In the fake news era, a combination of politics, big technology, and fear and animosity are blamed for the media mistrust and filter bubbles that are entrenching fragmentation in the public sphere. A partisan divide in the media and extreme political disagreements are nothing new, but new technology, such as social media and mobile push notifications, influences these years-old phenomena and plays an important role in current concerns. This paper explores how stories are represented differently by topic and across platforms, examining representation, polarization, and objectivity. Specifically, this paper looks at those issues from a novel perspective: through sentiment analysis of push notifications generated and archived from the Breaking News App on disasters, gun violence, and terrorism. Results indicate that partisan news organizations (1) emphasize different stories; (2) label the same events as categorically different; (3) hyperbolize and emotionalize different types of stories; and, (4) represent different categories of breaking news stories to different degrees of subjectivity.
Results and discussion
Digital and social news consumption is increasingly prevalent (Fenton, 2009). People dedicate less time to intentional and recreational news consumption, watching the nightly news or reading the Sunday paper (Kim, 2014), and instead subscribe to mobile push notifications (Sanfilippo and Lev-Aretz, 2017), check headlines on their phones (Kim, 2014), and share stories on Facebook or Twitter (Hermida, 2010). Along with changes in news consumption, is increasing fragmentation of the fourth estate and its social perceptions, as people develop filter bubbles around their information worlds (Flaxman, et al., 2016). Furthermore, real concerns have emerged about the implications for democracy (Fenton, 2009), as mistrust abounds and the label of “fake news” is applied based on political agendas rather than to actual instances of misinformation or deception (Spohr, 2017). Simultaneously, there is a social perception that sources and political parties are not even talking about the same thing, as different topical coverage and framing of stories paint dramatically different perspectives and agendas (Schnell, 2001; Stroud, 2010). The aggregate result is dramatic inequalities in understanding and worldviews (Stroud, 2010).
Current associations between social technologies and the changes in journalism distribution and consumption are hardly the first links between new technologies and fourth estate concerns (Newman, et al., 2012; Sanfilippo and Lev-Aretz, 2017). Historically, scholarship has examined the impact of new technologies such as radio, network television, cable news, and early Internet on news environments and content (Pavlik, 2000, 1999; Steensen, 2011), as well as sociopolitical implications of change (Baym, 2010; Entman, 1994; Kellner, 2018).
Research on push notifications in journalism, however, is more nascent. Push notifications are primarily understood descriptively at present. Technical (e.g., Warren, et al., 2014) and marketing research (e.g., Andrews, et al., 2016) map how to target specific audiences (O’Kane, 2013), opportunities to bring time sensitive and special interest information to those audiences immediately (O’Kane, 2013; Rowles, 2017), and the potential to configure notifications in ways most suitable to individuals, with options to suspend them overnight or to represent them in specific ways (Rowles, 2017). However, we know much less about push notifications in journalistic practice (e.g., Brown, 2017) and how they affect users as readers (e.g., Kim, 2014). Push notifications, while still blunt tools in journalism, are prevalent and feed into business needs, as digital journalism is pushed to be more responsive at a faster pace (Fenton, 2009; Pavlik, 1999; Steensen, 2011) and depends on clicks for revenue, despite the increasing dependence by users on headlines only (Kormelink and Meijer, 2018; Lee, et al., 2014). Behavioral research also illustrates that more emotional representation of news stories generates more understanding and more clicks (Bas and Grabe, 2015). There is also evidence that push notifications differ significantly from headlines in content and sentiment, sometimes including multiple pushes per headline (Sanfilippo and Lev-Aretz, 2017). As a result, there are rational concerns about what personalization, polarization, and subjectivity in push notifications may imply for the fourth estate. Yet, whether these concerns are substantiated remains to be seen.
Recent scholarship provides evidence that a combination of user behaviors, platforms, and algorithms contribute to the construction of filter bubbles (Flaxman, et al., 2016). Within filter bubbles, individuals are exposed preferentially to content that looks like what they have read or liked in the past (Thurman and Schifferes, 2012), ideas are reinforced in echo chambers (Flaxman, et al., 2016), and diversity of ideas is minimized over time (Tan and Weaver, 2013). These trends, relative to curation through news aggregation on platforms like Google News and social media platforms (Allcott and Gentzkow, 2017; Newman, et al., 2012; Spohr, 2017) are partisan in nature (Stroud, 2010; Van Aelst, et al., 2012). Furthermore, media outlets themselves increasingly have varying degrees of partisan alignment (Groeling, 2013; Schudson, 2001; Stroud, 2010) and preferentially frame content to suit their business needs. Some partisan framing polarizes both stories and understanding of events, including extensions into misinformation (Allcott and Gentzkow, 2017; Garrett, 2017). Given evidence that push notifications are more subjective and emotional than traditional headlines (Sanfilippo and Lev-Aretz, 2017) and emotion is tied to polarization, while subjectivity increases partisanship, is there evidence that partisan polarization is exacerbated in public discourse by push notifications?
Understanding of how push notifications actually impact content and behaviors is limited, despite the intuition that it does, given extrapolation from past journalistic and technological innovations, as empirical evidence on news filter bubbles, and anecdotal accounts. In order to begin to address this research gap and evaluate assumptions, it is necessary to empirically assess: whether sources represent the same stories in politically different ways; whether types of stories are represented differently; whether differences are politically motivated; and how subjective differences in representations are.
Drawing on the Breaking News app archive, including all push notifications and stories that were fed through the news aggregation app between 2012 to 2016, three exemplar topical categories — disasters, gun violence, and terrorism — were selected from all possible labels based on the following criteria: (1) they represent true breaking news events important to the general public and social wellbeing, and are not only local interest stories; (2) they have political ramifications; and, (3) exemplar stories within each topic have public service alert counterparts to serve as controls in assessing objectivity for the story. The first criterion is important, given that a large sample is necessary to assess polarization and partisanship, while the second is key to thinking of the implications of polarization and partisanship for an informed electorate. Coverage by only local organizations would result in small samples, just as categorical labels that pertain to more specific units of analysis would skew the sample based on the story. All three of these categories are broad enough to include a variety of stories and tens of thousands of notifications. Many of the stories included within these categories also correspond with government notifications to serve as neutral controls for journalistic notifications, meeting criterion 3.
Notably, no other categories, labelled within the Breaking News archive, met all three categories. Furthermore, all topics have also been the subject of past empirical research on framing in journalism (e.g., Houston, et al., 2012; Lewis and Reese, 2009; Schnell, 2001). This sample includes push notifications from hundreds of international news outlets, based on story relevance, in addition to popular domestic publications.
Sentiment analysis represented the most useful methodology to address questions about partisanship and polarization in a controlled and consistent way across a large sample and multiple contexts. Analysis was specifically conducted with ‘SentimentAnalysis’ and ‘sentimentr’ in R on the entirety of the samples generated from each of the three topics: disasters (n=57,681), gun violence (n=72,417), and terrorism (n=38,212). Additionally, the R package ‘quanteda’ was used in visualizing quantitative analysis of textual data relative to the topic models constructed and ‘tm’ was employed for the final topic model matrix visualization. Examination of significant stories within this topic allows for empirical assessment of stories representation and the labeling of stories with the frames provided by individual news outlets. Examining three exemplar stories from each of three categories, as presented in Table 1, supported this line of inquiry.
Table 1: Story sample sizes. Disasters Gun violence Terrorism Missing Flight MH370 (n=1,447) Pulse Nightclub (n=1,752) Boston Marathon Bombing (n=2,863) Hurricane Sandy 2012 (n=3,164) Aurora, CO (n=557) Charlie Hebdo (n=944) Nepal Earthquake 2015 (n=682) Sandy Hook Elementary School (n=1,205) Nice, France (n=756)
Exemplar stories were selected to include large sample sizes and reflect broad interest, in addition to local interest, much like the criterion for categorical labels. Furthermore, they include international and domestic stories, incorporating a wide variety of journalistic sources. While other stories would certainly meet these requirements, the exemplars are designed to complement the relatively abstract results from the total categorical samples by providing concrete and recognizable illustrations in context. They are not selected to be representative on average, as one major research question focused on interrogating the impression of unequal representation of stories.
Notifications were collected from the archive in May of 2017, after NBC had discontinued the Breaking News project, but before the archive was taken off-line. Across the 168,310 push notifications included in the study, a total of 387 unique sources were included, including both government sources and news sources. Journalistic sources included: major U.S. print sources, such as the New York Times and Washington Post; television news sources, such as CNN; international English language press, such as BBC and ABC (Australia); online only outlets, such as Yahoo and Huffington Post; and lesser known sources, such as Asiancorrespondent.com.
News sources were classified on a simply partisan spectrum (conservative, neutral, or liberal) based on historical data sets, over the same time period, from the American National Election Survey (ANES). ANES provides linked data on how individuals self-identify as their partisan affiliation and what news sources they consider. Aggregating these paired indicators across the representative surveys provided by ANES, allows sources to be classified within one of those three labels. In this sense, the affiliation is to people of particular partisan persuasions, on average, based on who reads a given news source, rather than actual or labeled partisanship of journalistic institutions. For example, Reuters is classified as neutral because, while many readers identified as neutral, readers were also normally distributed along the political spectrum. In contrast, most who consumed news through Fox News identified as conservative, and thus Fox was classified as having a conservative affiliation. Classification was not based on author opinions or applied manually.
Results and discussion
Different names for the same thing
Topical polarization was evidence in a number of ways, including unequal representation of stories (Unequal Representation of Stories) and politicized framing of particular stories in ways that contrast with emergent topics (Polarization and Topic Framing). In this sense, a significant part of polarization in push notifications stems from the fact that different outlets and platforms use different names when referring to the same story, and provide significantly different coverage for different stories. Labels associated with stories matter in shaping where coverage might be found and polar representations. Labeling a mass shooting as gun violence or terrorism implies different motivations and drives different political responses to the same tragic event; these different labels are applied by news organizations, just as by politicians or interest groups.
Unequal representation of stories
There are significant imbalances in terms of story representation, even within a general topical category. Some stories, with similar elements, receive more attention than others, both within and between news outlets. Figure 1 illustrates the distributions by topic of story sizes, measured as the number of notifications in the overall sample associated with a given topic.
This figure illustrates extreme skewness and non-normal distributions of story sizes, indicating that there are imbalances in attention provided to stories. While it would not be expected that all stories receive an equal number of notifications, given that the magnitude of impact of some stories or the time over which they are relevant would vary, it is unexpected that there is quite this significant an imbalance in the number of notifications sent per story within each topic over time.
Figure 1: Story sizes by topic. Note: Larger version of Figure 1 available here.
While very few stories generate thousands of push notifications, most stories generate few notifications, regardless of topic. All three categories are also multimodal, with notification modes for stories about gun violence at 10 and 15, and notification modes for disasters are at 1 and 19 and terrorism are at 24 and 32. Notably, over 95 percent of stories about gun violence contain fewer than 40 push notifications, while 10 percent of stories about terrorism and 15 percent of stories about disasters exceed 100 notifications.
Polarization and topic framing
Coverage imbalances are not only about frequencies, however, as polarization shapes representation. Frames are applied that represent the same story as very different events, based on news outlets and interests. Figure 2 explores document similarity in language and sentiment employed, operationalized at the term level, meaning individual words or groups of words with meaning in context (e.g., “gun control” or “gun violence” as opposed to “gun,” “control,” or “violence”). Measuring document similarity through matrix factorization at the level of terms is helpful both to cluster similar documents and structure subsequent sentiment analysis by story representation (Maas, et al., 2011; Wang, et al., 2008). This approach is used to illustrate that within categories of ‘disasters,’ ‘gun violence,’ and ‘terrorism,’ as labeled by humans, patterns of similarities are not clear, yet there are very similar term distributions between stories across these labels. Figure 2 illustrates the similarities and differences between the specific exemplar stories, highlighting, for example, similar language between notifications about mass shootings and the Nice, France terror attack, but relatively different language used in discussing the Nice and Charlie Hebdo attacks.
Figure 2: Term document similarity. Note: Larger version of Figure 2 available here.
While the Breaking News app labeled individual stories with their own metadata, aggregating stories by events, other general categorical labels were generated based on the content provided by news outlets, which differed across the sample. Figure 3 illustrates the emergent topics, generated through a topic modeling approach, as a comparison to those applied by news outlets and platforms. This approach to topic modeling generated emergent themes within stories, across all notifications, as a way to suggest labels for those stories based on text (Chaney and Bley, 2012), rather than depending on human perceptions. The visualizations specifically illustrate both frequency and consensus (Tang, et al., 2013), and similar approaches have previously been employed to compare social perceptions and newspaper coverage (DiMaggio, et al., 2013).
Figure 3: Emergent topics by story. Note: Larger version of Figure 3 available here.
In comparison to Figure 2, which generally illustrates how similar language actually used in push notifications is across stories, Figure 3 reveals what topics emerge based on language specific to a story, as a point of comparison to the tags used to label a given story. For example, the theme of “terror” is prominent in language used and labelling for the Boston Marathon attack, while it is not at all prevalent or widely represented in the actual language employed in notifications about the Nice or Charlie Hebdo attacks, despite these stories being labeled as terrorism. Interestingly, while “shooting” is a prevalent topic in all three exemplar stories about gun violence, guns themselves are only a prevalent and widely distributed topic in the Sandy Hook story. Notifications about the disappearance of flight MH370 present the most unusual relationship between prevalence and distribution of topics, given that there are really two major sub-stories, one on the disaster and the other on the long term search, representing very distinct emergent themes.
The results illustrate that while the news aggregation platform provided labels that better matched the content than the frames intentionally applied by outlets, there are real differences between what journalists and organizations deem relevant and appropriate framing, across competitors and in contrast within events and topics.
Talking past each other
Polarization by topic is diverse among sources and stories, as well. Results indicate that political coverage imbalances (Political Coverage Imbalances) and subjectivity of polarization (Subjective Polarization) are associated with intersections between polarization and partisanship of news sources (Intersection of Partisanship and Polarization), providing support for the common perception that liberals and conservatives, along with their preferred media sources, are having different conversations and talking past one another.
Political coverage imbalances
As alluded to in the section on “Unequal representation of stories”, many of the coverage imbalances by topic are partisan in nature. Figure 4 illustrates the partisan breakdown in how many stories were pushed on specific topics within each category included in the study, drawing on the same exemplars from Table 1. There is a skew toward coverage of U.S. based stories, overall, with Hurricane Sandy (2012) and the Boston Marathon Bombing receiving both disproportionate and skewed coverage. This is likely the result of only considering English language sources and a likely skew towards U.S. journalism in the archive.
Figure 4: Partisan story size imbalance. Note: Larger version of Figure 4 available here.
The implication is that political fragmentation has led the media to talk about entirely different things. While outlets with liberal affiliations spent considerable time talking about the Pulse Night Club tragedy, outlets with conservative affiliations glossed over it to talk more about terrorism, for example. This is not only a story of partisan preferences, but also a story about shaping public understanding of events’ significance. While some differences may stem from legitimate concerns, such as audience demographics that make international stories of greater interest for some outlets, there is also evidence that certain types of stories are prioritized or downplayed based on partisan agendas.
Overall, there were approximately twice as many notifications about gun violence as terrorism (Table 1), yet not all stories within those categories were treated equally. Over the entire sample, not merely the exemplary stories, categorical partisan balances are even more extreme, as Figure 5 illustrates. While disaster coverage is only slightly more prominent in news sources with a high conservative affiliation, there are extreme imbalances between liberally and conservatively affiliated sources on the topics of gun violence and terrorism, with ratios nearing 3:1 notifications from conservative to liberal sources on terrorism and from liberal to conservative sources on gun violence.
Figure 5: Partisan imbalance in news coverage. Note: Larger version of Figure 5 available here.
Intersections of partisanship and polarization
Partisan imbalances extend beyond coverage of stories to very different representations within content. Figure 6 presents results of sentiment analysis in terms of polarization of content across all notifications within the three topic areas. As a meaningful control, government notifications relative to the same topics are close to neutral, in terms of implying positive or negative emotional affect about stories. As stark contrasts, notifications from liberally affiliated sources on gun violence and conservatively affiliated sources on terrorism imply extremely negative emotions in the language used to frame stories within these topics. Comparisons across partisan clusters of news outlets illustrates how different representations look relative to gun violence stories, for example, with relatively neutral language in conservative sources, moderately negative language in sources with neutral affiliations and very negative sentiment expressed in liberally affiliated sources.
Figure 6: Partisanship of news outlets shapes content polarization by topic. Note: Larger version of Figure 6 available here.
Partisan imbalances are, of course, not the only factor driving polarization of content by topic, as neutrally affiliated news outlets present stories through more emotionally negative representations than government notifications. Yet the extreme polar differences in topical representations within partisan clusters of news organizations illustrates how particular topics are framed very differently based on agendas. Among liberal sources, for example, the spread between emotional representations of disasters and gun violence is significant, measuring greater than 1, with a nearly as extreme parallel spread between notification language about gun violence and terrorism by conservative sources.
Not only is polarization partisan in nature, but topical frames are increasingly subjective along partisan designations. Table 2 presents a heat map that illustrates a spectrum from perfectly objective at 0 to highly subjective at 1. These classifications are based on established measurements for subjectivity at the level of individual words in computational linguistics (e.g., Wiebe, et al., 2004), here assessed using specific sentiment analysis models on subjectivity. Subjectivity here refers to aggregate connotations and expressions of “opinions, emotions, sentiments, beliefs, and speculations”, while objectivity refers to neutral language and communication of themes . “Perfect objectivity,” even in brief text segments like push notifications, is rare, with only government notifications approaching a measurement of zero.
Table 2: Partisanship of news outlets shapes content subjectivity by topic.
While liberal representations of disasters and terrorism, as well as neutral representations of gun violence, are weakly subjective, liberal representations of gun violence and conservative representations of all three topics are subjective in representations. Neutral sources represent disasters and terrorism in a weakly objective manner, while only government notifications on these issues are objective.
This data notably illustrates that even while language may not portray a given story or topic in highly polar way, as either emotionally positive or negative to an extreme degree, it may still be solidly subjective, as with conservative representations of gun violence.
Polarization of topics is evident across a variety of news sources, when examining push notifications relative to breaking news stories on terrorism, gun violence, and disasters. Partisanship also exacerbates polarization by topic in news coverage.
The implications of this work are multifaceted. First, push notifications are contributing to filter bubble effects, with or without personalization, which is not addressed by this study, based on individuals’ source preferences. Given partisan differences in coverage, polarization, and objectivity relative to topics, and stories within topics, individuals are not only consuming news that supports their partisan views, but also that represent the same issue in ways that may not be recognizable to another audience.
Second, polarization of issues and stories is not a completely partisan problem, indicating that expectations of impartial, unemotional coverage of important news stories as duties of the fourth estate may be undermined by applications of new technologies and business models aimed to maximize readership.
Third, dependence on push notifications to be the first to know about important breaking news is undermined by practice. What is considered important by one news organization is not necessarily considered important by another. Further, as increased sentiment in framing notifications is more profitable, there is no incentive for news organizations to provide objective notifications. As a result, breaking news through push notifications is socio-politically constructed to a greater extent than past forms of journalism, particularly given the lack of transparency and logs of how notifications are developed and what was sent. It is particularly notable that aggregation services, which collected push notifications, such as the Breaking News archive on which this study was conducted are increasingly purchased by major media players and taken offline, including backlogs and archives.
About the authors
Madelyn Rose Sanfilippo is a postdoctoral research associate at the Center for Information Technology Policy at Princeton University. She is broadly interested in legal, social, and political issues surrounding information and information technology access, applying a social informatics perspective. Her research empirically explores governance of sociotechnical systems, as well as outcomes, inequality, and consequences within these systems, through mixed method research design. Madelyn is also currently collaborating on a large scale project, funded by the Tow Center for Digital Journalism at Columbia University, to examine how push notifications and personalized distribution and consumption of news manipulate readers and contributes to media anxiety, as well as what the implications of these changes in digital journalism may be for an informed electorate. Madelyn’s work is informed by her interdisciplinary background, as she studied political science, international studies, Spanish, and environmental studies at the University of Wisconsin, Madison as an undergraduate and completed her Master’s and doctoral studies in information science at Indiana University, Bloomington’s School of Informatics and Computing. Madelyn was also previously a postdoctoral research scholar at the Information Law Institute at New York University’s School of Law, where she studied knowledge commons governance, as well as social consequences and governance of artificial intelligence.
Direct comments to: madelyns [at] princeton [dot] edu
Yafit Lev-Aretz is an Assistant Professor of Law at Zicklin School of Business, Baruch College, City University of New York. Professor Lev-Aretz is also a research fellow at the Tow Center at Columbia Journalism School. As the digital environment constantly evolves, Professor Lev-Aretz studies self-regulatory regimes set by private entities and the legal vacuum they create. She is especially interested in the growing use of algorithmic decision-making, intrusive means of news dissemination, choice architecture in the age of big data, and the ethical challenges posed by machine learning and artificially intelligent systems. Her research also highlights the legal treatment of beneficial uses of data, such as data philanthropy and the data for good movement, striving to strike a delicate balance between solid privacy protection and competing values. Previously, Professor Lev-Aretz was a research fellow at the Information Law Institute at NYU, and an intellectual property fellow at the Kernochan Center for the Law, Media, and the Arts at Columbia University. Professor Lev-Aretz holds an SJD from the University of Pennsylvania Law School, LLM from Columbia Law School, and LLB from Bar-Ilan University in Israel.
We would like to thank the Tow Center for Digital Journalism at Columbia University for their financial support as we pursued this project; the opportunities associated with the Knight News Innovation Fellowship and their generosity made it possible to capture, store, and process large push notification data sets, as well as enabled our collaboration on this project across campuses. We are also grateful to our colleagues at NYU’s Information Law Institute and Privacy Research Group, Princeton University’s Center for Information Technology Policy, Data and Society, and the University of Wisconsin iSchool for all of their helpful suggestions and constructive feedback.
1. Montoyo, et al., 2012, p. 675.
Hunt Allcott and Matthew Gentzkow, 2017. “Social media and fake news in the 2016 election,” Journal of Economic Perspectives, volume 31, number 2, pp. 211–236.
doi: https://doi.org/10.1257/jep.31.2.211, accessed 14 August 2019.
Michelle Andrews, Jody Goehring, Sam Hui, Joseph Pancras, and Lance Thornswood, 2016. “Mobile promotions: A framework and research priorities,” Journal of Interactive Marketing, volume 34, pp. 15–24.
doi: https://doi.org/10.1016/j.intmar.2016.03.004, accessed 14 August 2019.
Ozen Bas and Maria Elizabeth Grabe, 2015. “Emotion-provoking personalization of news: Informing citizens and closing the knowledge gap?” Communication Research, volume 42, number 2, pp. 159–185.
doi: https://doi.org/10.1177/0093650213514602, accessed 14 August 2019.
Geoffery Baym, 2010. From Cronkite to Colbert: The evolution of broadcast news. Boulder, Colo.: Paradigm Publishers.
Pete Brown, 2017. “Pushed beyond breaking: US newsrooms use mobile alerts to define their brand,” Columbia Journalism Review (29 November), at https://www.cjr.org/tow_center_reports/push-mobile-alerts-brand-breaking-news.php/, accessed 14 August 2019.
Allison J.B. Chaney and David M. Blei, 2012. “Visualizing topic models,” Proceedings of the Sixth International AAAI Conference on Weblogs and Social Media, at https://www.aaai.org/ocs/index.php/ICWSM/ICWSM12/paper/viewFile/4645/5021, accessed 14 August 2019.
Paul DiMaggio, Manish Nag, and David Blei, 2013. “Exploiting affinities between topic modeling and the sociological perspective on culture: Application to newspaper coverage of U.S. government arts funding,” Poetics, volume 41, number 6, pp. 570–606.
doi: https://doi.org/10.1016/j.poetic.2013.08.004, accessed 14 August 2019.
Robert M. Entman, 1994. “Representation and reality in the portrayal of blacks on network television news,” Journalism & Mass Communication Quarterly, volume 71, number 3, pp. 509–520.
doi: https://doi.org/10.1177/107769909407100303, accessed 14 August 2019.
Natalie Fenton (editor), 2009. New media, old news: Journalism and democracy in the digital age. Los Angeles: Sage.
Seth Flaxman, Sharad Goel, and Justin M. Rao, 2016. “Filter bubbles, echo chambers, and online news consumption,” Public Opinion Quarterly, volume 80, number S1, pp. 298–320.
doi: https://doi.org/10.1093/poq/nfw006, accessed 14 August 2019.
R. Kelly Garrett, 2017. “The ‘echo chamber’ distraction: Disinformation campaigns are the problem, not audience fragmentation,” Journal of Applied Research in Memory and Cognition, volume 6, number 4, pp. 370–376.
doi: http://dx.doi.org/10.1016/j.jarmac.2017.09.011, accessed 14 August 2019.
Tim Groeling, 2013. “Media bias by the numbers: Challenges and opportunities in the empirical study of partisan news,” Annual Review of Political Science, volume 16, pp. 129–151.
doi: https://doi.org/10.1146/annurev-polisci-040811-115123, accessed 14 August 2019.
Alfred Hermida, 2010. “Twittering the news: The emergence of ambient journalism,” Journalism Practice, volume 4, number 3, pp. 297–308.
doi: https://doi.org/10.1080/17512781003640703, accessed 14 August 2019.
J. Brian Houston, Betty Pfefferbaum, and Cathy Ellen Rosenholtz, 2012. “Disaster news: Framing and frame changing in coverage of major U.S. natural disasters, 2000–2010,” Journalism & Mass Communication Quarterly, volume 89, number 4, pp. 606–623.
doi: https://doi.org/10.1177/1077699012456022, accessed 14 August 2019.
Douglas Kellner, 2018. Television and the crisis of democracy. New York: Routledge.
Mijung Kim, 2014. “The effects of external cues on media habit and use: Push notification alerts and mobile application usage habits,” Ph.D. dissertation, Media and Information Studies, Michigan State University, at https://d.lib.msu.edu/etd/3263, accessed 14 August 2019.
Tim Grut Kormelink and Irene Costera Meijer, 2018. “What clicks actually mean: Exploring digital news user practices,” Journalism, volume 19, number 5, pp. 668–683.
doi: https://doi.org/10.1177/1464884916688290, accessed 14 August 2019.
Angela M. Lee, Seth C. Lewis, and Matthew Powers, 2014. “Audience clicks and news placement: A study of time-lagged influence in online journalism,” Communication Research, volume 41, number 4, pp. 505–530.
doi: https://doi.org/10.1177/0093650212467031, accessed 14 August 2019.
Seth C. Lewis and Stephen D. Reese, 2009. “What is the war on terror? Framing through the eyes of journalists,” Journalism & Mass Communication Quarterly, volume 86, number 1, pp. 85–102.
doi: https://doi.org/10.1177/107769900908600106, accessed 14 August 2019.
Andrew L. Maas, Raymond E. Daly, Peter T. Pham, Dan Huang, Andrew Y. Ng, and Christopher Potts, 2011. “Learning word vectors for sentiment analysis,” HLT ’11: Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, volume 1, pp. 142–150.
Andrés Montoyo, Patricio Martínez-Barco, and Alexandra Balahur, 2012. “Subjectivity and sentiment analysis: An overview of the current state of the area and envisaged developments,” Decision Support Systems, volume 53, number 4, pp. 675–679.
doi: https://doi.org/10.1016/j.dss.2012.05.022, accessed 14 August 2019.
Nic Newman, William H. Dutton, and Grant Blank, 2012. “Social media in the changing ecology of news: The fourth and fifth estates in Britain,” International Journal of Internet Science, volume 7, number 1, pp. 6–22, and at http://www.ijis.net/ijis7_1/ijis7_1_newman_et_al_pre.html, accessed 15 January 2019.
Brendan O’Kane, 2013. “2013: The breakout year for mobile measurement,” International Journal of Mobile Marketing, volume 8, number 1, pp. 86–94.
John Pavlik, 2000. “The impact of technology on journalism,” Journalism Studies, volume 1, number 2, pp. 229–237.
doi: http://dx.doi.org/10.1080/14616700050028226, accessed 15 January 2019.
John V. Pavlik, 1999. “New media and news: Implications for the future of journalism,” New Media & Society, volume 1, number 1, pp. 54–59.
doi: https://doi.org/10.1177/1461444899001001009, accessed 15 January 2019.
Daniel Rowles, 2017. Mobile marketing: How mobile technology is revolutionizing marketing, communications and advertising. Second edition. London: Kogan Page.
Madelyn Rose Sanfilippo and Yafit Lev-Aretz, 2017. “Breaking news: How push notifications alter the fourth estate,” First Monday, volume 22, number 11, at https://firstmonday.org/article/view/8068/6562, accessed 15 January 2019.
doi: https://doi.org/10.5210/fm.v22i11.8068, accessed 15 January 2019.
Karen C.F. Schnell, 2001. “Assessing the democratic debate: How the news media frame elite policy discourse,” Political Communication, volume 18, number 2, pp. 183–213.
doi: https://doi.org/10.1080/105846001750322970, accessed 14 August 2019.
Michael Schudson, 2001. “The objectivity norm in American journalism,” Journalism, volume 2, number 2, pp. 149–170.
doi: https://doi.org/10.1177/146488490100200201, accessed 14 August 2019.
Dominic Spohr, 2017. “Fake news and ideological polarization: Filter bubbles and selective exposure on social media,” Business Information Review, volume 34, number 3, pp. 150–160.
doi: https://doi.org/10.1177/0266382117722446, accessed 14 August 2019.
Steen Steensen, 2011. “Online journalism and the promises of new technology: A critical review and look ahead,” Journalism Studies, volume 12, number 3, pp. 311–327.
doi: https://doi.org/10.1080/1461670X.2010.501151, accessed 14 August 2019.
Natalie Jomini Stroud, 2010. “Polarization and partisan selective exposure,” Journal of Communication, volume 60, number 3, pp. 556–576.
doi: https://doi.org/10.1111/j.1460-2466.2010.01497.x, accessed 14 August 2019.
Yue Tan and David H. Weaver, 2013. “Agenda diversity and agenda setting from 1956 to 2004: What are the trends over time?” Journalism Studies, volume 14, number 6, pp. 773–789.
doi: https://doi.org/10.1080/1461670X.2012.748516, accessed 14 August 2019.
Jian Tang, Ming Zhang, and Qiaozhu Mei, 2013. “One theme in all views: Modeling consensus topics in multiple contexts,” KDD ’13: Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 5–13.
doi: https://doi.org/10.1145/2487575.2487682, accessed 14 August 2019.
Neil Thurman and Steve Schifferes, 2012. “The future of personalization at news websites: L:essons from a longitudinal study,” Journalism Studies, volume 13, numbers 5–6, pp. 775–790.
doi: https://doi.org/10.1080/1461670X.2012.664341, accessed 14 August 2019.
Peter Van Aelst, Tamir Sheafer, and James Stanyer, 2012. “The personalization of mediated political communication: A review of concepts, operationalizations and key findings,” Journalism, volume 13, number 2, pp. 203–220.
doi: https://doi.org/10.1177/1464884911427802, accessed 14 August 2019.
Dingding Wang, Shenghuo Zhu, Tao Li, Yun Chi, and Yihong Gong, 2008. “Integrating clustering and multi-document summarization to improve document understanding,” CIKM ’08: Proceedings of the 17th ACM Conference on Information and Knowledge Management, pp. 1,435–1,436.
doi: https://doi.org/10.1145/1458082.1458319, accessed 14 August 2019.
Ian Warren, Andrew Meads, Satish Srirama, Thiranjith Weerasinghe, and Carlos Paniagua, 2014. “Push notification mechanisms for pervasive smartphone applications,” IEEE Pervasive Computing, volume 13, number 2, pp. 61–71.
doi: https://doi.org/10.1109/MPRV.2014.34, accessed 14 August 2019.
Janyce Wiebe, Theresa Wilson, Rebecca Bruce, Matthew Bell, and Melanie Martin, 2004. “Learning subjective language,” Computational Linguistics, volume 30, number 3, pp. 277–308.
doi: https://doi.org/10.1162/0891201041850885, accessed 14 August 2019.
Received 15 January 2019; revised 30 May 2019; accepted 2 August 2019.
“Topic polarization and push notifications” by Madelyn Rose Sanfilippo & Yafit Lev-Aretz is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
Topic polarization and push notifications
by Madelyn Rose Sanfilippo and Yafit Lev-Aretz.
First Monday, Volume 24, Number 9 - 2 September 2019