Employing topic modeling in the intermedia agenda-setting
theory: the interplay between the media, the corporation
and the public in Cambridge Analytica scandal
Name: Xijia Hu Student ID: 11351497 Master’s Thesis
Graduate School of Communication
Research master of Communication Science Supervisor: Sanne Kruikemeier
Date: February 01, 2019
This study explored the causal relationship between media, organizational and public agendas with the issue of Cambridge Analytica scandal. Topic modeling was the first time used in the agenda-setting theory to detect subtopics of media agenda and corporate agenda. Besides, online search behavior was regarded as the public agenda. Time-series analyses were employed to daily-level aggregated news articles (N = 313), Facebook press release (N = 62), and the popularity of Google search. Results of VAR model indicated that on the first level agenda-setting, the amount of news articles and press releases can influence the popularity of Google search about “Cambridge Analytica”. Subsequently, on the second level of agenda-setting, the use of lagged regression model demonstrated that subtopics flowed from media agenda to corporate agenda and public agenda, and from corporate agenda to public agenda.
During the last decade, Facebook has shown enormous user growth and even changed communication to some extent, leading to more interactive deliberations (Heggde & Shainesh, 2018). Users share personal activities on Facebook to their Facebook friends and even
strangers. Every day, Facebook users produce more than 500 terabytes of information (Tam, 2012). To some extent, the user generated content makes up the foundation of Facebook, and strengthens the connections between users and the platform (Heggde & Shainesh, 2018). Therefore, it is important for the company to maintain trust and encourage active content generation on Facebook.
However, on March 18th, 2018, Christopher Wiley, the CEO of Cambridge Analytica, disclosed that the personal information of more than 50 million Facebook users was used illegally to analyze their political preferences and to predict their voting intentions (Alpha.com, 2018). Cambridge Analytica was accused of taking advantage of this data to help Donald Trump with his 2016 presidential election campaign (Isaak & Hanna, 2018; Persily, 2017). As one of the top 10 stories of 2018 (Glusac, 2018), the scandal had a strong negative impact on the company: both Cambridge Analytica and its parent company shut down (Lumb, 2018), and
Facebook saw a sharp decrease in its market value (Shen, 2018). The leak of user information harmed the privacy of the public and subsequently attracted public attention. This scandal pushed many countries to amend their privacy laws and social media platforms to update their privacy rules as to protect privacy in the digital era (European Commission, 2018).
In addition to the backlash for the company that the scandal caused and the
microtargeting techniques (e.g., Tarran, 2018; Ward, 2018), it is also a particularly interesting case to explore the intermedia agenda setting effects between the media, the company (i.e., Facebook), and the public. Intermedia agenda setting theory, which is developed from traditional agenda setting theory, explains that the agenda transfers from one medium to another (Harder, Sevenans & Van Aelst, 2017; Tam, 2015). The existing studies have focused more on the interaction between two actors, either on issue transfer between different media outlets (e.g., Guo & Vargo, 2017; Vonbun, Königslöw & Schoenbach, 2016), between traditional media and social media (e.g. Rogstad, 2016), or between the media and the corporations (e.g., Kroon & van de Meer, 2018; Van Aelst & Vliegenthart, 2014). However, what happened when more than two actors joint the agenda-setting process?
Cambridge Analytica scandal presents a good case for intermedia agenda setting research between three actors (the media, the company, and the public), because these three main actors have different issue salience goals in this issue. For the media, the Cambridge Analytica scandal was highly newsworthy (Glusac, 2018). For Facebook, the scandal created a public relations crisis, which the company was expected to react quickly to, providing
information to the media and the public; because the scandal threatened the reputation and trust to the company this further affected their financial performance (Huang, 2006; Patriotta, Gond, & Schultz, 2011; Seeger, Sellnow, & Ulmer, 2003). Thus, the organization is another actor which engages in the agenda-setting process, and aims to lead the agenda to save its reputation. Equally important, the public actors also have an interest in joining the agenda setting process. As the scandal revealed the risks posed towards privacy, the public is likely to
follow up on the progress of the scandal investigation, and speak about their demands regarding privacy protection.
Hence, three actors all have the motivation to drive the agenda. Previous studies have, however, either focused on the relationship between the media and the public (Ragas, Tran, & Martin, 2014), or between the media and corporations (Kroon & van de Meer, 2018). Thus, there remains a gap to explore: When the media, the organization, and the public are all
included in the agenda setting route, which actor takes the lead? Therefore, this study aimed to fill this gap and to investigate the interplay between the media, the organization, and the public in the first-level and second-level agenda setting process, and used the Cambridge Analytica scandal as the context. Topic modeling was employed to detect the subtopics in media and corporation agendas. Time series analysis was used to explore the causal relationship between three actors.
Taken together, this study contributes to the understanding of intermedia agenda-setting theory and advances the theory by exploring the interplay between the media, the organization, and the public. The study explores the relationship between three actors in intermedia agenda setting, and deepens the understanding of how a company responded to a major scandal and their influence in a public relations crisis.
In addition, the use of topic modeling in this study expands the method by analyzing the subtopics in the newspaper and press releases. The implementation of topic modeling is a creative, new approach to the topic detection in the agenda-setting theory. Practically, this study contributes to company understanding of their position in the issue salience process, and guides them in planning campaign strategies.
Theoretical Framework
The Evolution of Agenda-Setting Theory
Since McCombs and Shaw (1972) introduced the agenda-setting theory, it has become one of the most well-developed theories in communication science. Based on interviews with voters, McCombs and Shaw (1972) found that news media can shape public attention and political reality by selecting which news coverage to make salient. Subsequently, scholars expanded on this by addressing different types of issues, media, audiences, etc. (e.g., Dearing & Rogers, 1996; McCombs, 2004; McCombs & Shaw, 1993; McCombs, Shaw, & Weaver, 1997; Wanta & Ghanem, 2007).
In the agenda-setting theory, the effect of agenda setting has been considered to be twofold— “what to think” and “how to think” (Lim, 2011). “What to think” refers to the first level of agenda building, which explores how the media selects the issue and influences the public’s opinion on it. The first level agenda setting effect refers to how the issues highlighted by the major media transfer to the secondary media (McCombs & Shaw, 1993). The emergence of this effect can be attributed to the small size of secondary media, which do not have enough
resources or journalists to obtain the information first-hand. Roberts and McCombs (1994) found that issue transfer from newspapers to television programs occurred during the 1990 Texas gubernatorial campaign.
In addition to the issue salience and transfer, attributes of issue agendas are transferred from one medium to other ones as well. For example, Kroon and van de Meer (2018) explored the reciprocal influence in affective attribute between organization agendas and tabloid agenda. Different from first-level agenda-setting influence on ‘what to think’, the second-level agenda setting describes ‘how to think’ about an issue.
Specifically, second-level agenda-setting is a social learning process by the salience of attributes of one medium has influence on these attributes of another medium (Lim, 2011).
Apart from affective and cognitive attribute, subtopics within an issue as an attribute category which proposed by Ghanem (1997) is not widely used in agenda-setting theory. This study prefers subtopics within an issue to explore the second-level agenda-setting theory. This is because an issue usually contains several subtopics. For example, in the articles of Cambridge Analytica scandal, the subcategories includes the fact issue (Aleksandr Kogan), the analysis of the impact (privacy), the role of Facebook (Mark Zuckerberg), etc (see Appendix 1.). To analyze the subtopics provides a more informative study in the issue salience (Lim, 2011).
Due to the development of online communication, the decentralized characteristic of the new media landscape has changed the way of both first-level and second-level agenda-setting route (Ragas, Tran, & Martin, 2014). Although media salience is still applicable, the traditional media has lost its dominant role in the agenda setting process (Meraz, 2009; Ragas, 2014). The public agenda, presented through digital platforms (e.g., online search behavior, social media, blog, etc.), has begun to show its influence on the agenda setting process (Harder, et al., 2017; Ragas, 2014; Ragas & Kiousis, 2010). Additionally, the existing research has also found an influence of large corporations on the media and the public (e.g., Kroon & van der Meer, 2018). In the meantime, intermedia agenda setting theory is proposed to explore the issue transference in a different medium.
Intermedia agenda setting is defined as “those instances when the media agenda is shaped by other media” (Sweetser, Golan, & Wanta, 2008, p. 199) to describe the flow of issue salience from one medium to others (Atwater et al. 1987; Harder, Sevenans, & Van Aelst, 2017). The main difference between intermedia agenda setting theory and the traditional agenda setting theory is that the former focuses on “the influence of issue attention amongst different media caused by the competitive nature of the media market” (Tam, 2015, p.118).
Most studies about intermedia agenda setting have focused on the transfer of issue salience in different mediums (e.g. Denham, 2014; Reese & Danielian, 1989). A few studies have identified the public as a supplementary actor in issue agenda development (e.g.,
Vliegenthart & Walgrave, 2008). However, the existing studies rarely explored the influence of public attention on the media or other actors in crisis communication. Therefore, this study aimed to extend the research on the relationship between the public agenda, media agenda, and corporate agenda following a trigger event.
Public Agenda and Online Search Behavior
In the social media era, the relationship between the media and the public has become complex and dynamic. Not only has news consumption changed, but now the public has the means to bypass the media entirely and select their own issues to make salient (Maurer & Holbach, 2016). By expressing public opinion and showing their attention to certain issues, the public has shown their impact on issue salience and transference. The public can now be regarded as another relevant actor in the agenda setting process.
The unilateral media-public agenda-setting route is thus broken, and there is a trend of mutual influence between the media and public agendas (Kleinnijenhuis, et al, 2015; Ragas, Tran, & Martin, 2014). Uscinski (2009) found that the public agenda (based on opinion polls) drove the media agenda (nightly network news), and made public issues, like energy or the environment, salient. The public agenda began to influence issue salience (Carroll & McCombs, 2003;Kiousis et al., 2016; Russell Neuman et al., 2014).
“Public agenda” was thus given a new meaning beyond “ranking of the relative importance of various public issues” (Dearing, 1989, p. 310). The comparison between blogs and news about political issues showed that agendas are not only driven by the media, but also by the public (Andrews & Caren, 2010; Meraz, 2009). The logic reversed from media attention driving public attention, to public attention driving media attention. The public uses these salience cues from the media to organize their own agendas, and to decide which issues and actors are most important. Over time, the set of priorities visible on the news media agenda becomes, to a considerable degree, the agenda of the public (Carroll & McCombs, 2003). At this time, the
central focus of public agenda studies is to explore how the public builds the agenda, and the consequence of an agenda driven by the public instead of the media (Andrews & Caren, 2010).
However, the public agenda is difficult to measure. Most studies measured the public agenda through experimentation or surveys (e.g. de Vreese, Boomgaarden & Semetko, 2011; Matthes, 2012; McComb & Shaw, 1972). These methods are usually time- and resource-costly (Maurer & Holbach, 2016). With the innovation of communication technologies, online search queries are regarded as an indicator of public agenda (Maurer & Holbach, 2016; Ragas et al., 2014). In a longitudinal study, Mellon (2013) found that Google Trend data can be used to find the issue salience in many fields (e.g., fuel price). Housholder, Watson, and Lorusso (2018) suggested a positive linkage between political advertisements and online search data, and also confirmed the feasibility of using online search queries to represent public attention.
Trigger Events and Corporate Agenda
A trigger event is an impressive and rare incident that brings an issue to the forefront (Birkland, 1997), which can be natural disasters or human-caused catastrophes, such as earthquakes, epidemics, scandal of a big company, or other incidents that may influence social security (Ragas et al., 2014). Generally, trigger events catch the public’s attention and cause discussions in both the media and the public (Ragas et al., 2014). According to Dearing and Rogers (1996), “agenda setting is, in some cases, an emotional reaction to certain trigger events” (p. 91).
Previous studies have tested the issue salience and transfer route after a trigger event happening. For example, Downs (1972) drew the issue-attention route of triggered events in the environmental issue. Cobb and Elder (1983) additionally found that trigger events can help an issue move to the policy agenda. Recently, more scholars have paid attention to the role of the Internet following trigger events. After the Deepwater Horizon oil spill in 2010, Ragas et al. (2014) found an interplay was present between the media agenda and the public’s online
search behavior, and Kleinnijenhuis et al. (2015) found support for the traditional agenda setting route and the effect of corporate press releases on the media agenda.
To further explore the issue-attention route, the Cambridge Analytica scandal is a perfect trigger event in the social media era. Cambridge Analytica was a British data analytics firm that focused on political campaigning during electoral processes (Ingram, 2018). In March 2018, the New York Times and other media outlets exposed Cambridge Analytica for collecting and using Facebook users’ personal data illegally (Martin, 2018). The information was collected for
academic research in 2014 and was meant to be deleted in 2015. However, the data was not deleted as promised, but rather, was used for Trump’s presidential campaign and Brexit
campaign (Ingram, 2018; Martin, 2018). Overall, Cambridge Analytica acquired nearly 87 million Facebook users’ personal data, most of which was acquired without users express permission (Martin, 2018). This scandal soon attracted the attention from the media and the public, and generated a massive news investigation and public discussion.
As a top news story in 2018 (Glusac, 2018), the Cambridge Analytica scandal was
newsworthy and attracted media attention. Further, the scandal affected Facebook’s reputation and increased the public’s concern about privacy. Therefore, this issue attracted the attention from the media, the company, and the public. As three actors all have the demands to lead the agenda, it is the perfect context for examining which could do so.
The Interplay Between the Media, the Corporation, and the Public in a Trigger Event In public relations research, previous studies focused on the reciprocal relationship
between the organization and the media, but ignored the role of the public in the agenda-setting process (Kroon & van de Meer, 2018). Regarding the interplay of the organization and the media, the studies pointed in different directions. In the case study on a Yahoo-Icahn contest, Ragas, Kim & Kiousis (2011) found that organizational information subsidies had unilateral effects on the news agenda. Similarly, in the study of BP’s performance following the Deepwater
Horizon crisis, Kleinnijenhuis et al. (2015) showed that the corporation had effects on the media agenda, but not the other way around. However, in financial (Ragas, 2013) and environmental news (Pollach, 2014), scholars have found a mutual effect between the corporation and the media. Even in a study of corporate reputation, a correlation between the company press release and the media coverage was not found (Kim, Kiousis, & Xiang, 2015). Overall, in the first level of the agenda-setting process, the relationship between the media and the corporation remains controversy.
Subsequently, public awareness is also involved in the process to build the agenda. Previous studies have supported the idea that the public agenda can influence the media agenda (e.g. Maurer & Holbach, 2016). Ragas et al., (2014) found that, in the Deepwater Horizon crisis context, there was a reciprocal relationship between media reports and online search behaviors, which gave evidence of the mutual influence between the media and public agendas.
Therefore, the relationships between the three actors (i.e., organization, media, and the public) is explored and leads to the research question:
RQ1: How does issue salience transfer between the organization, the media, and the public over time?
The study of agenda-setting theory found that media agendas shapes public perception (McComb & Shaw, 1972). Salwen and Matera (1989) found that television programs about foreign nations can influence the public’s perception about these countries. A study that examined online search behaviors and media reports about the Deepwater Horizon oil spill showed a reciprocal influence between media attention and search volume (Ragas, et al., 2014), and the increase of media reports about the trigger event can subsequently lead to increased searches for the event (Ragas et al., 2014). Besides, an empirical study of crisis communication, situated in the German bank crisis, proved media frames can be transferred to
corporate press releases (Strauß & Vliegenthart, 2017). Therefore, based on the first-level agenda-setting theory, it is proposed that:
H1: An increase in the amount of news about Cambridge Analytica will result in an increase of the Google search popularity of Cambridge Analytica.
H2: An increase in the amount of news about Cambridge Analytica will result in an increase of Facebook’s press releases.
Corporate reputation and crisis communication literature has also demonstrated that the company attention and public relation efforts can drive media and public attention (Kiousis, et al., 2016; Kiousis et al. 2007; Kleinnijenhuis, et al., 2015). Meanwhile, scholars have
discovered that a large percentage of news reports were based on corporate information subsidies (e.g. press release; Sweetser & Brown, 2008; Wanta & Ghanem, 2007). Therefore, two hypotheses are presented:
H3: An increase in the amount of Facebook’s press releases will result in an increase of the amount of news about Cambridge Analytica.
H4: An increase in the amount of Facebook’s press releases will result in an increase of the Google search popularity of Cambridge Analytica.
Media industry reports have shown that the media has paid more attention to the data from search engines and other online activities to discover news stories, and provided feedback to these online activities (Dick, 2011; Peters, 2010). Meanwhile, empirical studies showed a reciprocal relationship between the amount of news and search behaviors (Ragas & Tran, 2013). Besides, the increased amount of media and public attention directed towards the Cambridge Analytica scandal will lead Facebook to respond to these demands by releasing statements that answer questions from the media and public, and that protect its reputation. Thus, the fifth and sixth hypothesis is presented:
H5: An increase in the Google search popularity of Cambridge Analytica will result in an increase of Facebook’s press releases.
H6: An increase of the Google search popularity of Cambridge Analytica will result in an increase of news about Cambridge Analytica.
In the study of sub-topic salience, it is expected that the media agenda has the upper hand in the second-level agenda-setting process. In a study of the second-level agenda-setting
theory, the affective attribute of media agenda lead the affective attribute of organizational agenda (Kroon & van de Meer, 2018). Besides, in a study of 2008 Iowa Democratic presidential nomination, the media framing of the campaign transferred to the public framing (Heim, 2013). Therefore, it is proposed that:
H7: The media agenda drives the corporate agenda in the subtopics about the Cambridge Analytica scandal.
H8: The media agenda drives the public agenda in the subtopics about the Cambridge Analytica scandal.
Public relation studies focused on the relationship between organization press releases and the influence on the public. For example, corporations respond to a crisis and thus influence the public’s perception of their reputation (Coombs, 2007; Jin, Liu, & Austin, 2011). Householder et al. (2018) found an increasing amount of political advertising will result in an increasing online search volume. In addition, in the research of corporate reputation news on organization-public relationships supported the second-level agenda building in the substantive attribute (Kim et al., 2015). This result is expected to extend to the crisis communication.
A study about organizational and media agendas showed that, in some cases, reciprocal relationships between the two exist, but there is also a unilateral influence of the organization on the media’s agenda (Kroon & van der Meer, 2018). Therefore, it is proposed that:
H9: The corporate agenda drives the media agenda in the subtopics about the Cambridge Analytica scandal.
H10: The corporate agenda drives the public agenda in the subtopics about the Cambridge Analytica scandal.
Previous studies proposed a “reverse agenda setting” to explore after media/corporation salience an issue, if the increase of public attention would call for response from the
media/corporation (Ragas, et al., 2014). In the dissemination of subtopics, we expect
H11: The public agenda drives the media agenda in the subtopics about the Cambridge Analytica scandal reversely.
H12: The public agenda drives the corporate agenda in the subtopics about the Cambridge Analytica scandal reversely.
Method
Content Selection and Pre-Processing
Cambridge Analytica was exposed in early March 2018. To analyze the discussion of this topic among the media, the company, and the public, information from March 18th, 2018 to August 9th, 2018 was collected from three actors. After August 9th, there were few news reports and the Google search volume about Cambridge Analytica decreased.
News articles are used to determine the media agenda. LexisNexis was used to collect news reports about Facebook privacy scandal news with the “Cambridge Analytica” search query (N = 678). Then, to ensure the whole article was relevant to the scandal, only articles with a title containing the keywords “Cambridge Analytica” were selected (Kiousis, Popescu &
Mitrook, 2007; Ragas, et al., 2014). Ultimately, 313 articles were included.
Facebook press releases were used to gauge the corporate agenda. All press releases were sourced from Facebook’s official website, Facebook newsroom
(https://newsroom.fb.com/news/). A total of 62 press releases from Facebook newsroom were collected from the selected five-month period. No distinction was made regarding the relevance of the press releases to the scandal, as press release focused more on the impact of
Cambridge Analytica scandal instead of the issue itself (see the subtopics of press release in Appendix 1.).
The distribution of documents were aggregated at a daily level. Following collection, all documents were tokenized, removing stop words and punctuation, and lemmatized. Only the root of the words were kept in the documents. After this preprocessing step, the documents were prepared for topic modeling analysis. The data collection and cleaning were processed through a Python script. Subsequently, topic modeling was employed in the document of news and press releases to extract the subtopics of media agenda and corporate agenda (see Content Analyses).
The online search behavior data was collected from Google Trends
(https://trends.google.com/trends/) and used to measure the public agenda. Google Trends aggregated the amount of search requests about the keyword from Google, which is the most popular search engine in the world (Maurer & Holbach, 2016). Therefore, the data from Google Trends are assumed to be representative of the public agenda. Compared with Twitter or other social media sites, search queries in Google can reflect the public concern better as it covers a larger portion of the public.
The data from Google Trends showed the popularity of the searched keyword under the selected conditions. The range of the popularity score of each keyword is from 0-100; in the selected area and period, a score of 100 is given to the day with the most searches for the keyword (trend.google.com, 2019). The data can be requested on a daily, weekly, or monthly basis, and can then be used for comparisons with the media or corporate agendas. In this study, the search area is set as “Global” and the time period is set from March 18th, 2018 to August 9th, 2018. The popularity of each keyword was explored, and then used to represent the public attention in each news/press release subtopics. The popularity of the searched keywords was also collected on the daily level from Google Trends.
In this study, the search keywords are selected based on the news and press release topics (see Appendix 1.). Keywords which are related to the Cambridge Analytica scandal (e.g., “Robert Mercer,” the owner of Cambridge Analytica, or “Christopher Wylie,” the Cambridge Analytica employee who first revealed the scandal) were manually selected. General and ambiguous keywords (e.g., “want”) were excluded. Looking into corporate topics, the keywords are often too general and indirectly related to the Cambridge Analytica scandal; these keywords lead to broad and irrelevant search results. As such, keywords from the press releases were used alongside the phrase “Cambridge Analytica” as the search query to narrow the results. Therefore, the keywords collected from the first five news and corporate agenda topics, each, are presented in Table 1. It is worth mentioning that in the second press release topic, none of words are related to the Cambridge Analytica scandal (see Appendix 1.). Therefore, this
subtopic is considered irrelevant and the effect of the corporate agenda on the media and public agenda for this topic was not tested. After the popularity of the keywords was separately
collected, the scores of the keywords in each subtopic were averaged to evaluate the distribution of news and press release subtopics on the online search behavior.
Table 1
Search queries in Google Trends
Topic Keywords
News Topic 1 Aleksandr Kogan; Christopher Wylie
News Topic 2 Mark Zuckerberg
News Topic 3 Christopher Wylie; Brad Parscale; Cambridge Analytica; Robert Mercer
News Topic 4 Mark Zuckerberg; “Cambridge Analytica”+ privacy; Cambridge Analytica; Robert Mercer
News Topic 5 Aleksandr Kogan
Corporate Topic 1 “Cambridge Analytica” + data; “Cambridge Analytica” + information; “Cambridge Analytica” + election
Corporate Topic 2 N/A
Corporate Topic 3 “Cambridge Analytica” + information; “Cambridge Analytica” + post
Corporate Topic 4 “Cambridge Analytica” + privacy; “Cambridge Analytica” + information; “Cambridge Analytica” + data
Corporate Topic 5 “Cambridge Analytica” + election; “Cambridge Analytica” + information
Content Analyses
Topic modeling is a statistical model used to detect topics from documents. Humans determine topics by picking words from texts; the model then attempts to imitate this word-picking process and find the word composition from the topics (bag-of-word; Koltsova & Koltcov, 2013). The model assigned words to each topic with a value of probability (Koltsova & Koltcov, 2013). The probability is based on the co-occurrence of words from the bag-of-word dictionary with unusual frequencies in the text (Blei, Ng, & Jordan, 2003). Therefore, the output of the algorithm is the word and the probability of the word belonging to the topic (see Appendix 1).
Topic modeling was, thus, employed in this study to detect the five most popular topics from news reports and press releases, and then find the proportion of each topic distributed in all the articles. Afterward, the probability of topic distribution in the given documents was examined.
Latent Dirichlet Allocation (LDA) was employed as the topic modeling algorithm to extract topics from news reports and press releases. LDA was proposed by Blei et al. (2003), and is one of the most frequently used topic modeling algorithms. This algorithm is chosen because of its processing performance in terms of speed and quality when using a large amount of text, as compared to other topic modeling algorithms (Griffiths & Steyvers, 2004; Koltsova & Koltcov, 2013). In this study, the LDA was implemented using the Python Gensim package (Rehurek & Sojka, 2010). The Gensim package is based on the weighted term-frequency value, which is “the log-based term frequency-inverse document frequency (TF-IDF) weighting factor to the data prior to topic modeling has shown to be advantageous in producing diverse but
semantically coherent topics which are less likely to be represented by the same high-frequency terms” (Greene & Cross, 2017, p. 82).
The LDA model was trained by the news articles and then applied to press releases to find the probability of these news topics occurring in news and press releases. Then, the same procedure was implemented with press release articles, and applied to the news content. This process resulted in the topics from the media and Facebook press releases, and the distribution of these topics for two actors.
Based on previous literature (Chandelier et al., 2018; Robert et al., 2014), five topics for each actor were used to detect topic diversity. Each topic contains 10 words and the probability that they compose the topic.
While topic selection is objective and replicable, the interpretation of these topics relies highly upon the researcher’s knowledge about the issue. The output for news topics were all relevant to the Cambridge Analytica scandal, with topics three and four containing the company name “Analytica” specifically, and all topics containing at least one relevant person’s name.
Meanwhile, the press release topics did not indicate a strong relevance to the Cambridge Analytica scandal. Facebook did not focus on the issue itself, but rather, acknowledged the consequences or social issues which were relevant to the scandal (e.g. “privacy” and “election”).
The issue topics emerged from news and press release documents that were analyzed with LDA. This produced the distribution of issue topics that were aggregated on a daily level.
Results
Time series analysis was used to explore the time order of the three actors and to establish the causal relationship of issue transfer among them. As no clear unilateral causal relationship exists among the three actors in the theoretical framework, neither the media, corporate, nor public agenda can be considered exogenous in the analysis (Kroon & van der Meer, 2018).
When the relationships between variables are not clear, Vector Autoregressive (VAR) analysis is an appropriate way to build the multivariate time series model, and to explore the causal relationship between variables. VAR analysis aimed to evaluate the impact of the variables on each other over time (Brandt & Williams, 2007). Therefore, VAR modeling is suitable for analyzing the causal relationship between the three actors in this study. In VAR analysis, variables can be treated as both independent and dependent (Vliegenthart, 2014).
The time lag length in the model is determined by theoretical expectations and the boundary is established based on the theoretical arguments. Specifically, a maximum of 21 days is selected to allow for the fast-paced impact on news and public agendas, and a slow-paced impact on organizational agendas (Kroon & van der Meer, 2018; Vliegenthart, 2014).
Granger causality is employed to find the causal order of variables with a VAR analysis. Using the Granger causality test, it is assumed that a variable y can be predicted by the past values of y, and the prediction can be improved by adding the past values of x to the equation (Brandt & Williams, 2007). To employ this test, all variables must be stationary. Only if all
variables are stationary, can the parameter be estimated for reliable results (Vliegenthart, 2014). Therefore, the Augmented Dickey-Fuller (ADF) test was carried out to determine the stationarity of each variable. Table 2 indicated that in all instances, the null-hypothesis of non-stationarity can be rejected for the three variables.
Table 2
The ADF Results for Number of News Articles, Press Releases, and Google Trends
Variable T value p-value
Number of news articles -3.26 .017
Number of press release -7.40 .000
The popularity of “Cambridge Analytica” in Google Trends -3.33 .014
Subsequently, a VAR model was conducted to explore the relationship between the number of news articles, press releases, and the Google search volume about the Cambridge Analytica scandal. Firstly, the max lag order was set at 21 days, based on theoretical
expectations. The best model was found at lag 3 (AIC = 15.79, HQIC = 15.66), which meant that any one actor only influences the other two within 3 days.
The Granger causality test was then run to analyze the relationships between the number of news articles, press releases, and Google Trends. The details of the Granger causality test results are presented in Table 3. According to the findings, the number of news articles can be used to predict Google search queries in the next day, and the number of press releases can forecast Google search queries after one to two days. However, Google Trends do not influence the number of news articles or press releases reversely. H1 and H4 are therefore supported, but H2-3 and H5-6 are not.
The autocorrelations of the three variables’ residuals were then analyzed, and there was no autocorrelation for the number of news articles (p = .999), the number of press releases (p = .080), or the Google Trends (p = .913). Furthermore, the correlation is quite low between the
residuals of news articles and Google Trends (r = .10), or the residuals of the amount of press releases and Google Trends (r = -.14), which means there is no external element which
influenced the relationship between the amount of news articles and online search behavior, or the amount of press releases and online search behavior.
To further test the direction of the relationships between the variables, the impulse-response function (IRF) was used. Based on the Granger causality test findings, the IRF test was conducted to analyze the strength of the significant relationships. Forecasting error variance decomposition (FEVD) is used to predict how much variance is based on the
innovations of the variable itself, and how much variance is explained by other variables. In the IRF test, 16.7% of the search volume can be understood by the amount of news articles (FEVD = 0.69 in step 8). The press releases, however, can only predict 5.70% of the variance for the search volume.
Table 3
The VAR Model of the Number of News Articles, Press Releases, and Google Trends
Independent variable Dependent variable χ² Std. error Number of news articles Number of press releases
L1. -0.03 0.23 L2. 0.00 0.17 L3. -0.11 0.18 Google Trends L1. 0.14* 0.07 L2. 0.02 0.13 L3. 0.16 0.12
Number of press releases Number of news articles
L1. 0.06 0.32 L2. -0.26 0.28 L3. 0.06 0.20 Google Trends L1. 0.30** 0.07 L2. -0.30* 0.14 L3. 0.15 0.14
Google Trends Number of news articles
L1. 0.30 0.44
L2. -0.60 0.38
L3. 0.38 0.27
Number of press releases
L1. 0.56 0.34
L2. 0.00 0.25
L3. 0.53 0.27
Note: Granger causality coefficients (χ²) were reported * p < .05, ** p< .001
The transfer of subtopics between the media, organization, and the public was tested next. As the VAR model cannot be used to explore the transfer of sub-topics between three actors, the lagged regression model (LRM) was conducted to evaluate how the separate subtopics of one actor influenced the other two over time. To test how long one actor can influence the other two, trial-and-error was used to find that the maximum lag with observation in the LRM is six. Therefore, the independent variables were three actors * five topics * time lags (from lag 1-6), and the dependent variables were three actors * five corresponding topics. It is worth to mention that when testing the effect of search behavior, the time lag of LRM started from lag 2 to
measure the reverse agenda-setting effect after the media/corporation showing their impact on the public agenda. Overall, 30 LRMs were used to test the relationship between the media, the corporation, and the public.
Firstly, the effect of the media agenda on the public agenda was tested. LRM tested the impact of news subtopics on Google search query. For the fourth topic, the media agenda had a significant effect on the public agenda at lag 1, b = 0.40, t(33) = 2.32 p = .027. This means that when the media published news reports about topic 4, the Google search volume about this topic increased after one day. However, there is no significant result for other four subtopics. Therefore, H7 is partially supported.
Subsequently, the effects of the media agenda on the corporate agenda were explored. LRM tested the distribution of news subtopics in press releases. At lag 3, there is a significant effect for topic 2, b = -0.17, t(10) = -2.48, p = .048. This indicates that when the media published articles about topic 2, Facebook published press releases of the same topic after three days. At lag 2, there is a significant result for topic 3, which indicates that after two days of news reports about the third topic, Facebook will publish press releases with the same topic, b = 0.22, t(10) = 3.85, p = .008). H8 is, thus, partially supported.
Next, the effect of the corporate agenda on the media agenda was tested. LRM built to find the transfer of corporate subtopics in other two actors. Here, there were no significant effects found of the corporate agenda’s influence on the media agenda. This showed that in the subtopic transfer of the Cambridge Analytica scandal, the corporate agenda did not drive the media agenda. H9 is not supported.
The relationship between the corporate agenda and public agenda was tested. For the first subtopic, the corporate agenda significantly influenced the public agenda after one day, b = .19, t(12) = 4.20, p = .002. However, there were no other significant results for any other subtopic. This finding indicates that when Facebook published press releases about the first topic, the related Google searches increased after just one day. Therefore, H10 is partially supported.
In testing if the Google search volume can drive the corporate agenda, no significant results were found. Therefore, the subtopics of public agenda did not drive the subtopics of corporate agenda for the Cambridge Analytica scandal and H11 is not supported.
Similarly, the public agenda had no significant effect on the media agenda. Therefore, the public agenda did not drive the media agenda in the discussion of Cambridge Analytica scandal and H12 is not supported.
To answer the research question, in the first-level agenda-setting, the media agenda and corporate agenda both shaped the public agenda in Cambridge Analytica scandal, but there is no either reverse influence of public agenda on media and corporate agenda, or the relationship between media agenda and corporate agenda. In the second-level agenda-setting of the
transfer of subtopics, the subtopics transferred from the media and corporation to the public, and from the media to the corporation. No other interaction was found in the study.
Discussion
The purpose of this study was to examine the role of the media, the corporation and the public through unsupervised machine learning. Based on the literature, this study improved the
intermedia agenda setting theory, and filled this gap by exploring the agenda setting between the media, the company, and the public in the context of the Cambridge Analytica scandal.
The VAR model used to test the first-level agenda setting theory showed that after the amount of news articles increased, Google search volume about the issue also increased within one day. This result is consistent with the traditional agenda setting theory that states the media agenda drives public attention (McComb & Shaw, 1972). Contrary to what was expected, the increase of Facebook press releases decreased online searches about Cambridge Analytica after two days. The existing theory gives little inclination as to why this may be, but it can be speculated that the response from the company fulfills the public’s needs about the issue and thus decreases their attention towards it.
The LRM then explored the second-level agenda setting theory. This study indicated the media agenda leads the corporate and public agendas for some subtopics, but there is no evidence supporting any reverse relationships. This demonstrated that the media still took the lead in the agenda setting process; the subtopics transferred from the media to other actors. The model additionally found that the corporate agenda drove the public agenda. This is in line with both Householder et al. (2018), who showed that the increase of political advertising led to an increase of relevant online searches, and Jin et al. (2011), who found that a corporation’s response to a crisis can increase their perceived reputation from the public.
To conclude, a unidirectional relationship between the media and both the company and the public, and between the company and the public is still prevalent in the topic transferring process. In the crisis communication, no mutual relationship between three actors were found. The public attention is still driven by the media and corporation attention. In addition, based on the LRM, the response of public (i.e., after one day) is quicker than the corporation (i.e., after two to three days), which is consistent with the Vliegenthart (2014) and Kroon and van de Meer’s (2018) findings.
Few previous studies in public relations have focused on exploring three actors (i.e., the media, the organization, and the public) in one study. Typically paired relationships between the media and corporate agenda, the corporate and public agenda, or the media and public agenda are tested (e.g. Kroon & van de Meer, 2018; Ragas, et al., 2014; Verhoeven, 2009). This study, thus, contributed to extending the intermedia agenda setting theory by testing the issue transfer between all three of these actors, and the application of time series analysis helped to find the temporal order between them. Additionally, this study has deepened the research about crisis communication. While the existing literature about agenda setting theory in crisis
communication has only tested the relationship between the volume of news and press releases (Kleinnijenhuis, et al., 2015; Ragas, et al., 2014), this study included the second-level agenda setting to find the transfer of subtopics among three actors.
In the examination of issue transfer between three actors (i.e., the media, corporations, and the public) with unsupervised machine learning, the application of topic modeling provided scholars with a new tool to detect issue topics. Through topic modeling, the topics are extracted from the data rather than determining topics prior to analyses (e.g., supervised machine
learning). This method does not require human-coding, and as such, is more objective and lacks bias (Chandelier et al., 2018). Furthermore, unsupervised machine learning can process a large corpus of documents that would be more demanding if human-coded.
Four main limitations of this study should be considered. First, as a preliminary application of unsupervised machine learning in the agenda setting theory, this paper did not discuss the accuracy identify topics from documents with topic modeling. Although previous studies have shown that topic modeling can produce similarly accurate results to that of topic identification in supervised machine learning (Chandelier, et al., 2018; Roberts, Stewart, Tingley, & Airoldi, 2013), it would be valuable to compare the topic selections from topic modeling with other methods in future agenda setting studies. Second, this study focused on the distribution of topic words in a document rather than the interpretation of the semantic meanings. As a statistical
algorithm, topic modeling detected topics based on the probability of word co-occurrence. Therefore, it is difficult to represent the output as understandable topics in an academic paper. Besides, the study of subtopics is expected to be more informative (Lim, 2011). However, this study did not focus on the interpretation of the topics. Future research should thus pay more attention to the semantic meaning of topics. Third, this study used the LRM instead of VAR modeling in the analysis of topic transfer between three actors, due to the insufficient observations for VAR modeling. Future studies should collect data from a longer timeline to explore a longitudinal effects. Finally, newspapers served as the representation of the media in this study, Facebook press releases as the corporate agenda, and Google search queries as the public agenda. Future studies should expand the intermedia agenda setting to other
mediums. For example, Facebook company posts on social media could serve as the corporate agenda and the replies to these posts, as the public agenda.
In conclusion, this study made important contribution in the intermedia agenda-setting theory, and methodological innovation. Previous studies focused more on the interplay between the media and the organization, or between the media and the public (e.g. Kroon & van de Meer, 2018; Ragas et al., 2014). Herewith, this study has extended the research of intermedia agenda-setting theory to the agenda transfer between the media, the organization and the public. Besides, this study expanded the second-level agenda-setting to the transfer of subtopics. The results showed the subtopics of media agenda transferred to the corporate agenda and the public agenda, and the subtopics of corporate agenda transferred to the public agenda.
Reference
Birkland, T. A. (1997). After disaster: Agenda setting, public policy, and focusing events. Georgetown University Press.
Blei, D. M., Ng, A. Y., & Jordan, M. I. (2003). Latent dirichlet allocation. Journal of machine Learning research, 3(Jan), 993-1022.
Boumans, J. (2017). Subsidizing the News? Organizational press releases’ influence on news media’s agenda and content. Journalism Studies, 1-19.
Brandt, P. T., & Williams, J. T. (2007). Multiple time series models. Thousand Oaks, CA: SAGE.
Carroll, C. E., & McCombs, M. (2003). Agenda-setting effects of business news on the public's images and opinions about major corporations. Corporate reputation review, 6(1), 36-46.
Chandelier, Steuckardt, Mathevet, Diwersy, & Gimenez. (2018). Content analysis of newspaper coverage of wolf recolonization in France using structural topic modeling. Biological Conservation, 220, 254-261.
Coombs, W. (2007). Protecting Organization Reputations During a Crisis: The
Development and Application of Situational Crisis Communication Theory. Corporate
Reputation Review, 10(3), 163-176.
Curtin, P. A. (1999). Reevaluating public relations information subsidies: Market-driven journalism and agenda-building theory and practice. Journal of Public Relations Research, 11(1), 53-90.
de Vreese, C. H., Boomgaarden, H. G., & Semetko, H. A. (2011). (In) direct framing effects: The effects of news media framing on public support for Turkish membership in the European Union. Communication Research, 38, 179-205.
doi:10.1177/0093650210384934
Denham, B. E. (2014). Intermedia attribute agenda setting in the New York Times: The case of animal abuse in US horse racing. Journalism & Mass Communication Quarterly, 91(1), 17-37.
Dick, M. (2011). SEARCH ENGINE OPTIMISATION IN UK NEWS PRODUCTION. Journalism Practice, 5(4), 462-477.
Downs, A. (1972). Up and down with ecology: The issue-attention cycle. The public. European Commission, (2018), Data Protection, Factsheet Changed, retrieved from:
https://ec.europa.eu/commission/sites/beta-political/files/data-protection-factsheet-changes_en.pdf
Greene, D., & Cross, J. P. (2017). Exploring the political agenda of the European parliament using a dynamic topic modeling approach. Political Analysis, 25(1), 77-94. Ghanem, S. (1997). Filling in the tapestry: The second level of agenda-setting. In M.E. McCombs, D.L. Shaw, & D.H. Weaver (Eds.), Communication and democracy: Exploring the intellectual frontiers in agenda-setting theory (pp. 314). Mahwah, NJ: Lawrence Erlbaum Associates
Glusac, M. (2018 December, 10). 10 news stories that gripped the world in 2018, retrieved from:
https://www.thisisinsider.com/big-world-news-top-headlines-2018-11
Google Trend (2019 January, 04). Google Trend, retrieved from:
https://trends.google.com/trends/
Guo, L., & Vargo, C. (2015). The Power of Message Networks: A Big-Data Analysis of the Network Agenda Setting Model and Issue Ownership. Mass Communication and Society,
18(5), 557-576.
Guo, L., & Vargo, C. (2017). Global Intermedia Agenda Setting: A Big Data Analysis of International News Flow. Journal of Communication, 67(4), 499-520.
Heim, K. (2013). Framing the 2008 Iowa Democratic Caucuses: Political Blogs and Second-Level Intermedia Agenda Setting. Journalism & Mass Communication Quarterly, 90(3), 500-519.
Housholder, E., Watson, B., & Lorusso, S. (2018). Does Political Advertising Lead to Online Information Seeking? A Real-World Test Using Google Search Data. Journal of
Isaak, J., & Hanna, M. J. (2018). User Data Privacy: Facebook, Cambridge Analytica, and Privacy Protection. Computer, 51(8), 56-59.
Harder, R. A., Sevenans, J., & Van Aelst, P. (2017). Intermedia agenda setting in the social media age: How traditional players dominate the news agenda in election times. The International Journal of Press/Politics, 22(3), 275-293.
Heggde, G., & Shainesh, G. editor. (2018). Social Media Marketing Emerging Concepts and Applications.
Jin, Y., Liu, B., & Austin, L. (2014). Examining the Role of Social Media in Effective Crisis Management: The Effects of Crisis Origin, Information Form, and Source on Publics’ Crisis Responses. Communication Research, 41(1), 74-94.
Kim, J. Y., Kiousis, S., & Xiang, Z. (2015). Agenda building and agenda setting in business: Corporate reputation attributes. Corporate Reputation Review, 18(1), 25-36.
Kiousis, S., Popescu, C., & Mitrook, M. (2007). Understanding Influence on Corporate Reputation: An Examination of Public Relations Efforts, Media Coverage, Public Opinion, and Financial Performance From an Agenda-Building and Agenda-Setting Perspective.
Journal of Public Relations Research, 19(2), 147-165.
Kiousis, S., Kim, J. Y., Kochhar, S. K., Lim, H. J., Park, J. M., & Im, J. S. (2016). Agenda-building linkages between public relations and state news media during the 2010 Florida Senate Election. Public Relations Review, 42(1), 240-242.
Kroon, A. C., & van der Meer, T. G. (2018). Who Takes the Lead? Investigating the Reciprocal Relationship Between Organizational and News Agendas. Communication Research, 0093650217751733.
Kleinnijenhuis, J., Schultz, F., Utz, S., & Oegema, D. (2015). The mediating role of the news in the BP oil spill crisis 2010: How US news is influenced by public relations and in turn influences public awareness, foreign news, and the share price. Communication Research, 42(3), 408-428.
Lim, J. (2011). First-level and second-level intermedia agenda-setting among major news websites. Asian Journal of Communication, 21(2), 167-185.
Lopez-Escobar, E., Llamas, J. P., McCombs, M., & Lennon, F. R. (1998). Two levels of agenda setting among advertising and news in the 1995 Spanish elections. Political Communication, 15(2), 225-238.
Lumb, D (2018 May 02), Cambridge Analytica is shutting down following Facebook scandal, retrieved from: https://www.engadget.com/2018/05/02/cambridge-analytica-is-shutting-down-following-facebook-scandal/
Martin, A. (2018 April 05), Cambridge Analytica and Facebook: What happened and did the company shift many votes? Retrieved from:
https://www.alphr.com/politics/1008854/cambridge-analytica-facebook-what-happened
Matthes, J. (2012). Exposure to counterattitudinal news coverage and the timing of voting decisions. Communication Research, 39, 147-169. doi:10.1177/0093650211402322
Maurer, M., & Holbach, T. (2016). Taking Online Search Queries as an Indicator of the Public Agenda: The Role of Public Uncertainty. Journalism & Mass Communication
Quarterly, 93(3), 572-586.
Mellon, J. (2013). Internet Search Data and Issue Salience: The Properties of Google Trends as a Measure of Issue Salience. Journal of Elections, Public Opinion & Parties,
24(1), 1-28.
Meraz, S. (2009). Is There an Elite Hold? Traditional Media to Social Media Agenda
Setting Influence in Blog Networks. Journal of Computer‐ Mediated Communication,
14(3), 682-707.
Miller, A., & LaPoe, V. (2016). Visual agenda-setting, emotion, and the BP oil disaster. Visual Communication Quarterly, 23(1), 53-63.
Patriotta, G., Gond, J. P., & Schultz, F. (2011). Maintaining legitimacy: Controversies, orders of worth, and public justifications. Journal of Management Studies, 48(8), 1804-1836.
Persily, N. (2017). The 2016 US Election: Can democracy survive the internet?. Journal of democracy, 28(2), 63-76.
Peters, J. (2010 July 05). At Yahoo, Using Searches to Steer News Coverage. Retrieved from: http://www.nytimes.com/2010/07/05/business/media/05yahoo.html.
Pollach, I. (2014). Corporate environmental reporting and news coverage of environmental issues: An agenda‐ setting perspective. Business Strategy and the Environment, 23(5), 349-360.
Ragas, M., & Tran, H. (2013). Beyond Cognitions: A Longitudinal Study of Online Search Salience and Media Coverage of the President. Journalism & Mass Communication Quarterly, 90(3), 478-499.
Ragas, M. W., Tran, H. L., & Martin, J. A. (2014). Media-induced or search-driven? A study of online agenda-setting effects during the BP oil disaster. Journalism Studies, 15(1), 48-63.
Reese, S. D., & Danielian, L. H. (1989). Intermedia influence and the drug issue. Communication campaigns about drugs: Government, media, and the public, 29-46. Roberts, M., & McCombs, M. (1994). Agenda setting and political advertising: Origins of the news agenda. Political communication, 11(3), 249-262.
Roberts, M. E., Stewart, B. M., Tingley, D., & Airoldi, E. M. (2013, January). The structural topic model and applied social science. In Advances in neural information processing systems workshop on topic models: computation, application, and evaluation (pp. 1-20). Roberts, M. E., Stewart, B. M., Tingley, D., Lucas, C., Leder‐ Luis, J., Gadarian, S. K., ... & Rand, D. G. (2014). Structural topic models for open‐ ended survey responses.
Rogstad, I. (2016). Is Twitter just rehashing? Intermedia agenda setting between Twitter and mainstream media. Journal of Information Technology & Politics, 13(2), 142-158. Russell Neuman, W., Guggenheim, L., Mo Jang, S., & Bae, S. Y. (2014). The dynamics of public attention: Agenda-setting theory meets big data. Journal of Communication, 64(2), 193-214.
Salwen, M. B., & Matera, F. R. (1992). Public salience of foreign nations. Journalism Quarterly, 69(3), 623-632.
Seeger, M. W., Sellnow, T. L., & Ulmer, R. R. (2003). Communication and organizational crisis. Greenwood Publishing Group.
Shen, L. (2018 March 19), Why Facebook Suddenly Shed $35 Billion in Value. Retrieved from: http://fortune.com/2018/03/19/facebook-stock-share-price-cambridge-analytica-donald-trump/
Statista.com (2019). Most popular social networks worldwide as of October 2018, ranked by number of active users (in millions). Retrieved from:
https://www.statista.com/statistics/272014/global-social-networks-ranked-by-number-of-users/
Strauß, N., & Vliegenthart, R. (2017). Reciprocal influence? Investigating implicit frames in press releases and financial newspaper coverage during the German banking crisis. Public Relations Review, 43(2), 392-405. (title contain keywords)
Tam, D (2012 August 22). Facebook processes more than 500 TB of data daily. Retrieved from: https://www.cnet.com/news/facebook-processes-more-than-500-tb-of-data-daily/
Tam, L. (2015). News about corporate social responsibility (CSR): The interplay of intermedia agenda setting influences between corporate news releases and press coverage. Asian Journal of Business Ethics, 4(2), 117-130.
Tarran, B. (2018). What can we learn from the Facebook—Cambridge Analytica scandal?. Significance, 15(3), 4-5.
Uscinski, J. (2009). When Does the Public's Issue Agenda Affect the Media's Issue
Agenda (and Vice‐ Versa)? Developing a Framework for Media‐ Public Influence *. Social
Science Quarterly, 90(4), 796-815.
Van Aelst, P., & Vliegenthart, R. (2014). Studying the tango: An analysis of parliamentary questions and press coverage in the Netherlands. Journalism Studies, 15, 329-410. doi:10. 1080/1461670X.2013.831228
Vliegenthart, R., & Walgrave, S. (2008). The contingency of intermedia agenda setting: A longitudinal study in Belgium. Journalism & Mass Communication Quarterly, 85(4), 860-877.
Vliegenthart, R. (2014). Moving up. Applying aggregate level time series analysis in the study of media coverage. Quality & Quantity, 48, 2427-2445. doi:10.1007/s11135-013-9899-0
Verhoeven, P. (2009). Corporate actors in Western European television news. Public Relations Review, 35(3), 297-300.
Vonbun, R., Königslöw, K., & Schoenbach, K. (2016). Intermedia agenda-setting in a multimedia news environment. Journalism, 17(8), 1054-1073.
Ward, K. (2018). Social networks, the 2016 US presidential election, and Kantian ethics: applying the categorical imperative to Cambridge Analytica’s behavioral microtargeting. Journal of media ethics, 33(3), 133-148.
Appendix 1 Subtopics in news articles
Topic 1:
Words: 0.007*"claim" + 0.007*"tell" + 0.006*"investigation" + 0.006*"kogan" + 0.006*"wylie" + 0.006*"harvest" + 0.006*"leave" + 0.005*"know" + 0.005*"share" + 0.005*"personal"
Words: 0.008*"like" + 0.008*"congress" + 0.008*"zuckerberg" + 0.008*"target" + 0.007*"india" + 0.006*"know" + 0.006*"party" + 0.006*"committee" + 0.006*"profile" + 0.005*"report"
Topic 3
Words: 0.010*"profile" + 0.009*"employee" + 0.008*"voter" + 0.007*"wylie" + 0.007*"parscale" + 0.007*"analyticas" + 0.007*"mercer" + 0.006*"business" + 0.006*"effort" + 0.006*"know"
Topic 4
Words: 0.011*"report" + 0.009*"profile" + 0.009*"group" + 0.007*"voter" + 0.006*"investigation" + 0.006*"zuckerberg" + 0.006*"privacy" + 0.005*"analyticas" + 0.005*"mercer" + 0.005*"party" Topic 5
Words: 0.009*"obtain" + 0.009*"personal" + 0.007*"share" + 0.006*"kogan" + 0.006*"voter" + 0.006*"british" + 0.006*"party" + 0.006*"access" + 0.006*"help" + 0.006*"executive"
Subtopics from press releases Topic 1
Words: 0.041*"information" + 0.026*"local" + 0.024*"data" + 0.022*"research" + 0.021*"medium" + 0.020*"social" + 0.019*"election" + 0.015*"want" + 0.014*"political" + 0.013*"policy"
Topic 2
Words: 0.052*"business" + 0.048*"community" + 0.037*"small" + 0.033*"messenger" +
0.019*"friend" + 0.019*"fee" + 0.019*"program" + 0.017*"need" + 0.016*"video" + 0.016*"digital" Topic 3
Words: 0.029*"post" + 0.024*"community" + 0.020*"technology" + 0.020*"information" +
0.019*"today" + 0.019*"fee" + 0.018*"content" + 0.017*"feature" + 0.015*"build" + 0.015*"page" Topic 4
Words: 0.043*"photo" + 0.035*"video" + 0.032*"information" + 0.030*"article" + 0.024*"false" + 0.022*"privacy" + 0.021*"misinformation" + 0.018*"data" + 0.017*"story" + 0.014*"technology" Topic 5
Words: 0.034*"political" + 0.027*"page" + 0.026*"fee" + 0.024*"content" + 0.021*"election" + 0.020*"false" + 0.020*"issue" + 0.018*"information" + 0.016*"account" + 0.016*"story"
