggest further possible research.

(6)

6

2. VALUE OF EMOTIONS

The first chapter of the theoretical framework will focus on the value of emotions. Before starting a discussion about what is the value of emotions, it is important to elucidate the role of emotions within economics. Further, it is necessary to take a look at how emotions are classified, measured and subsequently quantified before they acquire value. Finally, this chapter will discuss the valuation of emotions on the Internet. In this chapter, I will try to indicate the lack of diversity in this study and the role of emotions within the field of economics and show how participatory culture presently affects valuation.

2.1 Role of emotions in economics

Psychological and behavioral studies have been used widely in the study of the origin and development of emotions, their differences and classifications, their influence and effect on humans, as they are one of the general psychological characteristics of human nature (Vetriselvi and Vadivel 719). According to Ekman, emotions manifest rapidly; their beginning can be untraceable due to this spontaneity (54). This factor allows human beings to react to current events and respond to them accordingly, whether alone or in a group (Ekman 46). The interpretation of emotions within these social interactions is important to understand their origin and means of action (Berezin 110),

because emotions are unstable. Therefore, they were (and still partly are) considered to be irrational and unpredictable within the field of economics.

According to Pixley, in the last twenty (presently thirty) years, there are two constituencies in economics – those strictly committed to the rational model1 and those who would challenge it and orthodox economics in general (70). However, both of them agree that expectations play an

important role within economics, whereas non-orthodox economics emphasize uncertainty, while for adherents of orthodox economics uncertainty is risk (Pixley 71). When an economic decision needs to be made while facing drastic uncertainty, emotions such as fear and panic become a prime predictor of possible outcomes within unorthodox economics (Pixley 71). Lately, more research about emotions in economics appeared due to the development of a new sub-discipline known as behavioral economics, which studies “cognitive, social, and emotional influences on people's observable economic behavior” and tries to “change the way economists think about people’s perceptions of value and expressed preference” (Samson 1). Even though behavioral economics is gaining more attention among the research community, the research is still mainly limited to

1

Rationality “refers to a means-end procedure of decision-making with the end overwhelmingly assumed in the literature to be interest-in economic terms the single 'maximand' is money” (Lane 43)

(7)

7

positive (happiness) or negative (fear, panic) emotions and rarely includes other types. There is a need of broader and more in-depth research on emotions and their effects on economics and society, since emotions are inevitable and inseparable parts of human existence and cannot be limited to negative/positive classification. As Pixley states “[emotions] cannot be ‘reduced’ and indeed they are not necessarily ‘irrational’. That is, the state of expectations cannot be arrived at without emotions, and trust, confidence, fear and so forth can play a rational role in actual decisions” (83). Behavioral economics is related to a sub-field of finance, which is known as behavioral finance and studies “the influence of psychology on the behavior of financial practitioners and the subsequent effect on markets” (Sewell 1). Behavioral finance2

investigates causes of market inefficiencies (Sewell 1). The emotions of society with their positive or negative sentiment have a certain influence on investors’ financial decisions and thus the overall market (Nofsinger 144). These sentiments can modulate the risk-adversity of investors, which in turn has an impact on the stock market, “the market in which shares of publicly held companies are issued and traded either through exchanges” (“Stock Market Definition”), which causes it to go up or become impermanent. Thus, emotions and the overall social mood have an influence on the “tone and character of financial and economic activity” (Nofsinger 145). As multiple studies have found, various events and subsequent emotions from these events might cause an increase/decrease in stock prices. For example,

Hirshleifer and Shumway3’s study found a positive correlation between sunny weather and stock prices which they inferred was related to happiness. Another study of Kaplanski and Levy4 found an inverse correlation between aviation disasters and stock prices due to emotions such as fear and anxiety. Trading on the stock market can be performed online through special platforms or by physically being on the floor of the market with the help of a stock trader or a stock trading company. Virtual stock trading is emphasized in this thesis, specifically high frequency trading (HFT) algorithms. HFT has “a rapid high turnover of capital in rapid computer-driven responses to changing market conditions” (Aldridge 1). HFT has its advantages, such as stabilization of market systems, increased market efficiency and innovation in computer technology (Aldridge 2). But also disadvantages, because HFT requires copious amounts of daily data, and gain/loss signals might be misinterpreted if they are not clear and executed speedily, which can be achieved only with

2_{The main difference between economics and finance is that “[behavioral] economics is philosophical, [behavioral]}

finance is pragmatic” (N. Smith).

3

D. Hirshleifer and T. Shumway “Good Day Sunshine: Stock Returns and the Weather” https://www.jstor.org/stable/3094570?seq=1#page_scan_tab_contents

4

G. Kaplanski and H. Levy “Sentiment and Stock Prices: The Case of Aviation Disasters” http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1084533

(8)

8

computer-based automation (Aldridge 4). Now, value creation happens with the help of HFT within several seconds. But what is value and how it can be created?

According to the online version of the Merriam-Webster dictionary the origin of value comes from 14th century and in Middle-English means “worth or high quality” (“Value”). Today’s

definition, according to the same source, is:

1. a fair return or equivalent in goods, services, or money for something exchanged 2. the monetary worth of something : market price

3. relative worth, utility, or importance <a good value at the price> <the value of base stealing in baseball> <had nothing of value to say>

4. a numerical quantity that is assigned or is determined by calculation or measurement <let x take on positive values> <a value for the age of the earth>

5. the relative duration of a musical note

6. a) relative lightness or darkness of a color : luminosity

b) the relation of one part in a picture to another with respect to lightness and darkness 7. something (as a principle or quality) intrinsically valuable or desirable <sought material

values instead of human values — W. H. Jones> 8. denomination 2 (“Value”)

It is interesting, that two first modern definitions of value are related to the market and economy, and only the third definition describes the general idea of “being worth”, as it is originally intended to. According to Graeber, value can be defined in three ways as:

1. ‘values’ in the sociological sense: conceptions of what is ultimately good, proper, or desirable in human life; 2. ‘value’ in the economic sense: the degree to which objects are desired, particularly, as measured by how much others are willing to give up to get them; and 3. ‘value’ in the linguistic sense, which goes back to the structural linguistics of Ferdinand de Saussure5, and might be most simply glossed as ‘meaningful difference’ (1-2).

Through the history of theory of value development, the main question or more a problem of it was (and still is) - “how much?” According to Bentham, who suggested to measure pleasure and pain based on their intensity and duration, value “has to be investigated and summed up in an account or an overall “balance” as a guide for decision-making and action; second, the pragmatic concern with

5

(9)

9

the technical questions of how to measure pain and pleasure requires calculative practices able to deliver numbers as material for utility calculations” (3). Adam Smith, the Scottish economist and moral philosopher, was one of the first to suggest that there are two types of value - ‘value of use’ and “value of exchange’, where in both cases, “labor, therefore, it appears evidently, is the only universal, as well as the only accurate, measure of value, or the only standard by which we can compare the values of different commodities, at all times, and at all places” and “at the same time and place, therefore, money is the exact measure of the real exchangeable value of all commodities” (A. Smith). This work of Smith’s motivated David Ricardo to discuss the value theory in his work “Principles of Political Economy and Taxation”, where he further developed Smith’s idea about correlation of labor and value by detaching value from money and suggesting that “the value of a commodity, or the quantity of any other commodity for which it will exchange, depends on the relative quantity of labor which is necessary for its production, and not on the greater or less

compensation which is paid for that labor” (1). Later, Gabriel Tarde, a French sociologist and social psychologist, suggested that value is both qualitative (subjective) and quantitative (objective):

“It [Value] is a quality, such as color, that we attribute to things, but that, like color, exists only within us by way of a perfectly subjective truth. It consists in the harmonization of the collective judgments we make concerning the aptitude of objects to be more or less-and by a greater or lesser number of people-believed, desired or enjoyed. Thus, this quality belongs among those peculiar ones which, appearing suited to show numerous degrees and to go up or down this ladder without changing their essential nature, merit the name quantity” (Latour and Lepinay 8)

According to Latour and Lepinay, already in the past, Tarde criticized economists for not being open-minded and playing it safe and “‘ignoring the wealth of human subjectivity,’ they strive to ‘quantify all’ at the risk of thus ‘amputating’ what is human from its ‘moral, emotional, aesthetic and social dimensions’” (12). Tarde is especially important for this thesis, since he was one of the first to believe that emotions are valuable and should be measured and quantified as an integral part of economics, which should give attention to the collective psychological attitudes of society. He believes that value consists of a combination of collective beliefs, desires, subjective things, ideas and willpower, which can be quantifiable and measured, and that fluctuations on the stock market cannot be explained without their psychological causes such as hope or fear within society

(Psychologie économique 630). Tarde’s thinking (together with Adam Smith and Jeremy Bentham) became the basis of behavioral economics, as discussed previously. According to Tarde, everything is subjective in economics, or actually inter-subjective, and that, as he argues, is the reason why it

(10)

10

can be quantified (Latour and Lepinay 7). However, the main issue for Tarde was that there was no proper way of constantly measuring content such as emotions. Before measuring emotions, they need to be categorized or classified, captured and finally measured, so at the end they can be valued. Therefore, to understand how emotions can be quantified the next subchapter will discuss

classification and measurement.

2.2 Classification and measurement

We, as a society and species, classify everything, it has been said: “To classify is human” (Bowker and Star 1). Starting from our birth, we are classified as females or males, healthy or ill, Caucasian or Asian et cetera. We classify furniture, languages, food, material or immaterial objects or anything else that we use or see in our everyday life. Categorization is also used within the governmental, medical and juridical and other fields, where it is used as a base of structuralization of service, strictly complied with and virtually unmalleable. Classification, a mostly invisible tool, has been long studied within various fields of research. “A classification is a spatial, temporal, or spatio-temporal segmentation of the world” (Ibid., 10). Another author defines classification thusly: “classificatory techniques may be applied in an information system in order to facilitate access, organization, use, and retrieval” (Hunter 4), such as in manual filing or electronic data processing system. Anything can be classified, even emotions and feelings. For a couple of decades, with the development of the fields of sociology and psychology, several different classifications of emotions such as Ekman’s list of emotions6_{, Plutchik’s wheel of emotions, Parrots’ classification of emotions}7 as well as cognitive-mediational theory of emotion8, and others have appeared in academia. As an example, Plutchik’s model is a classification system of twenty-four emotions, developed by psychologist Robert Plutchik. In 1958, Plutchik established eight basic primary emotions – anger, fear, sadness, disgust, surprise, anticipation, trust, and joy, by creating “wheels of emotions”, where these eight primary emotions are grouped into positive versus negative emotions (Plutchik 349).

6

Ekman (1999) determined a list of basic emotions, which includes amusement, contempt, contentment, embarrassment, excitement, guilt, pride in achievement, relief, satisfaction, sensory pleasure, shame, anger, disgust, fear,

sadness/distress.

7

Parrott (2001) classified over 100 emotions and arranged them into a tree structured list.

(11)

11

Image 1: Three-dimensional circumplex model of correlations between emotions taken from http://bit.ly/28P5uYH

The above Image 1 describes the relations between emotions, which are presented by their intensity and the degree of similarity between differing types of emotions according to their color, where eight sectors represent eight primary emotions and arranged into four pairs of opposites. Two, three or more emotions in the circumplex model can be blended to create a new emotion just as colors may be mixed to create new ones. This classification provides a broader view of the emotions, rather than dichotomizing negative or positive emotions, which is common in such research.

Classification also appears online, through search engines, or on personal Facebook

accounts, where a user needs to fill in data about themselves in specific information boxes or create, upload and sort out his/her photographs into albums.Classifications tend to constantly change their structures, actions and usage accordingly to the needs of the event/research, becoming conceptual and material simultaneously (Bowker and Star 289). For example, the social buttons on Facebook, which allow users to express their interests in certain content on the platform. The 24th of February 2016 became an important date for Facebook and its “Like” button design. Before this date,

Facebook users could express their feelings only through a “Like” button and now they may choose between Like, Love, Haha, Wow, Sad and Angry (Krug). After re-classifying the list of the

emotions, which Facebook researched, selected, designed and integrated, the bigger and more in-depth emotional information is available for collection and analysis. The main purpose of the

(12)

12

reaction buttons, according to one of the Facebook design directors Geoff Teehan, were universality and expressivity (Stinson). At the same time, users are not only clicking one button, but they decide and interpret their own feelings to click on the “right” button, which will express their feelings best. This fits well with the statement of Bowker and Star, where they suggested that “people as active interpreters of information who themselves inhabit multiple contexts of use and practice” (291). However, it is still unclear how well this pre-selected list of reaction buttons expresses the real emotions of users about content seen on Facebook, since emotions and standards differ per culture, society or even between individuals (Ibid., 293). Nevertheless, it obviously gives a broader range of emotions to research within the platform, nor can one expect ever be a perfect uniform classification system on any platform that will accurately represent society’s emotional range (Ibid., 322). But even these imperfect classifications yield the possibility to perform measurements and analysis of user’s reported emotional reactions.

These new ways of gathering data through various channels and devices require a

“gloriometer”, as Tarde stated, or “valuemeter” (Latour and Lepinay 18) which will be able to make this data readable and visible and which will allow one to calculate the value of users’ (emotional) data. Previously Tarde focused primarily on glory as a value, since fame and admiration were good indicators and measures of social value in the society (qtd. in Latour and Lepinay 10-11). Later, Latour and Lepinay suggested to look everywhere and elsewhere for the “valuemeters”, which are able to capture and collect “human souls” (22) and which can be not only fame, admiration or popularity. According to Tarde, the theoretical possibility of measuring emotions is already sufficient enough for economics, because it is impossible to measure belief, faith, emotions practically without just making it “a single entity with inadequate approximation” (On

communication and social influence 111). Today, the measurement is more focused on the attention dedicated to something like a brand, company or product rather than glory. According to Arvidsson, there are measurements, or in another term “brand valuators”, which measure “‘brand awareness’ (how many people know about a brand), ‘brand associations’ (if the brand gives rise to positive associations) and ‘brand loyalty’” (“Brands” 250-251). These measurements are collected through various sources such as social media platforms, surveys, loyalty cards barcodes and others. They try to capture and understand users’ feelings, perceptions, attitudes and understanding of a brand by attempting “to measure the type and intensity of sentiments that can be associated with brands, and to gauge the extent to which brands may function as nodes within social networks” (Lury and Moor 440). Those “brand valuators” are a modern incarnation of Tarde’s “gloriometers”, which are able to

(13)

13

collect diverse information and measurements and are not limited only to beliefs and desires and can additionally probe the intensity of feelings. Other aspects that became important for measurement are the impact and reputations of users and companies, where platforms play a role of measure of impact by analyzing recognition and status, where reputation indicates the quality of the impact (Arvidsson “The Ethical Economy” 333). As can be seen, the value and its measurement do not only rely on productivity or time any longer, but also take into consideration “affective investments” (Arvidsson and Colleoni 140), which are attention and reputation. However, as Lury and Moor pointed out, attention, feelings and beliefs of the consumers about the brand equity9 are personal perceptions of the consumers in their minds, which may or may not become an actual behavior in the future and might be measured with manipulation (questions framed to yield positive results) or are subjective in nature (445). Before assessing the perceptions and emotions of consumers, the researcher needs to choose how to classify them, which also brings additional subjectivity into the measurement and the future analysis of collected information and its valorization. Despite these issues, the classification and measurement of various states of society are popular among the research community, which is trying to investigate, analyze and discover new ways of classifying and measuring by developing new methods and tools. The development of multiple technologies, including online ones, gave researchers new fields to study, classify and measure.

2.3 Value of emotions on Internet

Currently, social media platforms are heavily dependent on their users, who create the content on their platforms and thus generate value. The most recent example of such a case is a video which was published on a personal Facebook page through the Facebook Live application10, of a woman named Candace Payne who just bought a Star Wars Chewbacca talking mask and decided to share her excitement with her friends. The video11 was uploaded on the 19th of May 2016 and has presently as of the 25th of May 2016 gained 145,399,926 views and been shared more than 3 million times. A significant effect of this video is that within couple of days this mask was

completely sold out in the department stores of Kohl’s, where Payne bought her mask and by Sunday it was sold out in other stores as well (Eordogh). Payne’s video created such a buzz around the mask, that Kohl’s store visited her and gifted her and her family with more Star Wars presents as well as gift cards ($2500 value and 10,000 award points) (Eordogh). According to Jenkins, this is an

9_{Which are a part of set of brand equity according to marketing professionals and academics, where brand equity is “ a}

measure that attempts to capture consumers’ perceptions of a brand” (Lury and Moor 440)

10

Facebook Live allows Facebook users to stream online videos on their profile pages https://live.fb.com/about/

(14)

14

example of participatory culture, which is formed online and can be defined as a culture where artistic expression and civic engagement are not limited; people are encouraged to create and share amongst one another; members believe that their voluntary contributions are valued (Jenkins et al. 5-6). So, the viewer has its personal online opinion and is encouraged to share it with others by means of the Internet and social media. This is then combined with opinions of others and allows

researchers and major corporations to study the preferences, needs and desires, the likes and dislikes, habits and concerns of their audience. All of this data is being collected, analyzed and “deployed to valorize its ‘audience power’” (Arvidsson and Bonini 163). In the case of the

Chewbacca masks, after such a rapid growth in popularity, these masks are now available only on eBay, but not for their original price of $30, but from $50 - $200 apiece (Chan). This example is of course a unique one and many vloggers12 cannot reach this amount of viewers in such a short time. However, this example is the best one (because it is so exaggerated, but in a positive sense) to show what Arvidsson and Colleioni wrote of. There was no intention of creating value by the user, at least not for the company that produced or sold the mask, and no intention of monetary gain. It was intended for her and her friends, because she was so excited about her purchase she could not wait to share it. Impatience, excitement and the surprise of the user are valuable for the product and brand in this case, because “their [products] value is not primarily dependent on physical characteristics but is self-generated by those who invent, make, and use the products” (Hutter 202). Her laugh was enough for the viewers to start sharing, liking, commenting, re-uploading the video and finally purchasing the mask itself. Generated content and following feedback and reactions of others is a new way of providing “new sources of data on prizing and appraising—new means to register value judgments in the economy” (Stark 327), where according to Dewey prizing is “a sense of the sense of holding precious, dear (and various other nearly equivalent activities, like honoring, regarding highly)” and appraising as “in the sense of putting a value upon, assigning value to” (5). This video obviously affected not only the companies that are producing and selling the mask and the users that saw the video, but also Facebook, since it is was used as the primary platform for sharing this video with its own live video application, thus this video generated value for it as well.

Therefore, today the meaning of value is indirectly changed and instead accented on “influence, engagement, ‘clout’, or, in marketing, loyalty and experience. In simple terms, it is considered valuable not only to have many users, but also to have passionate and engaged users who

(15)

15

are able to influence others” (Arvidsson and Bonini 159). The number of viewers is still considered important, but the spreadability13 of their opinions and feelings among their followers, friends, and subscribers becomes more relevant. This is the main argument of this chapter, where I attempt to show through theory and various examples, that the construct of value has changed as compared with previous decades, and that it is no longer related to time or productivity anymore rather the amount of attention, spreadability and the engagement of the users and consumers. This became possible because of the continuous development of various technologies, which lead to the

generation of large amounts of data, where all these users’ opinions, emotions, beliefs and desires become publicly available for collection, measurement and analysis. The next chapter of this thesis will focus on the datafication of emotions and how emotions can be analyzed after they are datafied by various available methods and tools.

3. DATAFICATION OF EMOTIONS

"If you torture the data long enough it will confess to anything" Ronald Coase (qtd. in Tullock 205)

After the discussion about value, its creation and role of emotions within economics in the previous chapter, there is a need to examine how this value is actually collected and further analyzed. In the previous chapter, the term “data” was repeatedly mentioned, but what does it actually mean? The word “data” is a plural form of the Latin word “datum”, first known use of which dates back in 1646, and means “something given or admitted especially as a basis for reasoning or inference” (“Datum.”). According to Merriam-Webster online version of vocabulary, data means:

“1: factual information (as measurements or statistics) used as a basis for reasoning, discussion, or calculation [...]

2: information output by a sensing device or organ that includes both useful and irrelevant or redundant information and must be processed to be meaningful

3: information in numerical form that can be digitally transmitted or processed” (“Data”). The ways of generating data are increasing with the rapid development of technology and nowadays includes not only manually collected information through questionnaires or interviews, but also

13

A concept created by Jenkins, who argues that “assumes that anything worth hearing will circulate through any and all available channels, potentially moving audiences from peripheral awareness to active engagement” (Jenkins, Ford, and Green7).

(16)

16

computer-based methods as online surveys or collection of information from websites. These methods were used both by researchers and companies, who were seeking understanding in what society is and wants. Today, new ways of generating data include wearable devices as bracelets, our smartphones, cars, television sets with receivers, drones with cameras, video game consoles,

medical equipment and social media platforms. This huge and constantly growing amount of data is now commonly called “Big Data”. The next subchapter will discuss what Big Data is and the process of datafication of information.

3.1 Big Data and datafication

With the development of Web 2.0 as a platform, software applications started to work as native web applications instead of desktop applications (O’Reilly 20). The idea of these new Web 2.0 applications was (and still is) that “delivering software as a continually-updated service that gets better the more people use it, consuming and remixing data from multiple sources, including

individual users, while providing their own data and services in a form that allows remixing by others, creating network effects through an "architecture of participation" (O’Reilly 17). Web 2.0 is known for its encouragement of users to produce, upload and share their own content,

user-generated content, and being participation-platform”, in comparison to Web 1.0 “Web-as-information-source” (Song 251-252). According to Gillespie, the current term “platform” combines four connotations such as computational, “an infrastructure that supports the design and use of particular applications, be they computer hardware, operating systems, gaming devices, mobile devices or digital disc formats” (349); political, where you can express yourself and be heard; figurative, a ground or foundation for an opportunity and action; and lastly architectural: “in its earliest appearances, the word appeared as two words – ‘platte fourme’ or a variation thereof – a clear emphasis on the physical shape” (350). This description suggests that the platform is “a progressive and egalitarian arrangement, promising to support those who stand upon it” (Gillespie 350), which suits most of the social media platforms descriptions such as Facebook, Twitter, Instagram, Pinterest, YouTube, where users are the main source of generated content. This content, which is uploaded online by the users, becomes datafied, where datafication, according to Mayer-Schönberger and Cukier, is “transformation of social action into online quantified data” (qtd. in van Dijck 198). According to IBM 2012 Report “Big Data is any type of data—structured and

unstructured data such as text, sensor data, audio, video, click streams, log files and more. New insights are found when analyzing these data types together” (qtd. in Andrejevic “The Big Data Divide” 1676). Boyd and Crawford explain Big Data as a phenomenon that includes cultural,

(17)

17

technological and research aspects, where technologies “gather, analyze, link, and compare large data sets”; analysis “make economic, social, technical, and legal claims”; and mythology “generates insights that were previously impossible, with the aura of truth, objectivity, and accuracy” (“Critical Questions for Big Data” 663). Another definition of Big Data, from a business perspective, states that data is “increasingly framed as an economic asset of critical importance, a commodity on a par with scarce natural resources” (Rotella). Nevertheless, the “bigness” of Big Data is not about the large quantities of it, but about new ways of relating these big data sets with each other with methods such as algorithms, which are “forging unpredictable relationships between data collected at different times and places and for different purposes” (Metcalf, Keller, and boyd 4). Some of the researchers suggest that all data is raw. However, Bowker disagrees and claims that “data are always already “cooked” and never entirely “raw” (qtd. in Gitelman and Jackson 2), because “all databases are richly contextual, with specific temporalities, spatialities, and materialities that require critical attention” (Metcalf, Keller, and boyd 5). Since data is already generated (Manovich The Language of New Media 224) according to the needs of the project by pre-selection, filtering and

categorization, it cannot be raw anymore and should be considered “cooked” and interpreted or framed in other words (Gitelman 5-6). Nevertheless, framing does not have a bad meaning, but needs to be clearly explained - why and how it was classified and what are the limitations - within the project. It seems like anything can become a part of a data set today, because the development of new technologies with constant advancement of the web made it easier to collect data, but at the same time made it harder to analyze because of its variety and volume. As Manovich stated “we can follow imagination, opinions, ideas, and feelings of hundreds of millions of people. [...] And we don’t need to ask their permission to do this, since they themselves encourage us to do by making all these data public” (“Trending” 1).The outcome depends on who is performing the analysis, social scientists, information engineers, or the corporation, on the methods and algorithms they are using, and how much they translate, interpret, process, transform and clean, aggregate and separate collected data.

All this data is not accessible for “ordinary” people and researchers (and even corporations),

because only the platform itself has full access to all of it and even then, the collected data is limited. These companies have a restricted access to the data that they gather within their platforms, which can be partially accessible for payment. Thus, the platforms themselves and other companies, who can afford paying money for the data, have the possibility to design and produce research. The ones, who cannot afford paying the fee, or have different access rights, cannot perform such research or even cannot test and judge the research that has all the access (boyd and Crawford “Six

(18)

18

Provocations for Big Data” 12). Nevertheless, this continuous collection of the data “promises to have tremendous impacts on economics, policing, security, science, education, policy, governance, health care, public health, and much more” (Metcalf, Keller, and boyd 3), by offering it with new generated insights. But again, there is a question how it is going to be used, since according to Metcalf, Keller and boyd’s report, there is a common tendency in apophenia, which means “seeing patterns where none actually exist, simply because enormous quantities of data can offer

connections that radiate in all directions” (4). The researchers need to imagine their data according to the needs and standards of the project before they are able to collect, analyze and interpret it, which makes them to have an interpretive base before the data collection (Gitelman and Jackson 3). Also, it is worth to point out that it is highly unlikely to be able to download the exact same database even with the same settings, because data collection is so dynamic. But bigger sets of data do not mean that it is better (Cormode, Krishnamurthy, and Willinger). Information that is collected from the personal accounts might be fake (people put “funny” religious beliefs, falsify their working experiences or tag themselves in locations where they are not at that moment), created by bots (see Twitter chapter) or be abandoned since the user does not update his/her information or left the page. Therefore, such information, or more precisely - noise, can be misinterpreted and used as a part of the collected data. Another issue, according to Kwak et al., the big success of social media

platforms created many problems including privacy issues and “large scale information diffusion” (599), and violations within “an analysis of users’ pleasures or of their exploitation” (van Dijck 18). For example, a collaboration study between Facebook and researchers from Cornell and University of California, which secretly scrapped and manipulated Facebook feeds of 689,003 users showed that the users’ moods are affected by the postings of their friends. The experiment filtered Facebook users’ news feed comments, postings of videos and images of their friends according to positive or negative sentiment, and stated that the expressed emotions of Facebook friends are affecting our own (Facebook users) moods14. This experiment caused a big discussion among users of the platform, lawyers and politicians, and news media outlets about the ethics and morality of such experiment and privacy issues, where personal data was easily manipulated by the platform itself and third-party participants - academics (Booth).

But what does make this social media platforms actually platforms? According to Anne Helmond, for web page/software program to become a platform, it “needs to provide an interface

14

Kramer, Adam D. I., Jamie E. Guillory, and Jeffrey T. Hancock. “Experimental Evidence of Massive-Scale Emotional Contagion through Social Networks.” Proceedings of the National Academy of Sciences 111.24 (2014): 8788–8790.

(19)

19

that allows for its (re)programming” (35), and by this software interface she understands API, Application Program Interface, as “an API is an interface provided by an application that lets users interact with or respond to data or service requests from another program, other applications, or Web sites. APIs facilitate data exchange between applications, allow the creation of new applications, and form the foundation for the ‘Web as a platform’ concept” (Murugesan 36). By allowing its

programmability through APIs, social media networks transform themselves into social media platforms (Helmond 35), where social logins system, “like” and “share”, “tweet” and “retweet” buttons of these platforms offer further integration of these platforms into the web through the third-parties and applications (Helmond 3). In common understanding, social media is limited to the biggest online platforms such as Facebook, Twitter, Instagram and YouTube. However, according to Aichner and Jacob, it includes a much more extended list of social media platforms (258). For example Wikipedia, Amazon or even World of Warcraft are considered social media platforms. Social media platforms are rapidly growing and changing with Web 2.0 continuous development alongside. For example, at the time of this thesis some of those previously mentioned social media platforms have:

● YouTube (2005) has over a billion users (“Statistics – YouTube”);

● Flickr (2004) hosts 1 million of photos shared per day (Etherington) with a total amount of over 10 billion images on the website (Stadlen) posted by more than 112 million

photographers (Roth);

● Instagram (2010) announced that had 400 million users in December 2014 (“Instagram Blog | Celebrating a Community of 400 Million”), who are sharing more than 80 millions of photos and videos per day (“Instagram Blog | A New Look for Instagram”);

● Facebook (2004) has 1.65 billion monthly active users as of March 31, 2016 (“Company Info | Facebook Newsroom”).

All these numbers present datafied content, which is growing on a daily basis for each of these platforms. One million of daily photos from Flickr, 80 million photos from Instagram, 1.65 billion users’ clicks, posts and shares on Facebook become captured and analyzed and datafied. Big Data gives possibilities to social media platforms and social spaces to become quantifiable, but “working with Big Data is still subjective, and what it quantifies does not necessarily have a closer claim on objective truth – particularly when considering messages from social media sites” (boyd and Crawford “Six Provocations for Big Data” 4). Datafication is a “sense-making process”, where sense-making process “refers to processes of organizing using the technology of language (e.g., labeling and categorizing for instance) to identify and regularize memories into plausible

(20)

20

explanations and whole narratives” (Lycett 383). By capturing, analyzing and monitoring, different organizations/researchers try to create their own patterns and make sense of the data that they

capture, even try to predict trends, events, moods, crimes or stock market prices. Predictive analytics is one of the methods for analyzing Big Data, since “data are widely considered to be a driver of better decision making and improved proﬁtability” (Waller and Fawcett 77). The next subchapter will focus on discussing predictive analytics and sentiment analysis.

3.2 Predictive analytics

“The goal is not to understand the world but to correlate its elements”

(Andrejevic Infoglut 87).

After capturing and collecting the data, there is a need of methods and tools to analyze and research it and investigate what it presents. Predictive analytics, which is one of these methods, is a “technology that learns from experience (data) to predict the future behavior of individuals in order to drive better decisions” (Siegel 11). Various studies have been done using predictive analytics as a part of their methodology, trying to predict emotions and attitudes among society, health issues or outcomes of events such as presidential elections. Predictive analytics do not provide assurance, but “rather, they distribute probabilities across populations and scenarios” (Andrejevic Infoglut 54), based on the previously analyzed patterns from historical data. Predictive analytics is helpful with analysis, interpretation and prediction of continuously growing data overload (Arvidsson and Bonini 167) and becomes a valuable research method for companies and their marketing research

campaigns or for researchers, who are trying to analyze society’s offline and online behavior in certain situations. If a positive pattern of consumers behavior is found in historical data about a product, then the brand is encouraged to repeat the details of this previous marketing campaign and by doing so it is promoting “a predictable form of affective pattern (a way of acting, experiencing and relating) across a wide variety of situations and in ways that can be subject to measurement” (Arvidsson “The Logic of the Brand” 19). In order to perform predictive analytics, the researcher needs to use tools such as data mining and/or sentiment analysis. Data mining “refers to the

automated discovery of "unknown patterns and trends" in data, and predictive analytics establishes “the durability of these patterns over time, in neither case is there a promise to explain these patterns to users” (Andrejevic Infoglut 58-59). Sentiment analysis translates the sentiment, which can be an emotional response or users’ opinions into computer recognizable data (Ibid., 17). Sentiment

(21)

21

analysis is an important part of predictive analytics for this thesis, since it focuses on the sentiments of the users, and will be discussed more in detail following this subchapter.

Despite the usefulness of predictive analytics, there are several limitations and data mining, which according to the author of the thesis might heavily affect the outcome of the analysis and also subsequent research. The first issue is clearly expressed in the Andrejevic’s quote (see

aforementioned quote of Andrejevic Infoglut 87) that describes the current usage of predictive analytics. It is more important to correlate two (or more) things together, rather than understanding their interrelationships and reasons of the correlations. Additionally, the predictive analytics are available only to a limited group of people, since they require collection, filtering and analysis of its big quantities of data, that only highly costly processing power and specific knowledge can actually get the probabilistic forecast results (Andrejevic Infoglut 50). Finally, the researchers and marketers by doing data mining and predictive analytics just look for the desired correlation, without giving any explanations why and how this correlation or prediction developed. Even if data mining discovers and indicates the pattern, previously unknown, there might be no explanation of it (Ibid., 58-59). Further development of tools, such as data mining, where sentiment analysis and opinion mining techniques are used, allows researchers to measure emotional experience of users (Cambria et al.149). For this purpose sentiment analysis is used, which is an important topic to discuss within the framework of this thesis, since sentiment analysis is part of value extraction (Kennedy 439).

3.3 Sentiment analysis

As already mentioned, one of the approaches is sentiment analysis, or opinion mining in other words, where the term “sentiment”15

links to “the automatic analysis of evaluative text and tracking of the predictive judgements” (Pang and Lee, 10). Here, the analysis itself is based “within the area of natural language processing, and can be defined as the computational treatment of opinions, feelings and subjectivity texts” (Martínez-Cámara et al. 2). Sentiment analysis is commonly used within various fields, such as studies of product (Ding and Liu) and movie (Pang and Lee) reviews, online reactions to events (Kim et al.). The goals of sentiment analysis is to continuously collect data as much and as fast as possible (Andrejevic “The Work that Affective Economics Does” 610), but not read and understand every single emotional statement, because it

15

Another definition of sentiment states that it is “an attitude based on a feeling, this emotive element

(22)

22

does not bring any value by itself , but only when it is combined with others. For companies and corporations this is a great opportunity to learn more about feelings, opinions and attitude of their consumers to their company, brands and products. Thus, they unendingly invest more and more money into (marketing) tools, which are able to scrape and collect sentiments of their consumers from various (social) media platforms. That gives them the opportunity not only to analyze current tendencies, but also predict future possible reactions and attitudes (Cambria 102). The aim of sentiment analysis and data mining is to modulate negative sentiment as a variable in accordance to maximizing users’ emotional investment and engagement. Thus, sentiment analysis and data mining, as parts of predictive analytics, are both pre-emptive and productive (Andrejevic Infoglut 95). Hearn sees a problem in that, by stating that this leads to monetization of users’ feelings and extraction of value, and thus being just one more tool of capitalism for generation of value (422) for big corporations, markets and businesses, but not users themselves. Another scholar believes that sentiment analysis and predictive analytics can lead to social discrimination, since they change ‘‘individual profiles into individual evaluations’’ (Turow 6). The actual profiles of people are being scraped and analyzed, and needed information is being collected, while other is not, and by these algorithms companies create their idea about their customer. As previously mentioned, some of the research is focused on analyzing only simplistic negative/positive sentiments, which can fail to bring up in-depth information and analysis of the issue that is being investigated. At the same time, simple sentiment analysis algorithms (negative/positive) will not be able to capture and correctly interpret human’s irony and sarcasm and emotions’ complex and affective qualities (Kennedy 438), which can bring a lot of useful data for research (Wright).

Social media platforms and especially microblogging such as Twitter are gaining more interest among academia and commercial researchers. According to Hu et al., emotions and

sentiment are important aspects of our social life and it is normal for people to express their feelings publicly (537-538). Like in speech, the written content on these platforms, holds not only

information, but also emotional meaning (Alm, Roth, and Sproat 579) by using certain attitudinal adjectives. By using rich and partly available public data from Twitter, it is possible to perform fast and cheap data sentiment analysis and opinion mining for real-time events for commercial,

governmental and/or academic purposes (Pak and Paroubek 1320).To be able to perform sentiment analysis of collected data sets, there is a need of a classification system based on various criteria such as ranking, scaling, labeling and categorization (Pang and Lee 23). However, from a practical perspective, sentiment classification of microblogging data is more complicated due to its noisiness

(23)

23

(unreliable information created by fake accounts, bots et cetera), embeddedness, connectivity between two or more (social) media platforms, with other social media platforms and excessive amount of it (Hu et al. 545). Whenever sentiment analysis is done there is a chance of some errors or limitations within the research. One of the limitations, which is extensively discussed, is the issue of privacy, where the data gathered from users is freely used by third parties without their permission. Twitter and other platforms allow other parties, such as big corporations and researchers, to collect data sets based on specific queries for additional payment. Lastly, there is concern about the possibility of manipulation of the data by third parties, such as companies which create “sock puppets” (Pang and Lee 91) - online identities - to promote their products, something that can be easily seen as immense amount of bots on Twitter. Therefore, when sentiment analysis is used as a research method and classification as a part of it, there is the need of high control and regulation of data before its analysis. Twitter is used as “a thermometer to measure feverish symptoms of crowds reacting to social or natural events” by datafying people’s emotions, opinions and sentiments in their tweets. This is a common belief among researchers who think these sentiments are not filtered or manipulated by social media platforms themselves (van Dijck 199).

This chapter tried to discuss and argue on platforms, big data, datafication and its methods and tools in data capturing. It is true that the growing amount of Big Data creates new possibilities for getting extensive information and data about human behavior and emotions. Nevertheless, due to its inaccessibility or limited private allowance, the trustworthiness and quality of research is always under question. Additionally, generalization of society based on collective Big Data, does not give a real representation of current issues, questions or behaviors of individuals. Thus, when a research performed is based on the analysis of Big Data through tools such as data mining and sentiment analysis, there is need of special attention to such methodology. Moreover, it is also important to pay attention to social media platform rules regarding accessibility of data sets when the research is performed. The next chapter will discuss Twitter, a social media platform, which is commonly used for data collection within the research community.

4. TWITTER

This is the last chapter of the theoretical framework for this thesis. The chapter will discuss the appearance and development of Twitter, the role of emotions and values on it, and data

(24)

24

discussed in the previous chapter. This chapter will investigate the opportunities and pitfalls within Twitter research. I will argue that Twitter, as a social media platform, can offer possibilities for sentiment analysis and predictive analytics research, but this research will be limited for reasons including restricted access to APIs and the decreasing popularity of Twitter.

4.1 What is Twitter?

“Twitter is your window to the world” (“Twitter About”) Twitter was founded in 2006 and is a microblogging service, “which allows users to

exchange small elements of content such as short sentences, individual images, or video links” (Kaplan and Haenlein 106), where users can create their private/public accounts, post messages (tweets) limited to 140 characters, follow others and get their own followers as well, without

previously knowing each other. The mission of Twitter is “to give everyone the power to create and share ideas and information instantly, without barriers” (“Company | About”). At the moment of writing this thesis, Twitter has 310 million active users with one billion unique visits on a monthly basis to sites with embedded Tweets (see Image 2).

Image 2: A screenshot from Twitter’s About Company webpage taken on 12-05-2016 at https://about.twitter.com/company

According to Richard Rogers, Twitter passed three stages of evolution since its creation, where the first stage can be described as bland, phatic and shallow (357), since the users and the platform did not have its own “culture”; the second stage was affected by a switch in the tagline of Twitter from “What are you doing” to “What’s happening” suggesting more a reporting style, rather than self-expression (Tate); and last third stage is the current state of Twitter, which is still in the process of developing and seen as a platform for event-following and news updates research among

(25)

25

academia (363). Since users are so strictly limited with their messages’ length, several common practices were developed and spread first on Twitter and later to other social media platforms as well. These include the use of the symbol @ as a tag for identifying user within a tweet or # (hashtag) a metadata tag, which helps users to search and unite within one “topic” easier. The hashtag plays a separate role as an additional way of collecting and analyzing Twitter information since they are searchable by Twitter’s APIs and firehose, which is the only way to get all requested real-time data from Twitter for prepaid fees (Milsom). Creation of hashtags is not moderated and a new hashtag can be suggested and used by anyone on Twitter. This leads to the uncontrolled rise of hashtags, which can be easily misused and misspelled by the users. Despite this fact, “processes of suggestion, imitation, and learning, as well as Twitter’s “trending topic”16

functionality promote a shared use of certain hashtags for current events, cultural expression, or engagement in ongoing conversations” (Schmidt 6). If the user decides to use a hashtag that is commonly used or if it is related to some certain event, such as US elections, a concert or a disaster, then there is a higher chances for the tweet to be seen by non-followers of this user. Such usage of common hashtags helps to create, track and coordinate the flow of information on Twitter (Bruns and Moe 17).

Although the first user of @mention is unknown, the hashtag has its initiator, who is Chris Messina. In 2007, he suggested to use the hashtag through his tweet (Halavais 36).

Image 3: An image of Chris Messina, initiator of a hashtag usage on Twitter, tweet taken at http://bit.ly/1TDMF9l

Finally, the users can retweet (RT) each other’s tweets, which encourages for content spreadability, as well as hashtags, where spreadability refers to “the potential — both technical and cultural — for

16

“Trending topics” appeared on Twitter in late 2008 and became “a feature that enabled users to group posts together by topic by articulating certain words or phrases prefixed with a hashtag sign (#)” (van Dijck 71), which users could use and actively participate in making something trending or passively just follow these topics through the “trending sidebar” on Twitter.

(26)

26

audiences to share content for their own purposes, sometimes with the permission of rights holders, sometimes against their wishes” (Jenkins, Ford, and Green 3), along Twitter (Kwak et al. 591). Since users are limited with the amount of possible characters in their tweets, they also use a lot of abbreviations such as OMG (Oh My God) or LOL (Laughing out loud) etc. or acronyms such as FB (Facebook), EM (Email), or TT (Translated Tweet). They don’t use articles or intentionally miss vowels (Gouws et al. 22) on Twitter. The use of abbreviations and acronyms has become so widespread, that special dedicated Twitter vocabulary webpages17 were created with examples and explanations. If the user wants to share a link, special URL shortening websites can be used to save up characters (Tao et al. 197). At the same time, users sometimes use repeated vowels or consonants to point out their level of impressionability or to better express their emotions (Ibid., 197). Because of the ability to share anything online, some think and criticize that Twitter is not more than a “phatic communication”18

channel with pointless messages and useless information (Weller et al. xxx) that is constantly shared by its users. However, the usage of hashtags, @mentions and RT allow the users to create “micro-content” (Dash) or “nanostories” (Wasik “And Then There's this”), that can be filtered via social connections or shared popular words, as hashtags or trending topics on Twitter (Schmidt 6). These nanostories do not necessarily contain only personal information or opinions, but can be reports of real-time events experiences or expressions of emotions about a certain issue/happening. As for example, the first information about Virginia's 5.8 magnitude earthquake appeared on Twitter and was sent by people who had just experienced it at the moment and location (Hotz); the death of Michael Jackson (and many other famous artists and celebrities) was widely discussed on Twitter, where people expressed their emotions and thoughts about it; the US airway crush in the Hudson river was first mentioned and photo captured by a Twitter user (Johnston and Marrone); an ordinary citizen in Pakistan, accidentally live tweeted the US raid that killed Osama bin Laden (McCullagh) or the Arab Spring protests in 2011, where social media platforms, including Twitter, were tools for organizing and mobilizing people during these protests and sharing the information about what was happening with the outside world (Hempel). These protests and the ones that appeared after (London Riots19, Occupy Wall Street20 and protests in

17

http://www.abbreviations.com/acronyms/TWITTER

18

Malinowski explained phatic communication as “they [words in phatic communication] they are neither the result of intellectual reflection, nor do they necessarily arouse reflection in the listener. Once again we may say that language does not function here as a means of transmission of thought] (315). Thus, phatic communication can be considered as a chatter, which “serves to establish bonds of personal union between people brought together by the mere need of companionship and does not serve any purpose of communicating ideas” (Ibid., 316).

19

Guardian article about London Riots 2011 http://bit.ly/24G64My

20

(27)

27

Russia in 201221) developed “a new form of political and social organizing, of “hyper-networked protests, revolts, and riots” (Wasik “#Riot”). Nevertheless, not only hazardous events and

interpersonal messages are live tweeted, but also other news, political debates, entertainment updates and brands and businesses marketing campaigns, because Twitter, due to its 140 character limitation, is actually closer to oral synchronous communication, even though it is written

(Puschmann et al. 429). That encourages people and organizations to participate live and be more active in the discussions and postings. There are not only accounts of people and companies, but also bot accounts, who actively participate in generating content on Twitter.

Twitter bots include spambots, chatbots or brand promotional bots.Originally, Twitter was designed only for humans and human-to-human communication. As one of the creators of the social media platform said “Twitter is about humans connecting with each other; and often in ways that they couldn’t otherwise” (Mowbray 183). According to Quartz article, there are 23 million active users that are not human on Twitter (Seward), which is 23 million bots out of 271 million monthly active users in 2014. First bots (software robots) appeared since computer creation, based on ideas of Alan Turing algorithms, an algorithm for having a conversation with a human being, in 1950s (Ferrara et al. 1). Some of Twitter bots just automatically collect content on Twitter or from other websites by searching for the keywords and share them on their Twitter timelines, while others perform tasks of brands’ and companies’ customer care services as chatbots (Ibid., 2). Bots can be fully automated and post tweets, retweet and add followers or start following others by themselves, or they can be semi-automated, where the author of the bot can take control over it whenever he/she wants. The same goes for the content of the tweets, which can be prepared and scheduled or

completely automatically algorithmically generated (Mowbray 183-184). There are various types of bots on Twitter, which are mostly commonly used for marketing purposes, but there can also be bots for news, weather forecasts, social functions, earthquake information, parody accounts, stock prices information and others. Some of the bots automatically post their content on Twitter, while other, social bots actually interact with real human beings by auto generating content based on the

previously sent tweets. Sometimes they can be easily distinguished from the human account, while others are attempt to hide and pretend that they are real people. For example, a bot named Jason Thorton22 pretends to be a founder of a startup company and regularly tweets, retweets and even replies to others. One thing that can identify this account as a bot account is that all his tweets contain a word “love” and they are actually not his tweets, but tweets that the bot stole from others

21

Guardian article about Bolotnya square protests in 2012 against Putin http://bit.ly/24G653c

(28)

28

and posted as his (Kincaid). Another example is that of Cynk Technology Corp, a company that had one employee, no product, no assets and $0.00 revenue on the market. It managed within a week to get $6 billion market value with penny stocks23. It is believed that this was fraud achieved by creating buzz on Twitter through multiply accounts and bots (see Image 4), which promoted Cynk Technologies Corp and thus increased the stock price of it (Regan). Later, it was investigated that these accounts were paid to do so (Machine).

Image 4: A screenshot of multiple Twitter accounts tweeting about Cynk Technology Corp taken at Mashable http://mashable.com/2014/07/10/cynk/#PrklzgWtKGqV

This leads to the last and least favorite type of bots: the spam bots, which are usually used for aggressive marketing promotion campaigns, but which at times also send dangerous malware virus to other users, mistreat and misinform, and usually get suspended by Twitter because they break Twitter’s Terms of Service24

(Mowbray 185) and spam the users of the platform. If they are not spam bots, bots are commonly considered to be harmless, but sometimes they might cause issues by retweeting content that has not been verified, as happened during the Boston bombings, when news bots started spreading rumors, fake and unchecked information25 causing false acquisitions (Ferrara et al. 2). This kind of content, created and spread by bots, might affect quality and trustworthiness of

23

Penny stock “generally refers to a security issued by a very small company that trades at less than $5 per share” (“Penny Stock Rules”).

24_{The Twitter Rules do not allow: a) malware and phishing, where the link or malicious content which is intended to}

cause harm to user’s software or hardware is published within a tweet; b) spam - automatically following/unfollowing large amounts of accounts; posting mainly links and no other (personal) information; spam complaints; posting duplicate content on one or multiple accounts; creation of false or misleading content; randomly and aggressively liking,

retweeting or following; if you continuously post others content as yours (“The Twitter Rules”)

25

Examples of various fake information and rumors that were shared within minutes after the attack http://bit.ly/261fvwb

#iamsohappy: A perspective on emotional content on Twitter and value creation