Dissemination Biases of Social Media Channels: On the Topical Coverage of Socially Shared News

In a marked departure from traditional offline media, where all subscribers of a particular news media source (e.g., New York Times) used to get the same news stories through printed newspapers, online news media presents multiple options for the readers to consume news. For example, the subscribers of a media source can get news directly from the news website, or from what their peers share over social media sites like Facebook and Twitter. It is, however, unclear whether there are any differences in the news disseminated on these different online channels. In this work, we analyze data from a popular online news media site (nytimes.com), and show that each of these different channels tends to highlight some types of stories more than other stories. We believe that consumers of online news as well as media organizations need to be aware of such differences in various online news dissemination channels. Introduction As the number of users receiving news via traditional offline methods (e.g., via print newspapers or weeklies) is in steep decline, online news media sites like nytimes.com and cnn.com are emerging as the primary sources of news for people world-wide. A recent survey by the Pew Research Center (Pew 2012) found that the proportion of Americans reading news on a printed newspaper halved to 23% in 2012 compared to 47% in 2000. On the other hand, 55% of the regular readers of The New York Times declared that they read the news stories online, similar to 48% of regular USA Today and 44% of Wall Street Journal readers. The wide-spread adoption of social media sites like Facebook and Twitter is fueling the growth of online news consumption further. In a separate survey (Mitchell et al. 2014), Pew Research Center found that around 48% of American Internet users got politics news via social media sites like Facebook, almost as many as those that got such news from local television channels. In a marked departure from traditional offline media, where all subscribers of a particular news media source (e.g., New York Times) used to get the same news stories through printed newspapers, online news media presents multiple options for the readers to consume news. For example, a Copyright c © 2016, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. user can directly visit the website (e.g., nytimes.com) and read the stories published therein. Whereas, another user may only read the news stories shared by her social contacts on Facebook or Twitter. Effectively these different news dissemination mediums become news channels for the users who can consume news from one or multiple such channels. Given the complex landscape of online news consumption, it is important to understand the differences in the news consumption process contributed by each of these different news dissemination channels. In this work, we investigate whether there are systemic differences in coverage of news stories over different channels. Specifically, we attempt to understand whether certain types of news stories are covered more in certain channels than others. For our study, we gathered extensive data from one of the world’s most popular online news media sites, nytimes.com (henceforth referred to as NYTimes). We collected all news stories published at NYTimes over a period of 6 months. We further gathered the stories which became most popular across different dissemination mediums (e.g., most viewed on website, most emailed, most shared on Facebook, most shared on Twitter) during this 6 month period. We analyzed the differences in the topical distributions of stories covered by different mediums. Our analysis demonstrates that there are significant differences in the topical coverage of news stories that gain popularity over different dissemination channels. For example, opinions and local stories (related to New York and US regions) tend to be more widely shared on Facebook, while business and world news stories tend to be shared more on Twitter. It is unclear whether the online news consumers are aware of such differences in the topical coverages of the different mediums from which they consume news stories. Additionally, designers of various news recommendation systems also use the popularity of news stories in different mediums as signals to rank / recommend news stories. Through this study, we want to spread awareness both among the news consumers and the designers of news recommendation systems, about the differences in news disseminated on various mediums. Finally, our work is an early attempt, and much future work still remains to be done on understanding the effects of the differences across the different mediums on the news consumers. Background and Related Work Comparing online and offline news media: There have been prior research on the coverage of news stories on the offline and online editions of media sources. For instance, (Althaus and Tewksbury 2002) investigated whether readers of printed newspaper and website of NYTimes get different perceptions of political news. (Quandt 2008) compared the distribution of articles per news sections in online news websites with printed newspaper and TV news channels, and found prevalence of news on national politics and economy consistently in all three mediums. Complementary to the above works, we compare different online mediums of news consumption, namely news websites, email, Facebook and Twitter, which was not explored before. Social media and propagation of news: As more and more people are relying on online sources for news, there have been several attempts to understand the flow of news stories on social media (Bhattacharya and Ram 2012; Jisun et al. 2011). There have also been studies on how different factors affect the coverage of news consumed by users. For instance, (Jisun et al. 2011) examined how indirect media exposure in Twitter expands the political diversity of news stories consumed by the users. Our prior work (Chakraborty et al. 2015) investigated whether coverage of trending stories can differ depending on the browsing habits of users. In this paper, we show that the coverage of news stories consumed by users can also vary with the medium which a user chooses to consume news from. Ideological segregation and filter bubbles: Researchers have investigated the impact of personalized search / recommendations on social media, where individual users get content based on their profiles (e.g., locations), social media neighborhood, past click behaviors, search histories, and so. The concern is whether such exposures increase ideological segregation (Flaxman, Goel, and Rao 2013) and filter bubbles (Pariser 2011). Our study here raises the concern that the social media channels selected by users to receive news may be implicitly filtering stories on certain topics. Dataset Used In this work, our objective is to understand how different news stories are covered across different dissemination mediums. We attempt to investigate this question in the context of one of the most popular news media site – NYTimes. Using the NYTimes developer API1, we collected all newsstories appearing on NYTimes during a period of 6 months, July – December, 2015. Overall, we collected 120, 231 distinct news-stories. Additionally, NYTimes API also returns sets of daily ‘Most Viewed’, ‘Most Emailed’, ‘Most Shared on Facebook’, and ‘Most Tweeted’ stories, all of which contain 20 news stories at a time.2 We collected all such stories returned by NYTimes API at 5-minute intervals. Table 1 shows the number of distinct stories that appeared in different sets during this period. developer.nytimes.com/docs/ For instance, the most emailed stories can be accessed at www.nytimes.com/most-popular-emailed. Type No. of distinct stories All stories published on site 120,231 Most viewed stories 3,008 Most emailed stories 2,667 Most tweeted stories 2,756 Most shared stories on Facebook 2,472 Table 1: Number of distinct NYTimes news stories that became popular in different dissemination channels during July – December, 2015 Each NYTimes news-story is published under a topical category assigned by the NYTimes site itself. Examples of some topical categories are Arts, Education, Politics, Sports, Science, and so on. We also gathered the topical annotations for every story using the NYTimes API. In this work, we compute the topical coverage of a set of stories (which have become popular on a particular medium) as the distribution of stories over these topical categories. Topical Coverage of Socially Shared News In this section, we compare the topical coverage of stories which are (i) most popular on the NYTimes website, i.e., most viewed stories, and (ii) most socially shared, which includes most emailed, most shared on Facebook, and most tweeted stories. To better understand the differences in the coverage between these two groups of stories, we also consider a baseline – the overall coverage of all stories published online at NYTimes. Figure 1(a) shows the topical coverages of most viewed, most socially shared and all published stories. Figure 1(b) shows a Venn diagram that represents the (non-)overlap between these three sets of stories. We observe that there are significant fractions of most viewed stories which are not most socially shared, and vice-versa. To further characterize the non-overlapping stories in Figure 1(b), we look at the topical coverages of stories which are either most viewed or socially most shared, but not both (shown in Figure 1(c)). From Figure 1(a) and Figure 1(c), we observe two interesting trends. First, the topical coverage of most viewed and most socially shared stories differs significantly from the topical coverage of all published stories, suggesting that online news consumers are expressing a preference for certain topical categories of NYTimes stories over other topical categories, both when viewing and sharing stories online. Stories on topics of broad interest like ‘Sports’, ‘Business’, ‘U.S.’ and ‘World’ are less viewed and socially shared, compared to their relative coverage in all published stories. At the same time, stories on topics of narrower (niche) interest like ’Health’, ’Fashion’, ’Science’, and ’Opinion’ are more viewed and shared socially than their share of all published stories. One possible explanation for the differences in the coverage is that users’ are more likely to view or socially share topical stories that are uniquely found on NYTimes compared to topical stories that can also be found on other online media sites.