DO YOU HEAR US TWEET?

(1)

DO YOU HEAR US TWEET?

An analysis of the use of Twitter by government

and news agencies during disasters.

Januari 2015 Simon A. Boer

Supervisor: Dr. Leonie Bosveld-De Smet Masterscriptie Computercommunicatie

(2)

Abstract

The increased utilization of mobile internet for communication during disaster situations elevates the importance of social media during these events. Specifically, microblogs like Twitter play an increasingly important role in the organization of distributing information and organizing relief aid. This study explores the use of the social medium Twitter by government and news media institutions during disasters in order to establish how the user orientation of their Tweets has adjusted before and after Twitter’s boom in popularity. Detecting this change is done by comparing the communication of institutions and non-institutional users during the 2009 Fort Hood shooting and the 2012 Aurora theater shooting. For both of these shootings, the activity of local, regional and national governmental and news media Twitter accounts and a sample of all non-institutional Twitter activity is gathered for an 8-day period following the event. In order to discover the user orientation of these collected tweets, the analysis has been done in two stages. First off, the sense-making functions of the tweets was annotated to find out the purpose of the Twitter message using a classification scheme consisting of information-, opinion-, action-, emotion-, and other-related functions. Secondly, the user engagement of Twitter messages is established by quantitatively looking at the properties of the Tweets. The engagement factors that affect user engagement that are researched are tweet length, use of hashtags, use of URLs, time of day and the use of retweet requests. The results of the study show that institutions did change their Twitter behavior between the two shootings, as they started using more effective user engagement tactics. However, institutions show little change in the sense-making functions that they use on Twitter: an overwhelming emphasis is on information sharing. This contrasts with an increased interest in other functions like opinion- and emotion-related among non-institutional users. This disconnect shows that institutions are not adopting a relational approach to communicating with the public and, through that, missing an opportunity to use the social in social media to develop trust for a long-term relationship with their audience.

(3)

2.4.1 Sense-making 16 2.4.2 User Engagement 18 3 Method 23 3.1 Research Questions 23 3.1.1 Hypothesis 25 3.2 The Disasters 25 3.3 Data Selection 26 3.3.1 Collection Method 26 3.3.2 Non-Institutional Corpora 27 3.3.3 Institutional Corpora 27 3.4 Operationalization 28 3.4.1 Sense-Making Functions 28 3.4.2 User Engagement 30 4 Results 32

4.1 Collecting the data 32

4.1.1 Non-Institutional Data Collection Results 32

4.1.2 Institutional Data Collection Results 33

4.2 Sense-Making Function Results 34

4.2.1 Testing the Classification Scheme 34

4.2.2 Non-Institutional 35

4.2.3 Institutional 40

4.3 User Engagement Results 45

4.3.1 Testing Engagement Factors 45

4.3.2 Engagement Factors Analysis 50

(4)

(5)

1 Introduction

In early 2014, a heavy snowstorm hit the Canadian province of New Brunswick, it left thousands of people snowed in and devastating the electrical grid. Locals found themselves in this crisis situation without electricity and were unable to power their televisions, radios or laptops to get informed on how the situation was being resolved. One of the only means of communication that was still available was through smartphones. Using this tool, social media like Twitter

became an important part in obtaining information and organizing help for those people that were stranded1_{. In this situation, the importance of social media in the aftermath of disasters nowadays}

became very clear. However, a notable player that was missing on Twitter during this disaster was the New Brunswick Emergency Measures Organization (EMO), a local governmental organization that coordinates provincial response operations during emergencies and who as such was on the front-line of organizing help. The EMO did not send out a single Twitter message from their account during the aftermath of the storm, disappointing people following their feed hoping to find some information on how the situation will be resolved. In contrast, the local electric power utility was very active on Twitter, trying to organize solutions for those people that were stranded without electricity: in the period between December 22 and January 2, the company posted 243 Tweets related to solving storm-related issues. This is a case in which using social media as a serious tool of communication can make the difference for a large group of people to help deal with the situation. People are relying increasingly on the networking capabilities of social media in order to organize relief efforts themselves following disasters. This self-organization was also the case during the 2008 Sichuan earthquake in China, which resulted in tens of thousands of casualties and many more people that were injured, made homeless or otherwise affected. During this large-scale disaster, authorities’ efforts in coordinating relief and organizing information exchange proved very difficult. As a result, as was the case during the storm in Canada, internet quickly became the main platform “for people to share information, express feelings and opinions, and exchange mutual support.” (Qu, Wu and Wang, 2009, p. 1) As the availability of smartphones, and those containing mobile internet specifically still increases, it is important to see what can be done to improve the way in which interaction through social media during disasters is realized.

1.1 Research Question

The examples from the previous paragraphs show the importance of the interaction following disasters, which is the reason why this research focuses on disaster communication. While there are a lot of different groups that communicate with each other following disasters, this research will focus on governments and news media in specific. The choice for these two groups is because they are among the most influential sources on social media. Many people look to their governments to come up with an answer to disasters, as most coordinated responses go through governmental bodies. This makes the government instrumental in dealing with the consequences of disasters. News media have a less direct involvement with the aftermath of disasters, as they do not perform any relief efforts like search and rescue. However, they are the main source

(6)

for updates on the situation for a lot of people. Their influence makes the news media’s role in disaster communication very influential, and makes it interesting to research to see how they take up this role. The combination of government and news media will be addressed in this paper with the term ‘institutions’, as compared to the main body of people that will be addressed with ‘non-institutional’. As was stated in the beginning of the introduction, social media have an increasingly pivotal role in disaster communication. This is why this research will focus on one of those social media: Twitter. Since Twitter was founded in 2006, it has grown to be one of the most influential social media on the web. It is also a very fast-paced environment in which topical developments follow each other up quickly. This makes it an interesting platform for interacting in disaster situations, as events happen quickly and the response needs to be swift. This research will therefore look at the communication on Twitter as an example to see how institutions adapt to using social media as a whole. There are many ways in which the use of Twitter by institutions can be discussed, however, in this paper the communication will be examined on the use of user orientation. This is a specifically interesting topic, since orientating utterances on twitter to the users has an impact on the reach of any messages that are posted. It is this reach that is important in a disaster situation, as usually a collective effort is needed to resolve the situation. Combining these notions leads to the following question that forms the foundation of this research: Has the user orientation of institutions on Twitter during disasters changed with the

popularization of the medium?

1.2 Importance of this research

This paper aims to add to the field of crisis communication by exploring the practices of online institutional disaster communication. This is done by applying the theories of sense-making and user engagement to the way in which institutions communicate on Twitter. By examining the way in which institutions have adjusted their orientation towards citizens online, it becomes clear whether or not this transformation in communication is happening rapid enough, if at all. The results of this comparison gives an insight in the way in which institutions deal with emerging technologies and how fast they adapt new ways of communicating. These conclusions can also be an incentive for both governmental as well as news media institutions to get their message across to the public more effectively. By addressing their role in the sense-making process as well as the level of user engagement that institutions utilize, this paper can be used to improve the best practices of Twitter use during disasters, which leads to the ultimate goal of all disaster research: to make sure that disasters are dealt with adequately in order to save time, resources and lives.

1.3 Outline of Report

(7)

communicate during disasters. This theoretical approach to communicating during disasters has to take place on a platform on which institutions can interact with the public. For that reason, the next section discusses the different media that can be used, and focuses on why social media and Twitter in specific can be a very useful tool during this process. The way in which institutions make use of social media presently is discussed in the following section. The final section of the theory chapter explores the theory behind user orientation and gives a definition on the key aspects of user orientation that are researched in this study: sense-making and user engagement.

The third chapter of this paper discusses the method that is used in finding an answer to the research question. This starts off by looking at this research question and narrowing it down to a handful of sub-questions that form the basis of the analysis. As a part of the methodology, two disasters are researched. These disasters, two shootings, are introduced in section 3.2. After selecting the disasters, the method for selecting the Twitter accounts that will be used is explained. Finally, the operationalization of the user orientation is described, which shows the main

methodology for the execution of the research.

In chapter 4 the results of the research are explained. The first step in this process is the result of the selection of Twitter accounts, and the Tweets that have been collected from these accounts. This is followed by the results of the analysis of the user orientation by these Twitter accounts. First off, the sense-making functions classification that was custom made for this study is tested, after which it is put into practice. Similarly, the second part of this section on user engagement factors starts off by testing out the factors and their actual effect on engaging users, before applying them to the collected data.

Chapter 5 describes the conclusions that can be made following these results. The

(8)

2 Theory

The following chapter will focus on the key fields that the research question touches upon: disasters, Twitter and user orientation. In the first part of this chapter the term disaster is defined in order to narrow down a term that has several different interpretations and synonyms. The characteristics of disasters and the way in which people and institutions communicate during disasters are also explored in order to give a basis from which to compare the institutional communication in the analysis with. The second part of the chapter focuses on the research field of social media, with an emphasis on microblogs like Twitter. In the social media section, the importance of the rise of social media is explored, and an explanation of the way in which it is used during disasters gives context to the research. The third part of the theory section explains what user orientation on Twitter looks like, namely, as sense-making and user engagement. The paragraph on sense-making shows the way in which people interact during disasters to collectively deal with different aspects like information gathering and opinion sharing. This will be shown to be relative to how institutions reach the Twitter community. The other side of user orientation is that of user engagement. This deals with the more direct ways in which users are reached over Twitter, such as Tweet length and the time of day during which the communication takes place.

2.1 Dealing With Disasters

In order to establish the meaning of the term ‘disaster’, it has to be seen in the context of a group of events often called ‘emergencies’. These emergencies come in many forms, but what is it specifically that differentiates a disaster from other emergencies? Regtvoort and Siepel make an attempt to clarify the differences by giving their definition of some of the terms that are used for emergency situations in their book on risk and crisis communication: (2007, p. 47, translated from Dutch)

- calamity: an unexpected event that can cause severe damage;

- crisis: an emergency that seriously upsets the functioning of a system; - incident: an occurrence that interrupts the ordinary course of things; - accident: something that went wrong and that (usually) causes damage; - disaster: a large, massive accident that affects many

(9)

constrain the harmful effects. A crisis is described as a situation in which a vital interest of society is affected or likely to be affected. (2007, p. 47) These descriptions might make the distinction slightly more clear, they still leave room for interpretation, since any situation in which the lives and health of many persons are threatened will probably be a situation in which a vital interest of society is affected. Schwartz (2010, p. 5) combines definitions of multiple sources when trying to get to a definition of a disaster, and summarizes that a disaster contains at least the following conditions:

- they disrupt the normal everyday life led by a community; - they cause damages (to environment, economy and human life); - they are beyond the capacity of the community to take care of;

- resources outside of the community are needed in order to help in its recovery.

The same elements of damages and a coordinated effort are reflected here as in the definition given by Regtvoort and Siepel. Because this research focuses on local as well as national response during crises, the emergencies that we will be looking at are severe enough to get extensive national coverage and response. This means the disaster is too big for the local community to take care of, and outside help is needed during the response. These characteristics of a disaster fit with the specific type of events that we will be looking at: mass shootings.

For crises that draw national attention, the Federal Emergency Management Agency uses the, what appears to be more general, term ‘emergency’. This term refers to ‘any occasion or instance for which, in the determination of the President, Federal assistance is needed to supplement State and local efforts and capabilities to save lives and to protect property and public health and safety, or to lessen or avert the threat of a catastrophe in any part of the United States. (FEMA, 2013, p. 1-2) This adds another distinction between ‘regular’ crises that require the coordination of several local or state efforts, and ‘disasters’, in which federal assistance is required.

Even though the disasters researched in this report might qualify under FEMA law as being emergencies, the term ‘disaster’ will be used to address the events, as this corresponds with the jargon used in the field of crisis communication during disasters. (Hughes, Palen, Sutton, Liu, & Vieweg, 2008) (Qu, Huang, Zhang, & Zhang, 2011) (Vieweg, Palen, Liu, Hughes, & Sutton, 2008)

2.1.1 Phases of Disasters

Disasters cannot be defined as a singular event. It is a process in which the situation changes constantly. Even though the initial event might be short: like an earthquake, its ramifications can last long after the initial event has transpired. It is therefore important to distinguish the following phases during disasters, as defined by Regtvoort and Siepel (2007): 1) Arise, 2) Observation, 3) Alarm, 4) First Response, 5) Scaling up, 6) Stabilization, 7) Ending the disaster situation, 8) Aftercare.2

These temporal stages are important to recognize and identify, as the necessary response both by aid workers and communication departments during these phases can differ. During the first-response phase, for example, it might be necessary to communicate about relief efforts, while

(10)

during the aftercare phase communication could help reconnect lost family members. However, it is important to note that the stages are not as clearly distinguishable as may appear from this scale. Certain events can be traced back to a particular point in time, like when the crisis is first observed or when first responders arrive on scene. Other events have less clear markers and are harder to identify, as phases like ending the disaster situation and aftercare will in most situations have an area of overlap. When reconstructing the timeline of a disaster, it is therefore necessary to identify events with a clear timestamp to map the events of the situation, and use broader timeframes to establish periods that are less clearly defined.

However, as is the case with defining disasters, there is no single answer on what phases occur during a disaster. For their exploration of online interaction during the Virginia Tech massacre, for example, Palen, Vieweg, Sutton, Liu, & Hughes make use of the following arrangement of so-called ‘socio-temporal’ stages: (2007)

Stage 0: PRE-DISASTER

State of social system preceding point of impact Stage 1: WARNING

Precautionary activity includes consultation with members of own social network

Stage 2: THREAT

Perception of change of conditions that prompts survival action Stage 3: IMPACT

Stage of “holding on” where recognition shifts from individual to community affect and involvement

Stage 4: INVENTORY

Individual takes stock, and begins to move into a collective inventory of what happened

Stage 5: RESCUE

Spontaneous, local, unorganized extrication and first aid; some preventive measures

Stage 6: REMEDY

Organized and professional relief arrive; medical care, preventive and security measures present

Stage 7: RECOVERY

Individual rehabilitation and readjustment; community restoration of property; organizational preventative measures

against recurrence; community evaluation

Figure 1: The eight socio-temporal stages of Disaster according to Powell (1954) and Dynes (1970).

(11)

of this focus on the community’s response, the model does not have a set point at which the disaster occurs or starts, rather it shows the point at which the disaster is perceived (the threat stage). Especially during shootings the community view is more often relevant: the actual shooting is witnessed by only a few people. However, once it gets picked up by a news outlet or is shared online the community will be informed. This is different from for example an earthquake, in which a large affected area instantly knows about the occurrence through personal experience. The community perspective that the model uses also makes it more suitable for identifying stages on social media, mainly because most interaction on social media is made by non-institutional users. This makes the interaction on Twitter reactive to the crisis developments, a result of which is that citizens can usually report on disasters faster than institutions can. These aspects make the model used by Palen et al. more relevant for researching the disasters in this paper.

2.1.2 Views on Communicating During Disasters

Disasters are situations in which effective communication can make the difference between mitigating the situation and having it escalate. (Joshi, 2012) However, what the best approach is to deal with the aftermath of a disaster situation is not clear cut. In the field of disaster communication, a distinction can be made into preventative and responsive action. If communication is used preventatively it is called risk management. Risk management deals with informing the public on the disasters that could possibly occur, the steps taken to prevent these disasters and to inform about correct behavior during disasters. The aim of managing the risks is to limit the amount of possible damage during a disaster. (Regtvoort & Siepel, 2007, p. 94, 95) Basically, risk communication is to prepare the public for the event of a crisis or disaster. However, in this research we are interested in looking at communicating in response to disaster situations. This is called crisis communication and concerns the communication during and following a crisis. There are two dominating views on how to deal with crises as they emerge, the relational view and the reputation management view. The relational view assumes that crises are “episodes in the ongoing relationship between an organization and its stakeholders” (Coombs, 2000, p. 73). This means that it is important to have good relationships with stakeholders in order to develop effective responses to crises. In practice, this means that approaches like being honest pay off in the long run, as this will improve the relationship with stakeholders in the run-up to a possible next disaster. The reputation management view states that the reputation that an organization has before a crisis will protect the organization during the crisis. Coombs and Holladay (2006) found that using this strategy limits reputational damage because of the ‘halo as shield’ effect (p. 134). This implies that stakeholders “may be inclined to discount or ignore the negative information about the organization” (p. 125) because of its positive previous reputation. Even though this last approach might work for certain companies that already have a high reputation, in the case of government institutions this reputation is structurally low.3_{This means that in order to build a}

more trusting relationship with their citizens, it is important for governments to adopt a relational approach to their communication. Part of this approach is communicating on platforms that are widely used among citizens, which include social media.

(12)

When putting the relational view of crisis communication in practice, it is vital during the early stages of a disaster to “recognize the crisis as soon as possible and make responses to the crisis so as to mitigate negative impact” (Liang, 2012, p. 249) When the crisis is over, communication has to focus on “learn[ing] from the crisis, monitor[ing] the public’s impression, and mak[ing] sure that the crisis is truly over.” (Liang, 2012, p. 249) To improve the way in which agencies apply crisis communication, Veil et al. have established a list of best practices: (2011, p. 111, 112)

- Establish risk and crisis management policies and process approaches. - Plan pre-event logistics.

- Partner with the public.

- Listen to the public’s concerns and understand the audience. - Communicate with honesty, candor, and openness.

- Collaborate and coordinate with credible sources. - Meet the needs of the media and remain accessible. - Communicate with compassion, concern, and empathy. - Accept uncertainty and ambiguity.

- Provide messages of self-efficacy.

These can be used by organizations and government agencies to streamline their risk and crisis communication to be as effective as possible. A few of these are of particular interest for this study on the use of social media during crises, specifically where the interaction with the public is concerned. For example, ‘partner with the public’ is about being open in sharing information on the crisis and the risks that people face. This is a two-way street, in which both the agencies as well as the public contribute information to the conversation. ‘Communicate with compassion, concern, and empathy’ refers to the emotional response that people having during and after crises. Utilizing social media, organizations can open up the exchange of emotions and humanize the crisis response. The last practice, ‘provide messages of self-efficacy’, relates to the actions people might want to take in response to a crisis. Through helping others by donating or taking action, people can retake a sense of control that a crisis or disaster takes away. These examples show that communication that is interactive towards information sharing, but that also includes helping people with other functions, like aiding the physical actions of people or relating to their feelings, helps build a relationship with more trust that is more durable.

2.2 Medium Choice

(13)

a higher message sharing rate. Also, when used in combination with blog posts, Twitter scored high on post-crisis reputation. “People who had read the Tweet were less likely to boycott the organization or to talk negatively about the organization than people who had read the blog or the newspaper article.” (p. 25) This suggests that using Twitter during crisis response is a valuable asset for an organization or agency, next to the still relevant traditional media of, for example, newspapers (Utz, Schultz, & Glocka, 2013). It is therefore important that these groups make adequate use of the tool that is Twitter.

2.2.1 Social Media During Disasters

Throughout the process of disaster communication, social media can play a significant role. During disasters, a speedy exchange of information is important for all actors to stay up-to-date with the latest developments. With the information and communication technology available, however, keeping up to date has become easier than ever. Social media, and Twitter in particular have become an increasingly popular means for emergency communications. (Vieweg, Palen, Liu, Hughes, & Sutton, 2008, p. 1079)

During the first half of the 2000s, the increasing availability of internet gave rise to what got called web 2.0. This meant that the web is used “as a platform whereby content and applications are no longer created and published by individuals, but instead are continuously modified by all users in a participatory and collaborative fashion.” (Kaplan & Haenlein, 2011, p. 61) The most prominent way in which this became visible was the rise of social media, most popularly Facebook and Twitter. These websites offer ways in which users can interact and create their own content, which in turn generates great data to research human behavior and improve on the possibilities of the social media as a subset of computing science. These new possibilities are not only in the social sphere of “friendly chatter and individual expression”, but increasingly includes more serious fields such as “health care, energy sustainability, environmental conservation, disaster response, and community safety.” (Schneiderman, Preece, & Pirolli, 2011)

Research into social media becomes especially useful during disasters. One example of this is disaster detection. Using the vast amount of utterances on social media, algorithms can monitor the conversations looking for patterns of words that can indicate a disaster is occurring. An example of this is Yin et al’s paper on their ESA (Emergency Situtation Awareness) system that uses burst detection that sends out an alert when an unexpected burst is detected (2012, p. 2). This can then be combined with other information on the location, nature and severity of the disaster that is derived from (meta)data contained in the social media utterances. Detecting emergencies in time is not only crucial for first responders, but also for stakeholders that need to proactively respond to these emergencies. In this and other ways, social media can be used not only for social chit-chat, but for aiding real-world processes like disasters.

2.2.2 The Rise of Twitter

Out of all of the social media, perhaps the most interesting one to research is Twitter. The reasons for this are that Twitter has a large user base, among the top five of all social media4_{, and}

the fact that all the data that is being shared on the website is publically accessible. This makes for a large amount of social data that is at the disposal of the scientific community. Twitter is a

(14)

microblog that got started in 2006 as a way for people to easily send short status updates to one another via mobile phones. It quickly gained traction in the years following, with a great increase in 2008 of 1,382% in one year. (Armstrong and Gao, 2010, p. 222) This is when the public really started to pick up on it, as can be seen in the following chart showing the number of Twitter accounts over the years.

Figure 2: Growth in the amount of Twitter users between 2006 and 2012 by D. Steven White5_.

This clearly shows the increase in popularity between 2008 and 2012. Between 2012 and 2014 the rate of growth of users has decreased, with a current user count of 883 million6_{. This}

growth period influenced the choice of the disasters in this research, with one disaster on the onset and one at the end of this growth spurt so Twitter practices can be compared between these events. From its start in North America, Twitter first grew to users in Europe and Asia and then to the rest of the world. (Java et al., 2007) Even though Twitter started out as a service on which people interacted socially on topics such as personal activities and popular culture, it has grown to include more ‘adult’ topics.

2.3 Governments and News Agencies on Social Media

Governments around the world, both locally and nationally, are increasingly using ICT under the title of ‘e-government’ with the aim of getting “greater efficiency, deeper transparency, higher service quality, and more engaged citizen participation.” (Sandoval-Almazan & Gil-Garcia, 2012) In all, e-governance can improve many processes that a (local) government deals with on a daily basis. Up to now this has been mostly through streamlining the back-office processes by automating costumer requests through websites, online human resourcing and encouraging democracy through online voting systems. However, with the increasing availability of mobile devices and social networks, governments are also starting to explore the possibilities of online interaction with citizens, businesses and other actors: ‘Government 2.0’. The use of these social media is specifically aimed at enhancing the transparency and increasing the effectiveness of public services. (Henman, 2013, p. 1415) Aalberts and Kreijveld make a distinction between four ways in which governments interact with their citizens online (2011, p. 37):

5 http://dstevenwhite.com/2013/02/09/social-media-growth-2006-to-2012/ 6

(15)

- Informing citizens on policies - Getting opinions from citizens

- Give citizens the option to ask questions - Getting new ideas from citizens

However, in a survey among Dutch cities that they conducted, results showed that the governments were mostly using ‘old fashioned’ online interactions like forums and regular websites. The adoption of social media was not widespread. Even though most cities did have a Twitter account, the rates of interaction with citizens was low and they were mostly using for sharing information rather than starting a dialogue. They see the adoption of social media therefore to be invaluable to the future of government communication, however, they also see some pitfalls that need to be addressed before it can be implemented effectively. For one, it is unclear to what extent the voices and opinions on social media are representative for the population as a whole, making it very difficult to use these voices in the policy making process. Furthermore, by giving people the option to voice their opinion, the expectation that their opinion will be used will make for more disappointment when this is not possible. This is part of the reason why governments are slow to implement social media and stick to using the internet for informing their citizens.

Twitter specifically can be an important tool for governments to collaborate with citizens. Research by Waters and Williams has shown that governments use the medium not only as a means of one-way communication to inform and educate, but that many researched Tweets tried to invoke a dialogue. It is this mixture of communication approaches that were most effective in serving the organization’s interests as well as the audience’s expectations. (2011, p. 359) This research focused on looking at Tweets from 60 randomly selected American national government agencies like the National Aeronautics and Space Administration and the Department of Homeland Security. These Tweets were then evaluated for their functions of public relations. This showed that the most used model was that of public information: informing the public on the agency’s own situation happened in 85% of the Tweets. The second highest occurring role though, was that of the two-way symmetrical model, which deals with communication between the organization and its Twitter followers. This model was used in 42% of the researched Tweets. (Waters and Williams, 2011, p. 358) This dialogue encourages citizen-government collaboration on social media, as was the case during the 2011 riots in England. Panagiotopoulos et al. discovered that during these crises, Twitter’s conversational features helped enabling different forms of collaboration, like “cleaning the streets, disproving rumors and identifying suspects.” (2014, p. 356) However, the adaptation of Twitter for this purpose by UK local government authorities or councils seemed to be spontaneous rather than planned. The authors therefore recommend government officers to review their readiness for these events and come up with a plan for the use of social media.

(16)

media does not, however, incorporate the full potential that social media offers. The very nature of social media allows for one-to-many and many-to-many communication: news organizations could as well as merely informing their readers start a dialogue with them. In line with this, scholars “suggest being authentic and brief, providing useful material, creating community gradually, listening more than talking, participating more than spreading, continuously linking information, being very selective with sources, not being overly didactic, providing novelty, and being patient.” (Herrera, 2012, p. 81) Basically, this means that news organizations need to have a more personal touch to their social media content in order to engage their users and to build a community around their online presence, with the goal of creating a longer lasting presence on social media. (Messner, Linke and Eford, 2011)

Armstrong and Gao have researched the use of social media by news institutions more in-depth. (2010) Looking at the topics that were tweeted by news agencies, they found that there is a difference in the topics that local and regional news media Tweet about, than those of the national Twitter accounts. The local and regional accounts most frequently tweeted about crime, whereas national news media Tweet most about public affairs/politics. A reason that they give for this discrepancy is the target audience: Tweeting about crime is closely related to the community, whereas politics and public affairs related more to the general population. Given that the

disasters that are researched in this paper are of a criminal nature, the expectation is that local and regional news outlets will have more Tweets on the topic of the disaster than the national accounts.

2.4 User Orientation During Disasters

In order to research how the governmental and news media institutions connect to the Twitter public, the user orientation of these institutions is of interest. This user orientation is in this paper divided up in the concepts of sense-making and user engagement. Firstly, the term sense-making and how this is used in disaster situations is further explained in this section. Secondly, the concept of user engagement will be discussed in the next section.

2.4.1 Sense-making

(17)

the ambiguity of the information that they are receiving (why is this disaster happening?), so they search for meaning. (Weick & Sutcliffe, 2005, p. 419)

All of the above functions that occur online following a disaster are commonly identified as sense-making functions. The interaction takes place on many different platforms, like websites (Fox, Rainie, & Madden, 2002), blogs (Macias, Hilyard, & Freimuth, 2009), forums (Qu, Wu, & Wang, 2009), and most prominently on social media. It has been shown that the freedom that people get to organize on social media gives rise to both small-scale, concentrated, reliable information collection (Vieweg, Palen, Liu, Hughes, & Sutton, 2008), as well as a platform for large-scale information interaction (Starbird, Palen, Hughes, & Vieweg, 2010). In their study of these sense-making functions, researchers have operationalized it in different (disaster-dependent) ways. However, the different classifications they have come up with all have a common core of four main functions: information, opinion, action and emotion. These functions seem to be closely related to John Searle’s speech act theory. This theory identifies the illocutionary act of an utterance (or it’s intended meaning) and classifies it as one of the following: assertives, directives, commissives, expressives and declaratives. (Searle, 1975) Assertives and commissives are statements that commit the speaker to the truth of the expressed proposition or commit them to an action, which is in correlation with the information-related functions found in the classifications for sense-making. Directives relate to the action that the hearer is requested or commanded to take, which correlates to the action-related sense-making functions. Expressives are utterences in which the speaker expresses their attitudes and emotions, and correlate with the opinion and emotion functions. Declaratives relate to real-world actions that are taken in accordance with the utterance, which cannot (yet) exist in an online environment.

One classification scheme relating to the sense-making functions after disasters has been proposed by Qu, Wu, & Wang (2009). In their study, they used the sense-making classification scheme as seen in see Figure 3 to establish the functions that forum threads had in the community’s sense-making in the aftermath of the 2008 Sichuan Earthquake.

Category Description

Information-related

Sharing Providing factual information to the community Seeking Soliciting specific factual information from other

community members Gathering and

Integrating Gathering and integrating factual information from other community members to form a “knowledgebase” collectively on a specific issue or topic

Opinion-related

Criticizing Criticizing the government, organizations, or individuals

Other opinion Appraising; providing comments without explicit appraising or criticizing; or seeking opinions from others

Action-related

(18)

Individual participation Declaring actions performed by, or to be performed by the author

Coordinating Organizing actions among a group of forum users Emotion-related

Expressing Expressing personal feelings such as anxiety, sadness, anger, proud, etc.

Emotional support Demonstrating social and emotional support, including mourning, blessing, comforting, encouraging and expressing concerns for victims. Community building

Moderation-related Posts by the moderators; the forum users’ request for moderation; and users’ responses to moderation activities

Norm shaping Forum users’ attempts to regulate others’ behaviors in the forum

Sense-making Making connections among pieces of information in order to understand the occurrence of the earthquake or to interpret other related events Anti-Social

Flaming Using insulting or hostile language to personally attack a person or a group

Trolling Posting irrelevant or off-topic messages with the intention of baiting community members into an emotional response

Off-topic Messages irrelevant to the earthquake

Figure 3: Qu, Wu and Wang’s forum thread classification scheme of disaster reactions.

This scheme was constructed on data that was gathered on forum discussions, and as such includes descriptions specifically designed for a forum thread study. It includes the four commonly used functions: information, opinion, action, emotion. Next to that, three categories that did not directly pertain to disaster communication in particular have been added: community building, anti-social and off-topic. For each function the different ways in which they are identified in the data have been noted, including a description on the form they take on. Even though this scheme was produced with the aim of researching forum activity, the same main functions apply on microblogging services like Twitter. With that in mind, this scheme was used as the basis for the annotation in this research, adapting it to the particular disasters and functions found in the data that was collected for the research reported on in this thesis.

2.4.2 User Engagement

The use of these functions can only be effective in the sense-making efforts of the

(19)

the message. Doing so will increase the chance of having them retweeting the Tweet, which will expose the message to a large audience: the followers of the person that retweets it will have it show up in their timeline. No matter the amount of followers that an account has, it has been shown that once a Tweet starts spreading via retweets it will reach, on average, at least 1000 more recipients. Moreover, this number increases significantly when the initial account has more than a thousand followers. (Kwak, Lee, Park, & Moon, 2010) Especially during a disaster it can be important to reach a large amount of people with the correct instructions or information.

It is important to think about engaging users on Twitter, since research has shown that it is the social medium with among the lowest rates of user engagement. The following graph shows a breakdown of the user engagement on different social media.

Figure 4: User Engagement with brands among different social media in 2014 according to research by Nate Elliott7_.

The low level of user engagement by default means that extra effort has to be put into engaging users with content. Many strategies can be used to get users to engage en have them retweet a message. Research into the effects of these strategic factors has found that usage of content features like hashtags and URLs correlates with retweetability, while mentioning users does not. Contextual features that increase the amount of retweets are the number of followers and followed as well as the age of the account. The number of past Tweets from an account however does not seem to influence the amount of retweets. (Suh, Hong, Pirolli, & Chi, 2010) Other research done by Dan Zarella, a Social and Viral Marketing Scientist, has also found correlations in an non-scientific setting between retweet rates and usage of certain words, syllables per word, readability, novel word usage, part-of-speech percentages, RID Content Type (Emotional, Primordial, Conceptual), asking to retweet, content types, gender, hour of day, depth of reproduction, and use of ‘rich’ media. (2009)

To find if government and news organizations during disasters are Tweeting in such a manner that the engagement will be optimal, a number of the factors will be analyzed quantitatively in this research. Not all of the factors could be examined in this paper: because we are looking at how Twitter was used in 2009 and 2012, it is impossible to retrieve most of the account information during that period. Therefore we cannot look at the amount of followers, favorites and gender of a Twitter account. Also, the limited data returned by Topsy also does not include rich media like images or video’s, making it impossible to research the use of these. Moreover, because of the limited resources available in this research, a number of factors would prove too labor intensive to explore. These include word use, part-of-speech tags and RID content types. However, a number

(20)

of factors are possible and relevant in the scope of this research. These are the following: using the correct length of the Twitter message, using hashtags, mentioning user names, Tweeting at the right time, and asking for users to retweet. What these factors represent, and how they influence the user engagement is explained in the following paragraphs.

A note has to be made on the fact that these engagement factors have, for the most part, not been properly scientifically researched. Most of the information on these factors are from web experts that give lists of best practices in order to get users engaged on social media.

Tweet Length

The length of a twitter message can range from 0 to the famous Twitter limit of 140 characters. This means that the message has to be conveyed in a short and concise manner. However, shorter is not always better, as multiple sources, including Twitter’s own best practices page, indicate. The ideal length for a Twitter message is said to be in the range of 71-100

characters (excluding links and mentions), with an ideal length of 100 characters, as this length yields the highest percentage of retweets. The reason for this is that this particular length leaves enough characters for someone that is retweeting it to add their own comments, while still having enough space to make the original message compelling. The following chart by Track Social8_shows

the distribution of the amount of retweets by length of the Twitter message in their research of 50,321 brand-related Tweets in 2012.

Figure 5: Graph by Track Social on the retweet rates of Twitter messages of different lengths shows a spike for those in the 71-100 range.

This figure shows that the shorter Tweets get a significantly lower amount of retweets than longer Tweets, with a peak at the 71-100 message length group. The increase gained by using Twitter messages of this length came out in multiple analyses to be between 17 and 20 percent compared to shorter messages. This suggests that making Twitter messages of this length can have a positive effect on the engagement with users.

Hashtags and URLs

A widespread strategy for getting other users engaged with your Tweets is the use of hashtags and URLs. (Zangerle, Gassler, & Specht, 2011) Hashtags are user-defined index terms that link

(21)

messages that are related to the same topics and events together, increasing the exposure in information retrieval and navigation (Yang, Sun , Zhang, & Mei, 2012), which occur as words starting with the special pound sign character, or ‘hashtag’ (#) that named the practice. The first hashtagged word on Twitter was used in 2007, and the implementation gained popularity since.9

However, while using hashtags is a popular recommendation for increasing the engagement with users, recent research suggests that using more than 2 hashtags will show a decrease in engagement as opposed to one or two hashtags.10_{URLs, like hashtags, have a special function}

within a Tweet. They give the reader the opportunity to find more information than just the 140 characters that are allowed, thereby making the message potentially more interesting. Boyd et al. have shown a correlation with higher retweet rates, hashtags and URLs were overrepresented compared to a random sample of Tweets. (2010, p. 4) It has to be noted that although there is a correlation overall, it does matter which hashtags and links are provided in the Tweet: not all of them create an equal amount of interactivity. (Suh, Hong, Pirolli, & Chi, 2010, p. 182) While using hashtags is a popular recommendation for increasing the engagement with users, recent research suggests that using more than 2 hashtags will show a decrease in engagement as opposed to one or two hashtags.11

Time of Day

Because there are Twitter users in all time zones, it is important to know where your target audience lives and to Tweet at the right times. A separate tool has even been developed to make a strategy on which time of day would fit best with your followers on Twitter: Tweriod. This shows how important it is to catch your readers when they are most active on the medium. But even if you know where your readers live, what is the best time of the day to get them involved with the messages? Dan Zarrella has explored the times of day when Twitter messages are most retweeted, as shown in Figure 6 below. (2009)

9 https://www.hashtags.org/platforms/twitter/history-of-hashtags/

10 Neil Patel states that “Tweets that use more than two hashtags have a 17% drop in

engagement” http://www.socialmediaexaminer.com/twitter-tactics-to-increase-engagement/ 11 Neil Patel states that “Tweets that use more than two hashtags have a 17% drop in

(22)

Figure 6

The graph shows that most engagement on Twitter is during the afternoon and into the evening, with a peak between 12pm and 2pm EST. This distribution means that in order to get the most engagement with users, Twitter messages should be sent during that time period for the appropriate time zone. The disasters that are analyzed in this report are in different time zones, yet have gained national attention. Because of this, it is important for institutions to not only reach the local people that were affected, but also add to the national narrative surrounding the disasters. It is therefore interesting to see how the Twitter behavior is distributed for the local as well as the most popular time zone on twitter in the U.S.: EST.

Requesting Retweets

A seemingly obvious factor in retweet rate is the Twitter user asking his or her followers to retweet the message. By specifically requesting the action of retweeting, users are reminded of the option and will be more inclined to perform the action. Studies have shown that including this call to action can lead to up to 43 times12_{increase in retweets. Even though adding a retweet request}

takes up valuable real estate in the 140 character limit, this would serve as a great way to get users engaged, and as such could be very helpful for institutions during disasters.

(23)

3 Method

This chapter on the methodology of the research starts out with a narrowing down of the research question into usable sub-questions. The theory from the previous chapter is then used to formulate a hypothesis on the research question. Section 3.2 gives a brief description of the disaster-cases that are researched in this paper: the two mass shootings. These cases are then used to set up a selection method for the data that is necessary to analyze. The operationalization of the variables included in user orientation, sense-making and user engagement, is done in the final two paragraphs of this chapter.

3.1 Research Questions

Narrowing the research question down to tangible sub-questions is the topic of this first section of the method chapter. The main research question of this paper is aimed at trying to analyze the behavior of institutions during disasters:

Has the user orientation of institutions on Twitter during disasters changed with the popularization of the medium?

Since it is practically impossible to research the institutions’ response in all disasters that have occurred, a selection is made of two disasters in the form of shootings. The two shootings that are analyzed, the 2009 Fort Hood shooting and the 2012 Aurora theater shooting, are picked on their similarities for best comparison. The dates on which they occur also coincide with the period just before and after the boom in popularity of the medium Twitter. The details of the shootings will be further discussed in the following chapter: 3.2. In order to establish whether or not government and news media institutions have actually adjusted their way of communicating on Twitter to the public between these two shootings, the definition of user orientation from the theory section is used to formulate sub-questions. This definition was the separation of user orientation into sense-making and user engagement. This research will therefore be divided up by focusing on the sense-making aspects on one hand, and on the user engagement aspects on the other hand.

The first part of user orientation that will be researched is the use of sense-making functions. As was stated in the literature, these functions signify what the topic of conversation on Twitter is about at any given time. It is important for institutions to tune into these functions and not be Tweeting about irrelevant topics so they will relate to non-institutional Twitter users. Therefore, a comparison can be made between the two disasters to see if the use of these functions has changed. This leads to the following question:

(24)

The following sub questions are formulated that have to be answered in order to be able to make a statement about the use of sense-making functions:

- How do institutional and non-institutional Twitter accounts use sense-making functions during the shootings?

- Has the use of sense-making functions by institutional and non-institutional Twitter accounts changed between 2009 and 2012, and how?

- How does the change in using sense-making functions by institutional Twitter accounts relate to that of non-institutional accounts?

The first of these questions will look at each shooting specifically, and identify the way in which both non-institutional users, governments and news media use the functions throughout the shootings. The second question builds on this, by asking if the groups are using different functions in 2012 than in 2009. Because this measurement of change is done for both institutional and non-institutional users, an answer to the last question can be derived, namely, how these changed relate to one another. Once these changes have been mapped for both the non-institutional and the institutional Tweets, a conclusion will be made about whether or not the institutions adjusted their use of the sense-making functions, and if so, whether these adjustments correspond with how the use of sense-making by non-institutional users changed.

The second aspect of user orientation that shows to what degree institutions get their

message across to the Twitter public is by optimizing their user engagement. By engaging users with Twitter messages, it becomes easier to spread information or to come in contact with the stakeholders (citizens). In order to measure whether or not institutions are employing the right methods to engage users, the following question is formulated:

- Have governments and news media adjusted their user engagement during disasters between 2009 and 2012, and how?

As is clear from the theory on user engagement (chapter 2.4.2), there are a lot of factors that influence the engagement of users with a specific Twitter message. The factors that are identified in the theory as being of interest to this research are used to address the question of user

engagement. The following questions aim to seek an answer to whether or not the institutions have adjusted their user engagement to that of non-institutional Twitter users.

- How did institutions and non-institutional Twitter accounts use the factors to increase user engagement during the disasters?

- Have institutions and non-institutional Twitter accounts changed the way in which they use the factors to increase user engagement during the disasters, and how?

- How does the change in using engagement factors by institutional Twitter accounts relate to that of non-institutional accounts?

(25)

over time, and finally see how this relates between institutional and non-institutional users. The use of user engagement factors is analyzed by setting up a quantitative research that looks at all the available Twitter data. By measuring the data and looking for significant differences between institutions and non-institutional users, and between 2009 and 2012, the sub questions can be answered. The full operationalization for answering all questions is made in chapter 3.4.

3.1.1 Hypothesis

Now that the questions for this research are set up, an informed hypothesis can be made on what the outcome of the analysis will be. Namely, if institutions have adjusted their user-orientated between 2009 and 2012, then the use of sense-making functions and user engagement factors of institutions will have gotten closer to that of non-institutional users over time. In 2009, Twitter just started to become mainstream, which means that people had not yet had time to develop elaborate strategies for using the medium properly. However, during the three years the popularity increased exponentially. The result of this growth is that not only the amount of users but also the amount of articles and best practices about using Twitter increased. It can be expected that institutions have picked up on this and further professionalized their online social media presence in this period. If reaching the public is an important aspect of this online presence, the assertion can be made that institutions will have become more user-oriented. For that reason, the expectation is that the user orientation did in fact go up in the specified period, which would show as having a more similar utilization of sense-making functions with non-institutional users after three years, along with better use of the user engagement factors.

3.2 The Disasters

The best method to find out whether or not institutions have changed their behavior, is by assessing this online behavior at a point in time before and after the popularization of Twitter. As is mentioned before, two disasters have been selected for this comparison: the 2009 Fort Hood shooting and the 2012 Aurora theater shooting. The selection of shootings in particular is made because these are the only type of disaster that, regrettably, occurs regularly enough to find disasters of comparable severity in the same country and during the selected years. The upcoming paragraphs include a more detailed description of the events during and following the disasters.

The first shooting that is the subject of this research is the shooting on the Fort Hood military base that occurred on November 5, 2009. Nidal Malik Hasan, a U.S. Army major and psychiatrist shot and killed 13 people and injured 31 more.13_{Even though the shooting only}

directly affected some of the local population, the event quickly gained national news and political attention.

The second shooting during which the Twitter activities are analyzed is the shooting at a movie theater in Aurora, Colorado on July 20 of 2012. In this instance, the perpetrator was James Holmes, who killed 12 people and injured another 58.14_{In contrast to the Fort Hood shooting,}

this incident affected mostly civilians, whereas the victims of the Fort Hood shooting were mostly 13

http://www.reuters.com/article/2013/08/28/us-usa-crime-forthood-idUSBRE97Q11A20130828

(26)

military personnel. The victims being civilian caused for an even greater amount of national attention than was the case in 2009.

3.3 Data Selection

In order to study the Twitter behavior of government and news media institutions, their activity on Twitter during both of the shooting incidents will be retrieved. Beside these institutional Tweets, another corpus containing non-institutional Tweets will be collected in order to compare the activities of the institutions to the Twitter conversation as a whole. The selection process that is used to retrieve data for both the non-institutional and the institutional corpora is described in the following sections. Firstly, the collection method for this data is explained in section 3.3.1. Secondly, the selection process for non-institutional Tweets (section 3.3.2) and institutional Tweets (section 3.3.3) is described.

3.3.1 Collection Method

(27)

3.3.2 Non-Institutional Corpora

The first dataset that will be collected is that of messages of non-institutional Twitter users. For the purpose of comparing the changes between the two shootings, two sets of Twitter messages have been collected: one of the online interactions during the Fort Hood shooting, and one of the Aurora shooting. For the collection of these sets, there will be no selection on the source of the Tweets, instead a one-term query will be used to catch a broad selection of Tweets (the queries that were used are “Aurora” and “Fort Hood” respectively). The timeframe in which the data will be collected, is from the start of the particular disaster (on a whole hour before the first event), until the same time 8 days later, 192 hours after the occurrence. This timeframe is chosen because previous research on blog functions has shown that this period is long enough to account for all shifts in functional trends that take place after a disaster. (Qu, Wu, & Wang, 2009) (Heverin & Zach, 2011)

To be able to detect hourly differences in the use of functions in these datasets, it is necessary to collect enough Tweets per hour to accommodate for a random selection of Tweets per hour. For this purpose at least ten Tweets per hour will have to be annotated to be able to view temporal differences between hours. In order to make a random selection of ten Tweets per hour, at least 100 Tweets will be collected per hour to make sampling possible. Moreover, because preliminary tests showed that a large (inconsistent) proportion of the Tweets that are collected through Topsy are duplicates (in some examples up to 30%), 150 Tweets per hour will be withdrawn in order to accommodate for at least 100 results for every hour.

3.3.3 Institutional Corpora

In order to acquire Twitter data from government and news organizations, a selection will be made of the institutions to be included in the data. Because the disasters in this research are mainly of a local or regional nature - a shooting only affects a few people directly, and their families and friends indirectly - the local and regional institutions will be included. The size of the disasters, however, also sparked national attention. For this reason several national governmental and news media institutions will also be included.

(28)

3.4 Operationalization

The process through which the variables sense-making functions and user-engagement will be measured in the data is explained in this chapter. Firstly, the measuring of sense-making functions will be done through a custom classification scheme based on those used by other researchers. Secondly, the operationalization of the quantitative user engagement factors is relatively straight-forward, since this will mostly consist of measuring the number of occurrences of the factors.

3.4.1 Sense-Making Functions

(29)

Function Description 1. Information-related

Situation Update Share factual information on what is happening with the disaster and the political reactions to the events.

Technology Inform about other outlets to find information on the developments. Neutral Quote Quoting a news source or other third person in order to inform on

developments.

Announcement Announcing a future event, like a television program.

Personal Sharing personal experiences or activities relating to the disaster. 2. Opinion-related

Announced Explicit announcement of an opinion.

Opinionated hashtags Using hashtags that indicate that the Tweet is opinionated, or commenting on the content of the Tweet.

Give qualification Using terms to give a qualification to a person, institution, situation etc. Criticizing

institutions Criticizing actions or policies of (governmental) organizations. Sarcasm/rhetorical

questions Using sarcasm or rhetorical questions to implicitly give an opinion. Selective correlations Presenting a personal view on a situation by illogically relating events

causally.

Loaded quotation Quoting an opinionated source. 3. Action-related

Requesting

information Asking other users to share information on the events. Asking for relief aid Asking users to donate for the victims of the disaster. Ask to share Asking other users to retweet the message.

Request online

behavior Specifically asking other users to click on a URL, or other online behavior Call for physical

action Asking users to physically help in dealing with the disaster. 4. Emotion-related

Express solidarity Expressing solidarity with first responders, families of victims or the victims themselves.

Express feeling Expressing emotions like feeling sad or regret. 5. Other

Off-topic Tweet does not relate to the disaster at hand.

Ambiguous Not enough information is given to decide what the intention of the user was.

Figure 7: Classification scheme of sense-making functions.

(30)

pertain to the disaster at hand or not. All other messages that will be annotated fit into one or more of the other categories.

The first category is that of the information-related messages. This category contains all messages that pertain to the sharing of information on the disaster and its aftermath, and the way in which information on the disaster can be obtained. The second category, opinion, contains all messages that include a personal or biased view of a situation that happen in obvious and more subtle ways. In a few cases it is very obvious that an opinion is used because this is announced, like by starting the Tweet with “opinion: “, or by using hashtags that indicate an opinion (for example, #tcot is a popular hashtag signifying a conservative political view). A less obvious

occurrence of opinions in Tweets is by using sarcasm and selective correlations, which can only be observed by considering the context. Messages in the subset of loaded quotations contain sources that advocate opinion, like opinionated blogs. The third function that is discussed during disaster is that of action. These Tweets ask something of the person reading it. In the observed data these action-related tweets had the goal of asking for information, (monetary) aid and physical action. Some actions also did not pertain directly to doing something about the disaster, but sharing the messages on Twitter or asking for other online behavior like following a link. Emotion-related functions are used in those Tweets in which people share their emotions in the form of supporting people that are touched by the disaster, or just by sharing how they feel about the situation.

Annotating the messages according to the sense-making functions is manual work, which makes establishing the functions of every collected message too much work for this thesis alone. However, this is not necessary in order to get a view of the temporal changes in the use of these functions. Instead, a sample of Tweets will be taken from the data that will be annotated and analyzed. The governmental corpora proved to be small enough (see paragraph for the results of the data collection) to be annotated in their entirety. For the other corpora a random selection of 10 Tweets per hour will be made for the first 24 hours after the disaster occurred, with the thought that most developments around the disasters themselves take place in this period. Also, for the 8-day period afterward, a 100-Tweet random sample will be taken from the daily results to be annotated in order to view the changes per day. This will result in six groups of Tweets that are taken from the respective corpora, which will then be annotated manually using the custom-made annotation scheme.

Because this research is performed by one researcher only, it raises the question of the reliability of the annotation: the results cannot be compared with those of other annotators. In order to minimize this effect, an attempt will be made to make the scheme as non-ambiguous as possible. To test how well the scheme performs, a second annotator will be used on a selection of 100 random Tweets to see how their results match with those of the main researcher. By comparing the results an inter-rater agreement level will be established showing the accuracy of the model. The results of this test can be found in chapter 4.2.1.

3.4.2 User Engagement

(31)

Message Length

In order to analyze the length of the Twitter messages used in the different corpora, they will first be processed to remove all links and username mentions (@username). These are not part of the message in the Tweet, as they only serve as adding contextual information. Also, messages that are retweets will have the retweet tags (‘RT @username’) and any text that was added by the retweeting person removed. As the original message usually follows the retweet announcement (Boyd et al, 2010, p. 3), any text before this announcement is assumed to be added by the person retweeting. Deleting this leaves just the message of the original Tweet. The clean dataset that remains will be used to count the length of the message in every Tweet. This length will be defined in the amount of characters of any kind: letters, numbers, spaces and special characters, this ratio-type data can have a value between 0 and 140. The premises will then be tested by comparing the lengths of Twitter messages from the different corpora.

Use of Hashtags and URLs

Quantifying the use of hashtags and URLs in Tweets will be a matter of scanning messages and identifying entities as hashtags or URLs, after which a message can be annotated to whether or not it contains one or more of these entities. Finding the use of hashtags and URLS will be done by parsing each twitter message using PHP regular expressions. For hashtags this is “/#[\S]+” , or any entity starting with a hashtag that starts with ‘#’, and for URLs “[\S]+\.[a-z]+\/[\S]+” and “http(\s|\S)+”, or any entity that includes any domain name followed by a folder, or starts with ‘http’. A count will be made for each Tweet stating whether (1) or not (0) a hashtag has been used, how many hashtags have been used (#), and whether (1) or not (0) a URL has been used. This information can then be used to relate to the size of a particular corpus to find differences in the use of these entities.

Timing

The timestamps that will be collected with the data are in the Unix timestamp format. This 10-digit code will be converted to show the hour of day during which the message was sent. After this the number of messages sent per hour can be easily counted and plotted. This analysis will only be done for the institutional corpora, since the corpora of non-institutional Twitter messages will be collected by scraping the amount of Tweets per hour. This means there are just as many Tweets every hour of the day, and thus cannot be used to find which time of day yielded most results.

Requesting Retweets

DO YOU HEAR US TWEET?