Attention to News and Its Dissemination on Twitter: a Survey✩ Claudia Orellana-Rodriguez *, Mark T

Computer Science Review 29 (2018) 74–94

Contents lists available at ScienceDirect

Computer Science Review

journal homepage: www.elsevier.com/locate/cosrev

Survey Attention to news and its dissemination on Twitter: A survey✩ Claudia Orellana-Rodriguez *, Mark T. Keane Insight Centre for Data Analytics, School of Computer Science, University College Dublin, Ireland

article info a b s t r a c t

Article history: In recent years, news media have been hugely disrupted by news promotion, commentary and sharing Received 16 February 2018 in online, social media (e.g., Twitter, Facebook, and Reddit). This disruption has been the subject of a Received in revised form 3 July 2018 significant literature that has largely used AI techniques – machine learning, text analytics and network Accepted 11 July 2018 models – to both (i) understand the factors underlying audience attention and news dissemination on social media (e.g., effects of popularity, type of day) and (ii) provide new tools/guidelines for journalists to better disseminate their news via these social media. This paper provides an integrative review of Keywords: Computational journalism the literature on the professional reporting of news on Twitter; focusing on how journalists and news Digital journalism outlets use Twitter as a platform to disseminate news, and on the factors that impact readers’ attention Social media and engagement with that news on Twitter. Using the precise definition of a news-tweet (i.e., divided into News articles user, content and context features), the survey structures the literature to reveal the main findings on Twitter features affecting audience attention to news and its dissemination on Twitter. From this analysis, it then Journalism considers the most effective guidelines for digital journalists to better disseminate news in the future. News © 2018 Elsevier Inc. All rights reserved. Audience engagement Audience attention

Contents

1. Introduction...... 75 1.1. Online developments: News moves online & social...... 75 1.2. Emerging developments: From computational to citizen journalism...... 76 1.3. Structure of the review...... 76 2. The good & the bad of news on Twitter...... 77 2.1. The good: How Twitter enhances journalism...... 77 2.2. The bad: How Twitter undermines journalism...... 78 3. The anatomy of news-tweets & attention to them...... 79 3.1. The news-tweet: Definition & features...... 79 3.2. Attention to news-tweets: Definitions & distinctions...... 79 4. Features of news-tweets: Content, user & context...... 80 5. Content features of news-tweets...... 81 5.1. Headlines & what they convey...... 81 5.2. News categories & topics: Reader & journalistic aspects...... 83 5.3. Tweet elements: Hashtags, mentions, replies & URLs...... 84 5.4. Textual elements: Stand-out phrases, commentaries, quotes & memes...... 84 6. User features of news-tweets...... 85 6.1. Account ownership: Corporate versus individual accounts...... 85 6.2. Branding & reputational features...... 85 6.3. User features conveyed by tweeting style...... 86 6.4. Features reflecting demographics: Gender, nationality & ethnicity...... 86 7. Context features of news-tweets...... 87 7.1. Temporal aspects of news tweeting...... 87 7.2. Locational aspects of news tweeting...... 87

✩ No author associated with this paper has disclosed any potential or pertinent conflicts which may be perceived to have impending conflict with this work. For full disclosure statements refer to 10.1016/j.cosrev.2018.07.001. * Corresponding author. E-mail addresses: [email protected] (C. Orellana-Rodriguez), [email protected] (M.T. Keane).

7.3. The visibility of news tweets...... 87 8. Studying the interactions between features...... 87 8.1. A sample study on feature interactions...... 89 8.2. Structuring the diversity of research...... 89 9. Guidelines for news providers...... 90 9.1. Social media guidelines for journalists...... 90 9.2. Specific guidelines on audience engagement...... 91 10. Future directions...... 91 Acknowledgments...... 91 References...... 91

1. Introduction 1.1. Online developments: News moves online & social

In recent times, the emergence of the online world, and notably Throughout the 1990s, most major news providers moved social media, has transformed the news industry in many unfore- online establishing significant digital presences. For example, by 2 seen ways. Traditional news providers have been assaulted by one 1996, most major American newspapers – including U.S.A. Today, 3 4 disruption after another from people sharing news on social media, The New York Times and The Washington Post – had established websites and were publishing digital versions of their papers. In- to everyone becoming a potential ‘‘citizen journalist’’, to a tweeting evitably, this led to digital-only news providers; in 2000, the South- President that refuses to give traditional press briefings. Impor- port Reporter5 was launched as the first online-only UK newspaper. tantly, each of these disruptions are progressively undermining Such offerings continued apace with the emergence of digital-first the business model of these news organizations, damaging their news; for instance, by 2006, The Guardian was offering a ‘‘web traditional income streams and depleting their paying audience. As first’’6 service where news was first published online before it always, however, such disruptions present as many opportunities appeared in the physical newspaper. Nowadays, the online pub- as they do challenges; opportunities that depend on understanding lication of news typically precedes newspaper publication thus the dynamics of news in this brave new world of social media. becoming the first and, in some cases, primary source of most of If traditional news outlets are to survive then, critically, they the news that people read. need to understand the dynamics of news production, consump- In the 2000s, with the emergence of social media sites – Face- tion and dissemination in social media. Social media enable users book (2004), YouTube (2005), and Twitter (2006) – there was yet to create, read and share news content in a social way, via sites another turning-point in online news delivery and consumption. such as Twitter, Facebook, and YouTube. Twitter, in particular, has These sites enabled the real-time sharing of online news and gave attained a special status as the social media platform for news; it the audience the opportunity to interact with news providers, di- has become a venue where newsworthy tweeters, news consumers rectly. Furthermore, as these social media platforms were ‘‘always- and journalists converge to report, read, discuss and share the on’’, they fast became key fora for collating updates, gathering news. Hence, insight into the dynamics of news in this social media comments and reporting breaking news. Indeed, in many cases, world, relies on research that helps us understand the impact of they became the news themselves [4]. news on readers [1,2], the features of news that engage people [3], Of all social media sites, Twitter has emerged as the platform and the factors that drive the sharing of news on these social media for news [5]. Unlike other social media platforms (e.g., Facebook, platforms [4]. Instagram, Youtube) where news is mainly encountered as a side- In this survey, we focus on the dynamics of news on Twitter; effect of other interactions, many Twitter users specifically use specifically we review recent research on how news providers the platform to track news developments [4,6,7]. Accordingly, tweet about their news and on the features of these news-tweets this micro-blogging platform is favored by journalists and news that impact audience readership, attention, and engagement. Here organizations to live-tweet news events that are being broadcast we define audience readership as the community of Twitter users in real time or to share incremental reporting threads and other content [8]. that attend in any way to these news-tweets and attention as the Fig. 1 shows the top-25 newspapers, by worldwide circulation, passive attention (e.g., just reading or scanning a tweet) or active ordered by the dates on which they created their official Twitter attention (i.e., commonly called ‘‘engagement’’ where users re- accounts. The New York Times joined Twitter on March 2nd 2007, tweet, quote, mention, etc.) that these readers give to these news- being the first of the five most popular American newspapers to tweets.1 In the following sections, we provide basic definitions for use the service, swiftly followed by The Wall Street Journal (March the components of this ecosystem, as well as identifying newly- 31st 2007), and the German tabloid Bild7 (October 2nd 2007). Most emerging concepts. We also address the important issue of span- of the other newspapers in this list, created their official accounts ning the gap between these research findings and journalistic prac- between 2008 and 2010, though two other outlets – Rajasthan tice by considering practical guidelines for optimizing journalistic Patrika and Eenadu from India – did not officially join Twitter until use of social media platforms when publishing the news. As such, late 2013. the review aims to establish how the main findings of the literature might concretely inform the news industry’s response to their 2 http://static.usatoday.com/about/timeline/. current challenges. 3 http://www.niemanlab.org/2016/01/20-years-ago-today-nytimes-com- debuted-on-line-on-the-web/. 4 1 https://www.washingtonpost.com/apps/g/page/national/washington-post- The terminology used in the literature is often inconsistent and confusing. co-timeline/374/. Engagement is often used to refer to active interactions (e.g., re-tweeting) but 5 http://www.southportreporter.com/. these actions are also often labeled simply as attention (with no passive/active 6 distinction). Though we propose more precise definitions, throughout the paper, https://www.theguardian.com/media/2006/jun/07/theguardian. to be consistent with the literature we often use the less precise terminology pressandpublishing. ‘‘attention and engagement’’ to refer to user behavior. 7 http://www.bild.de/. 76 C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94

Fig. 1. Dates on which Top-25 newspapers (by circulation) created official Twitter accounts.

By the late 2000s, news providers were not only reporting news Such offerings popularized the concept of Data Journalism as the on Twitter, but, with the increasing popularity of the platform, intersection between journalism, design, computer science, and were also using it as a tool to post immediate updates from citizens statistics [16]. and eyewitnesses. Journalists had also started sourcing news from The importance of social media, and particularly of Twitter, tweets, some of which were posted from anonymous or unverified for news production and distribution became evident in 2010 Twitter accounts; this ‘‘publish first, ask questions later’’ strategy with the emergence of Citizen Journalism and Ambient Journalism. marked the coverage of certain events, such as the popular protests Citizen Journalism refers to the power of the audience to create in Iran in June 2009 [9]. Of course, in time, a cottage tech-industry and disseminate news, often as part of their direct participation has grown up around verifying and curating such Twitter sources in target, news events.10,11 Though the concept already existed (e.g., [10]). Now, different new outlets (e.g., BBC, NYT) and different by 2010, it reached a peak of popularity during The Arab Spring news platforms (e.g., TV, online, in print) compete using a wide (December 17th 2010–December 2012). The coverage of the Arab variety of strategies and journalistic styles played out on social Spring saw ordinary citizens shaping and spreading the news based media [11]. on millions of tweets from ‘‘young, urban, relatively well-educated Just over a decade after its creation, nearly all major news individuals, many of which were women’’ [17]. This Citizen Jour- providers use Twitter to both advertise their news articles and nalism concept has inspired the idea of Ambient Journalism, where promote interaction with their audience. And, while social me- Twitter is viewed as an awareness system that offers a ‘‘means dia platforms have clearly changed journalistic practice in many to collect, communicate, share and display news and information positive ways, they also appear to have changed news publication serving diverse purposes’’ [5]. Table 1 presents these emerging in negative ways. Many worry that the social aspects of Twit- types of journalistic practice comparing them in terms of the main ter undermine journalistic integrity; for instance, while a highly actors involved and the platforms on which their news is typically interactive journalist benefits from more engagement with their delivered. audience they may do so at the expense of appearing less objective by virtue of these interactions [12,13]. 1.3. Structure of the review

1.2. Emerging developments: From computational to citizen journal- In the last 20 years, the news business has experienced major ism changes brought about by the impact of online news and social media. The technologies used for news collection, production and These major changes in the news ecosystem have created new dissemination are currently in constant flux, as many new tools are conceptions of journalistic practice; changes that have re-defined being developed for journalists and newsrooms to address these the skill-sets of those who call themselves journalists (as in Compu- changes [8,18–20]. This paper reviews the literature surrounding tational Journalism) and, perhaps, widened the definition of ‘‘jour- the professional reporting of news on Twitter. Hence, we survey nalist’’ to include ordinary citizens (as in Citizen Journalism). how journalists and newspapers use Twitter as a platform to dis- Computational Journalism was coined in 2006 as a descriptor seminate news and the factors that affect readers’ attention and for the then newly emerging form of journalism8 that draws engagement with that news on Twitter. We also consider whether on innovations in computer science and informatics. It changes any practical advice for journalists and news professionals can be ‘‘how news stories are discovered, presented, aggregated, mon- gleaned from this research, to optimize the use of Twitter for news. etized, and archived’’ largely through the use of new computa- The survey covers five main topic areas: tional technologies [14]. Fig. 2 shows how different journalistic and news-publication processes can be enabled and augmented • Good and Bad of News on Twitter: which reviews the broad, by information technology ‘‘all while upholding the core values of positive and negative impacts that Twitter has had on journalism such as accuracy and verifiability’’ [15]. For example, journalistic practice and people’s consumption of news for the process of information gathering, information technology (e.g., real-time information gathering, emergence of fake has facilitated the finding and characterization of source experts, news; see Section2) the cross-referencing of eyewitness reports and the determina- • tion of originating source of information. As for the communica- The Anatomy of News-Tweets & Attention to Them: which tion/presentation process, journalists are now able to make inter- introduces the definition of a news-tweet and how attention active models or data that inform more than a static story. to one might be measured (see Section3) The Guardian’s Datablog was introduced in 2009, as a new service that specifically gleaned news stories from data analysis.9 9 https://www.theguardian.com/news/datablog/2011/jul/28/data-journalism. 10 http://www.hypergene.net/wemedia/weblog.php. 8 http://www.gatech.edu/hg/item/182791. 11 http://archive.pressthink.org/2008/07/14/a_most_useful_d.html. C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 77

Table 1 New & old types of journalism, the actors performing them and the platforms on which their news is typically delivered. Style Actors Platforms Citizen Journalism Public Web, social media Traditional Journalism Professional journalists Web, newspaper Computational Journalism Professional journalists, data scientists Web, social media, newspaper Data Journalism Professional journalists, data scientists Web, social media, newspaper Ambient Journalism Public, professional journalists, data scientists Web, social media, newspaper

Fig. 2. Journalistic and publishing processes impacted by computing. Source: Adapted from [15]

• Features of News-Tweets: which distinguishes the main com- Real-time information gathering & validation. During rapidly- ponents of news-tweets – content, user and context feature- changing news events, Twitter can serve as a real-time monitor categories – and then maps the literature into these feature- for journalists and first responders (e.g., during natural disasters, categories (see Sections4–8) disease outbreaks, elections or civil disturbances). For example, • Guidelines for News Providers: which takes a step back from maps of weather events can be built from crowdsourced infor- the Twitter-analytics literature and tries to consider how its mation, by asking people to report how much snow they have in their backyards during a snowstorm [22]. Twitter has also been findings might be used to develop guidelines for journalistic used as an early warning system for disease outbreaks, based on practice, to improve the impact of tweeted news (see Sec- people reporting their symptoms (e.g., the 2011 e-coli outbreak in tion9) Germany [23]). It has also been used as a direct source of news • Future Directions: which sketches some fruitful directions in elections, where politicians’ tweets themselves become the for future research (see Section 10) subject of news reports [24]. Notably, journalists also use Twitter to perform real-time news collection by questioning eyewitnesses using tweets. For example, during the Arab Spring, journalists 2. The good & the bad of news on Twitter made extensive use of Twitter to satisfy a variety of information needs: to check facts, to find information sources, and to solicit Most digital innovations impact our daily lives in good and bad and clarify opinions [25]. For breaking news, Twitter can be seen ways and the use of Twitter for news is no exception. Many positive as a distributed sensing platform, through which eyewitnesses – effects have arisen from Twitter’s arrival in the news ecosystem; people in situ during the event – share their experiences, verify for instance, journalists can now track breaking news more effi- details of the event and provide personal opinions [10]. ciently and directly [21]. However, many negative effects have also A conduit to news websites. A large number of tweets posted by arisen; for instance, arguably, Twitter has undermined some of the news outlets and journalists provide links to a news organization’s key tenets of traditional journalism. It is important to appreciate website or to other news sites.12 Twitter can be used to direct traffic these wider aspects of the news ecosystem, to contextualize the to a news-provider’s home-page while making it easier for users to technological innovations arising from the use of Twitter before share one’s news with followers, leading to wider dissemination moving to considering the literature in detail and what it has found. (e.g., by retweeting) [26]. It has been shown that having a Twit- ter presence increases traffic to a news-provider’s website [27]. 2.1. The good: How Twitter enhances journalism However, this behavior is changing in several ways. Increasingly, readers are moving away from accessing news via the homepages The use of Twitter for news has transformed many aspects of of news providers in favor of accessing news via social media journalistic practice in positive ways. Twitter has become an im- and news-aggregator sites (e.g., Facebook, Twitter, and Google- portant tool for journalists by (i) supporting real-time information News) [28]; between 2013 and 2015, the number of Americans gathering and validation, (ii) creating a conduit to websites for accessing news via Twitter increased from 52% to 63% [29]. People news articles, (iii) providing a self-selected, target audience, (iv) engaging audiences in news commentary, and (v) enabling self- 12 http://www.journalism.org/2011/11/14/how-mainstream-media-outlets- promotion and branding. use-twitter/. 78 C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 are also accessing news on Twitter by following news categories; were found to be more ‘‘open’’ to posting tweets that mixed news while the breaking news category has been used in this way for with opinions and personal life stories [35]. Similar behaviors have some time, people are now following other categories such as been observed in female journalists, who in contrast to their male national government and politics, international affairs, business and counterparts, tend to be significantly more transparent about their sports [29]. This trend has led to the category-based promotion of activities and themselves, leading to increases in relatedness and news articles by some news providers. Indeed, it is now common audience engagement [36]. practice for news organizations to use several Twitter accounts to post tweets reflecting different news categories. As of 2011, The 2.2. The bad: How Twitter undermines journalism Washington Post, The New York Times, and The Wall Street Journal each had more than 70 different Twitter accounts, covering a wide The use of Twitter for news can also negatively impact the range of news categories.13 Furthermore, recent work has shown provision of news and journalism. Twitter can have negative im- that providing links to external sites (e.g., other news outlets) to pacts on news publication by: (i) making journalists act in more support ‘‘story-focused reading’’ – where users read the same story subjective ways, (ii) increasing the perception of bias in the news, from multiple sources – results in higher user engagement with an (iii) facilitating the emergence of filter bubbles, and (iv) supporting news-provider’s home site [30]. the spread of fake news. Provision of a self-selected, target audience. As a social media plat- Making journalists more subjective. Traditionally, journalists are form, Twitter delivers a self-selected, target audience for jour- meant to be ‘‘committed observers’’ which means that though they nalists and news organizations; an audience that is known to be are part of the community, the need to maintain a detachment interested in either them or their news or both. Recent research has from that community to maintain a perspective on events [37]. shown that this audience is a younger demographic that routinely Traditional journalism attempts to draw a clear line between ob- uses social sharing. The 2016 Reuters Digital News Report [28] jective descriptions of events and subjective evaluations of these found that younger people (ages 18–34) in developed economies events; a distinction that was enshrined in the separation of news (i.e., the U.S., Germany, France, Japan, Ireland, UK, and Sweden) and op-ed sections in traditional newspapers. However, Twitter use their smartphones as the main device for accessing news. encourages journalists to act in more subjective ways undermining Indeed, this demographic segment tends to rely on social media traditional journalistic norms of objectivity [38–40]. People use sites as their main news source, checking these sites repeatedly Twitter to engage with a journalist or news provider, directly; in throughout the day (as often as five times a day). Also, the ‘‘con- the same way that they follow famous people with whom they cise delivery’’ of news in Twitter may be an attractive feature would normally have no direct contact [41]. In this medium, social for this audience. With limited characters per tweet, journalists interaction and personal disclosure are the norm and one-to-one provide this audience with a preferred, stripped-down summary interactions are not only encouraged but expected [13,42]. On of the news (typically, a headline or link text). A recent analysis of Twitter, the journalist is actively pressured to be more subjective; the retweets received on two Financial Times’ corporate accounts they are expected to converse with their readers and express 14,15 (i.e., @FinancialTimes and @FT ), showed that the account that themselves more freely. Indeed, Twitter users tend to pay more tweeted the news with headlines alone, attracted higher levels attention to tweets reflecting such interactions (e.g., mentions: of sharing than the one tweeting news with the headline plus @user). Readers also prefer to follow and interact with accounts commentary [31]. belonging to individual journalists (e.g., @AnnaHolmes) rather than Audience engagement in news commentary. News distribution is those that are official corporate accounts (e.g., @nytimes), showing no longer a one-way street where journalists and news providers the personalized nature of this type of news consumption [34,41]. push news out to passive readers. Twitter has made news dissem- The perception of bias. Twitter’s subjectivization of the journalistic ination a three-way interaction, transforming it into a participa- role has a number of unfortunate side-effects, most notably an tory activity where users interact, engage, ask questions, express increase in the perception of bias. With Twitter, the context for viewpoints, have conversations, not only with news providers but, news publication has changed radically from the closely-edited, also, with other readers [22] (i.e., active attention, see Section 3.2). objective delivery of news to a loosely-edited, real-time personal- Twitter now represents a medium through which journalists and ized interaction [43]. So, just like non-journalistic tweeters, a high audiences can bond. For journalists, these interactions have created proportion of journalists’ tweets deal with personal life stories, a new channel for feedback on their work. For audiences, these humor, and opinions and not the news per se [36,44,45]. Hence, interactions have created a new sense of being part of a dialogue they come to be perceived as ‘‘just’’ another member of a news- about the news, possibly increasing their motivation to consume sharing community. This more informal delivery of the news, often it [12,32,33]. without source accountability or editorial inspection [45], strongly Journalistic self-promotion. Journalists can also use their profes- suggests that ‘‘the traditional conception of the objective journalist sional Twitter accounts to promote themselves and, even, to create does not apply on Twitter’’ [12]. Indeed, in [46], the authors have a personal brand. We have found that the top-6 most active Twitter recently proposed quantified measures of a media outlet’s political accounts, for news in Ireland, have engagement levels for individ- and socio-economic bias using Twitter data. ual journalists that sometimes outstripped the corporate accounts The filter bubble. Furthermore, arguably, Twitter also creates the of their employers [34]. It has been shown that the extent to conditions for yet another negative, social media effect on news; which journalists promote themselves depends on several factors, namely, the filter bubbles that emerge from, amongst other things, including the organization that employs them and their gender. following and sharing behaviors. Filter bubbles (aka ‘‘echo cham- Elite journalists (working for major news organizations) are less bers’’) refer to the narrowing of people’s perspective on current prone to use Twitter to create a personal image than non-elite events – created by social media features, personalization and journalists (working for ‘‘lesser’’ organizations); indeed, the latter recommendation – as they literally only see the news that confirms their prior beliefs [44,47–49]. So, the technology actively restricts 13 See Footnote 11. their exposure to a diversity of views, thus confirming their ex- 14 https://www.ft.com/. isting beliefs about current events. On Twitter, as in other social 15 Both accounts tweet at approximately the same days and times. media platforms, filter bubbles can emerge when users (including C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 79 journalists) follow, receive and share news from a few selected 3. The anatomy of news-tweets & attention to them sources; sources that are already aligned with their own viewpoint and interests, thus remaining unaware of information from other In the previous section, we outlined the board positive and sources, commentaries and events, that could provide a wider and negative impacts that Twitter has had on journalistic practice and more informed perspective [49,50]. For example, in the 2016 US people’s consumption of the news. In the current section, we turn Presidential election, an analysis of 1B journalistic tweets showed to the core literature dealing with the dynamics of professional that the majority of journalists followed Hilary Clinton rather than news publication on Twitter, beginning with definitions of the Donald Trump and based their opinions on official polls, leading news-tweet and the attention it attracts. to a serious underestimate of support for Trump in the electorate This review concerns itself with news-tweets (see Section 3.1); (see The Electome Project16). Indeed, it has been shown that many those tweets that are newsworthy or specifically designed to of the news audiences in the US election stayed within their filter convey news items. These tweets are quite different from the bubbles for much of the campaign (see MIT analysis17), which aligns millions of other tweets that discuss personal events, topics and with the idea that regular readers of partisan articles are almost happenings, that have nothing to do with the news (e.g., spam, exclusively exposed to only one side of the political spectrum [51]. birthdays, adverts and jokes). The review also addresses how news providers use such news-tweets to gain attention for their news Fake news. Twitter also, perhaps inadvertently, supports the and on what features of these tweets impact audience attention. emergence of ‘‘fake news’’. Fake News refers to fabricated arti- Broadly speaking, we use the term attention to cover all forms of cles presented as news, that convey deliberate misinformation or engagement with a news-tweet whether that is simply looking hoaxes, designed to either attract attention or mislead an audi- at/reading the tweet (i.e., impression) or actively engaging with it ence [52–54]. Many major cases of fake news have emerged in re- (i.e., by liking, quoting, or retweeting it). However, we also advance cent elections with a view to influencing voters, leading to regula- more precise definitions for several different types of attention (see 18 tory steps being taken by some governments : including, the 2016 Section 3.2). U.S. presidential election [52], the 2016 Italian Constitutional Ref- erendum [49], the 2017 French presidential election [55] and the 3.1. The news-tweet: Definition & features 2017 German federal elections [56]. While the phenomenon of fake news is not new (e.g., the ‘‘Yellow Journalism’’ of the Hearst era) it News tweets are tweets by journalists that, typically, (a) link to has gained a new life with the advent of social media platforms. articles they have written, (b) link to articles from other sources, or Twitter has contributed to the fake news phenomenon in two (c) include a journalist’s commentary on news items or tweets by main ways. First, by encouraging more personalized interaction non-journalists that report significant events/comments/opinions with news providers, Twitter has increased the subjectivization that are news (e.g., by eyewitnesses, celebrities, or politicians). of journalism, leading to the perception that many news sources Fig. 3 shows actual examples of these three main types of news- are biased, blurring the lines between ‘‘fake’’ and ‘‘real’’ news. tweet. Second, the social media functionality of Twitter (e.g., following, All of these news-tweets can be divided into three categories retweeting, liking) allows the rapid spread of fake news articles of features; content, user, and context features (see Fig. 4). Content before they can be debunked; as the constant repetition of the fake features refer to what is written in the tweet itself; for instance, news item lends it a spurious truth. For example, Shao et al. [57] the topic of the tweet (e.g., sports, politics, or education) and/or found evidence that social bots (software-controlled profiles or its inclusion of hashtags or URLs or other media (e.g., photos or pages)19 support the spread of fake news as they are early spread- videos). User features are those related to the user who posts the ers of claims and tend to target influential users, though they tweet: the gender, number of followers/followees, the volume and may not spread fake news any faster than human spreaders [54]. frequency of his/her tweeting, the status of the user (e.g., verified However, the ways in which rumors come to be propagated or or unverified, personal or corporate account), or the news topics debunked on Twitter are quite complex [54,60–63]. Early research he/she discusses in the tweets. Context features relate to the condi- found that Twitter does well in debunking inaccurate information tions under which the tweet is posted, for example, time and day and scotching rumors [60], though other findings question this of publication, location of the author, or geographical focus of the conclusion [54,61]. Recently, more temporally-focused work has tweet. These features hold for both original tweets and retweets, found that while the overall tendency is for users to support unver- even though the content-part of the latter may itself contain an ified rumors in the early stages, there is a shift toward supporting embedded tweet. We have found this tripartite division of the true rumors and debunking false rumors as time goes on [62,63]. features of news-tweet to be a good organizer for the literature in Indeed, journalists appear to play a key role as power users in the field (see Section4). this debunking process [64,65] Having said this, a recent study has found that ‘‘falsehood diffused significantly farther, faster, deeper, 3.2. Attention to news-tweets: Definitions & distinctions and more broadly than the truth in all categories of information, and the effects were more pronounced for false political news than Every news-tweet competes for the limited attention of users for false news about terrorism, natural disasters, science, urban depending on where it falls in their respective timelines [66,67]. legends, or financial information’’ [54]. So, while Twitter is not When a news-tweet is posted, assuming it is seen, it may be solely responsible for fake news (indeed, Facebook may play a attended to in several different ways. The attention to it can be a larger role in the phenomenon), it has certainly contributed to the passive attention, where it is simply scanned or read by a user, or conditions that facilitate its occurrence and promotion. an active attention, where it is actively engaged with by users, by liking, retweeting or quoting. Indeed, very few tweets attract active attention; attention to news-tweets, as measured by number of 16 http://www.electome.org/. retweets has been found to follow an exponential distribution, 17 https://news.vice.com/story/journalists-and-trump-voters-live-in-separate- with a few tweets receiving a lot of attention and most tweets online-bubbles-mit-analysis-shows. receiving no attention in a long tail (e.g., see Figs.5a and5b, [34]). 18 http://www.politico.eu/article/germany-election-campaign-fake-news- angela-merkel-trump-digital-misinformation/. Furthermore, these two classes of attention can be further sub- 19 Notably, this emergence of social bots has made the ‘‘author profiling task’’ an divided into primary attention and secondary attention. Consider important challenge for the community, see e.g. [58,59]. each type of attention, in turn. 80 C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94

Fig. 3. Examples of three types of news-tweet: where a journalist refers to (a) their own news article, (b) another journalist’s article, and (c) a third party comment/opinion/viewpoint.

active attention also exposes the tweet to those who follow that user; thus, increasing the size of the audience for the news item and, possibly, attracting further attention to the post, either in the form of more retweets and likes, or even in new followers for the tweet’s author [69]. Users may also actively react to news- tweets by engaging in conversations with journalists by either mentioning journalists in their tweets or replying to journalists’ posts. In many cases, journalists themselves start these conversations, leading to more active attention, especially from users who favor participatory journalism (i.e., journalistic environments as ‘‘social systems’’ in which audiences participate actively [70]). All of these engagement actions potentially promote the popularity of a given news-tweet and, hence, its visibility to other users, increasing the likelihood of receiving more attention from a wider audience. Fig. 4. Types of tweet features: content, user and context. Primary attention refers to the passive or active attention of a user to an original posting of a news-tweet. It concerns the user’s direct evaluation and potentially active response to the news- Passive attention can be operationalized in terms of impressions, tweet in its original form as posted by the journalist (see Fig. 6). based on a simple count of the times a news-tweet is shown to Secondary attention refers to the passive or active attention of a user.20 Impressions indicate that the tweet has appeared in a a user to a tweet referring to an original news-tweet; that is, the user’s timeline or in search results without necessarily attracting reading or engagement with a tweet that has been liked, retweeted further active engagement (e.g., likes or retweets). In analyzing or quoted by any user, other than the original tweet’s author. impressions, the amount of attention that a reader has given to Secondary attention, is bound in some ways, to be different from a tweet is unclear; that is, it is not known whether it was just primary attention as the original news-tweet is now being presented in a new context (see Fig. 6). For example, if @nytimes posts visually scanned as a user scrolls through their tweets, or read a news-tweet about an article on the US Presidential Election and quickly or carefully considered. As such, passive attention does not Hilary Clinton retweets it with some comment, known in Twitter tell us much about a user’s response to a news-tweet. Notably, 21 as a quote, then the original tweet is framed in a new context, a passive attention is the modus operandi for most Twitter users. context that would be very different to one in which Donald Trump Although reading a news-tweet is indeed a signal of attention, for quotes it with a different comment. Understanding the dynamics of wider dissemination, active attention is much more desirable, as it primary attention to news-tweets is likely to be significantly easier broadcasts the news to other Twitter users [68]. than understanding the dynamics of secondary attention; though, Active attention refers to the explicit engagement of a user with it should be noted that, an understanding of the latter is critical to a news-tweet (by liking, retweeting or quoting); it is much more explaining dissemination and diffusion. informative about the user and important to dissemination. This 4. Features of news-tweets: Content, user & context

20 https://support.twitter.com/articles/20171990. In the previous section, we clarified the definition of a news- 21 http://guardianlv.com/2014/04/twitter-users-are-not-tweeting/. tweet and described how it can be cast a collection of content, user C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 81

Fig. 5. Distributions of active attention to ∼1.1 M news-tweets posted by Irish journalists between August-2015 and January-2016; shown as (a) rank-ordered frequency of retweets to each tweet and (b) natural log of these retweet frequencies. Source: Figure extracted from [34].

content, a reader’s knowledge of the tweet’s author and the overall context act together to influence people’s behavior. For any given reader, the deployment of attention will depend on filtering out some stimuli, balancing other stimuli and, perhaps, assessing their emotional significance [76]. However, ultimately, attention to a news-tweet must depend on various facets of the tweet itself. These facets can be sub-divided into three categories of features: content features based on the news-tweet’s content, user features based on the news-tweet’s author and context features that capture the context surrounding the posting of that news-tweet (see Fig. 4). These classes of features have the potential to impact (i) primary or secondary attention (though, the literature primarily addresses the former) and (ii) passive or active attention (though, the literature has primarily addressed the latter because it provides more visible Fig. 6. The original news-tweet produced by a journalist receives primary atten- output measures). tion from several users, some interacting passively (e.g., reading) others actively Table 2 summarizes these three main categories of features, (e.g., re-tweeting, shown as white circle with new tweet). These active attention interactions then enter the sphere of secondary attention where users may, in turn, lists the particular features within each category and references passively or actively attend to them. the literature that addresses different aspects of the feature in question. It is probably the case that this list of features does not exhaust all the relevant factors affecting attention to news- tweets, but it does present those that have been identified, to and context features; we also considered some key definitional date. In the following sections, we survey the specific studies that distinctions in the notion of attention (e.g., passive versus active, have been carried out on how these features impact attention primary versus secondary). In the current section, we begin to con- to news-tweets, before considering the literature that addresses sider how the literature might be organized using these feature- categories of the news-tweet. interactions between these features. There is now a substantial literature on the dynamics of professional news publication on Twitter. However, it is hard to get 5. Content features of news-tweets a coherent picture of what has been discovered, because this literature addresses a babel of different tasks: from predicting As we have seen, there are there main categories of features popularity [31,34,67,71], to tracking the news life-cycle [66] and that can be explored in understanding attention to news-tweets – topic coverage [34,72], to identifying gender imbalances in jour- content, user and context features – into which we can map the nalistic environments [73,74], and the role of journalists and news existing literature (see Section4). Content features refer to what audiences in social media [75]. Hence, there is a pressing need to is written in the news-tweet itself; for instance, the topic of the put some structure on the literature to understand its findings and news-tweet (e.g., sports, politics, or education), or, indeed, its use to identify clear recommendations for journalistic practice. Our of hashtags or URLs, or embedded media (e.g., photos or videos). approach to putting some shape on the findings from this work, There is an extensive body of research examining a wide range of rests on two steps: (i) focusing on ‘‘attention’’ to news-tweets as specific content features (see Table 2). We consider each of these a fundamental phenomenon of importance (acknowledging the features, in turn, summarizing their main findings for attention to different sub-categories of attention, see previous section), and (ii) news-tweets. partitioning the literature based on the different feature-categories of news-tweets that impact attention in different ways (namely, 5.1. Headlines & what they convey content, user and context features). In an environment where new posts and updates are constantly In the age of social media, headlines have become very impor- emerging, news providers have a narrow time-window in which to tant, as they may be the only visible artifact that gives readers attract attention to their news-tweets. The attraction of attention access to a news provider’s articles; that is, increasingly readers to news on Twitter is a complex process in which a news-tweet’s only encounter news items in social media (indeed, in many cases 82 C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94

Table 2 Features that impact the tweeting of news studied in the literature. Type Feature Related research Content Headlines Social deviance and sensationalism [75,77], News worthiness, values and style [78,79] News category & Topic Sports journalism [80], breaking news [81–83], political journalism [84], financial news [85], all news categories [34], popularity and overlap [72] eventful vs. non-eventful topics [86], real world events [71], climate change [87], topic coverage [88], disaster events in rural contexts [89], riots [82], personal, informal or formal content [90], serendipity [91], soft vs. hard news coverage [92], broadcast vs. targeted news [93] Tweet Elements Hashtags [34,71,72,94], mentions and replies [31,34,72,92], URLs and propagated web-links [68,95,96] Textual Elements Standout phrases [97], commentaries with personal interactions [90], quotes & memes for news dissemination [92,98,99], quotes with named entities and people [72,92] User Account owner Corporate vs. individual journalist account [81,93,100,101], trust and tweet lifespan [102], elite vs. non-elite news organizations [93], broadcast vs. targeted communication [81,93], event type [103], mentioning patterns [100] Branding & reputation Meformer vs. informer [71], personal brand creation [45,45], gatekeeping role [104,105], impartiality and nonpartisanship [105], emotions and personal voice [72], perception of journalists [13] Followers & followees Social media adoption, online readership and the size of newspapers’ social media network [27], social network structure [71], influence and number of followers [95] Activity volume Daily tweets [93,95] Journalistic process Accountability [45,105], elite vs. non-elite sources [93], source review [10], perception of journalists [13] Gender Gender, sharing and impressions [73], gender and journalistic transparency [36], follower bias [74] Nationality & ethnicity Arab and British journalists [93], Australian journalists [81], Irish Journalists [34], white males [74] Context Temporal aspects Time & date of publication, news crowds [95], visitation patterns [66], breaking news on Twitter [100] Locational aspects Geographical aspects of users and real world events [71], news recommendation [108], reporting with a crowd [89], events of local-interest [107] Visibility Visibility and divided attention [109]

it may be the only aspect of the article read [96]). As news read- Social deviance. Many of the effects of headlines on attention in ers on Twitter are exposed to a continuous stream of attention- Twitter rely on well-established practice in writing headlines from demanding headlines, from which they try to find the most rele- traditional news, such as emphasizing social deviance. Banner vant ones [108], these headlines are increasingly tailored to capture headlines on the front pages of newspapers were an innovation users’ attention [109,110]. Although news providers use different of the late 1800s; before then, front pages were filled with ad- strategies when posting news-tweets, the story headline + link verts, often, in very small type [112]. Traditionally, news headlines format remains the most commonly-used strategy [13,38]. This were fashioned to draw readers in, encouraging them to either way of presenting headlines is, presumably, designed to invite the buy the newspaper and/or to read the target article. So, there is reader to click on the link, to read the article and/or to like or to a long tradition of reporting social deviance in news headlines share it. In many respects, using headlines in this way in news- (e.g., the infamous tabloid headline ‘‘Freddy Starr Ate My Hamster’’ 23 tweets leverages tactics that have a long history in newspapers published by The Sun in 1986 ). Socially deviant news involves (e.g., using headlines with particular styles and content, such as items about violations to social or legal norms (e.g., robbery or 24 social deviance). However, news providers are increasingly adopt- homicide ). In Twitter, the effects of socially-deviant headlines ing more sophisticated strategies for headline use; for instance, The on audience attention have been analyzed with a view to predicting shareability. Diakopoulos and Zubiaga [75] analyzed 107,066 Washington Post now uses several different headlines for the same tweets, using the ⟨storyheadline⟩+⟨link⟩ format, posted by the eight story, which are A–B tested online to be personalized for different top U.S. newspapers (by circulation) with more than 100K Twit- segments of their readership.22 Furthermore, several research ef- ter followers between November 2011 and October 2012. Using forts have developed tools and models to suggest ‘‘good headlines’’ for journalists [110,111]. 23 https://www.thesun.co.uk/archives/tv/895099/freddie-starr-ate-my- hamper/. 22 https://agency.reuters.com/en/insights/articles/articles-archive/how- 24 http://www.aejmc.org/home/wp-content/uploads/2012/09/Journalism- washington-post-data-driven-product-development-engages-audiences.html. Quarterly-1991-Shoemaker-781-951.pdf. C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 83

Amazon Mechanical Turk (AMT)25 the tweets were labeled as ‘‘de- Notably, these news category effects are, perhaps, most clearly viant’’ or ‘‘non-deviant’’ (with deviant meaning that the headline in manifested in the ‘‘breaking news’’ category; where tweets about the tweet refers to an event that breaks social norms). This study sudden or developing events receive considerable attention.26 In found that newspapers have a preference for tweeting deviant situations where real-time updates on events are critically impor- headlines, especially tabloid newspapers (e.g., the NYPost and NY tant (e.g., earthquakes, floods, or sport events) news-tweets that Daily News). Furthermore, it was also found that news audiences accurately convey breaking news tend to gain traction on Twitter on Twitter pay more attention to deviant headlines than to non- before other social media (e.g., Facebook or Google Plus [83]). deviant ones, and are more likely to retweet them. More generally, Indeed, Keane et al. [86] have shown, by means of daily LDA topic the use of social deviance is another manifestation of the use of modeling of 350 K Reuters news articles and 2 billion tweets, that sensationalism, which is widely used across all news categories certain types of ‘‘eventy’ topics (such as Robin Williams’ death) (i.e., including ‘‘hard news’’ categories such as government affairs receive consistently higher levels of attention in Twitter. Notably, it and science) to attract online attention, even though readers do not has been shown that, compared to the news wires, Twitter exhibits necessarily respond to such sensational treatments [77,113]. a larger number of hyper-local or real-life events news-tweets that attract the attention from audiences in search of increased Newsworthiness versus style. Traditional news media has always event coverage [88,116]. Furthermore,these news audiences tend made judgments about the newsworthiness of events with a view to engage more with topics presented informally and that tend to to publishing or prioritizing stories. Accordingly, headlines have be more personal, reflecting a more reciprocal relation between been fashioned to convey this newsworthiness, reflecting so-called news values (i.e., prominence, sentiment, magnitude, proximity, readers and journalists [90]. surprise, and uniqueness). Indeed, Trilling et al. [114] have argued However, the effects of news category are not restricted to that newsworthiness is a good predictor of ‘‘shareworthiness’’. breaking news alone. Orellana-Rodriguez et al. [31,34], have shown Headlines have also been written with a distinct style represented that distinct attention behaviors occur across several news cat- by brevity, simplicity, and unambiguity. These features of head- egories: in analyzing a 2.9 M news-tweets crawl by 564 jour- lines have been analyzed to determine their effects on atten- nalists’ accounts they found, for example, that readers of sports tion [78,79,96,115]. In [78], Piotrkowicz et al. used headlines alone news-tweets responded differently at different times of the day to predict the popularity of news articles. They used a collection and week than readers of business news-tweets. In other work, 27 of news headlines from The Guardian and The New York Times using data collected from Feedzilla covering 31 news categories and their corresponding popularity scores in Twitter (tweets and (e.g., world news, business, sports and art), Bandari et al. [72] retweets) and Facebook (likes and shares). The authors partitioned found that technology-related and health-related news stood out. the features of the headlines into news-value and style features Technology-related news receives a higher average number of and using regression models found that style features on their own, tweets than other news categories, while health-related news achieve better performance than news values, suggesting that the appears to have a loyal, niche audience that is ready to tweet and style of the headline, independently of the article content, plays a share its links, even if the number of these links is lower than in greater role in capturing the attention of social media users (see other categories. also [79]). However, this does not mean that the content of the On Twitter, news agencies and their readers may often differ article is irrelevant; as we shall see in the next section, it appears in what they consider relevant or interesting; there may be a gap that the content features of a tweeted article also significantly between what newspapers publish online and what people share affect the extent to which it garners attention. in social media. Through their analysis of news and tweets on climate change, Olteanu et al. [87] showed that ordinary events 5.2. News categories & topics: Reader & journalistic aspects receive more coverage in Twitter than in mainstream media, and that individual actions from non-elite people ‘‘neither rich, nor Different readers have different interests, preferring some news powerful, nor famous’’ generate peaks of attention comparable to stories over others; indeed, such interests may manifest them- those involving important organizations or even governments. selves in broad preferences for one category of news over another Finally, it should be said that sometimes the content of a news- (e.g., sports over politics). These preferences seen in traditional tweet can capture attention, serendipitously, just because it is news consumption, carry forward into the domain of tweeted news surprising in some way (and not necessarily in a socially deviant and are reflected in new behaviors appropriate to this ecosystem. way). As readers search through news-tweets apparently non- Similarly, journalists working in diverse areas of news may mani- relevant, surprising content can attract users’ interest and capture fest different norms in their professional use of social media; that their attention [91]. is, sports journalists may tweet their news and interact with their Journalistic practice within news categories. Just as readers tend to readership in different ways than, say, business journalists. Overall, respond differentially to different categories of news, so too do these behaviors act to make news content and news categories journalists; specifically, journalistic practice in the use of Twitter important factors impacting attention to news. varies over content categories. Readers’ engagement with news categories & topics. In Twitter, In breaking news, Bruns [81] found that individual journalists readers’ preferences for a given news category are reflected in their covering a story on Twitter, adopted a more collaborative dynamic engagement with tweets related to topics covering similar events. in which several parties worked together to find and share the For example, by looking at the content of a user’s past tweets and at latest updates of the developing events. Along the same lines, those of their followees, it is possible to predict future engagement Vis [82] studied two journalists – Paul Lewis (The Guardian) and on sports, politics, or business topics. For instance, a user who Ravi Somaiya (The New York Times) who actively covered breaking normally tweets about a geographical area, is also likely to engage news about the UK summer riots of 2011 – finding that a type with news covering sports teams in that area [71]. This influence of of tweeting activity that is similar to live blogging, gained them news category has been shown for several different types of events, significant national and international attention on Twitter. including real world events [71], climate change [88], disaster events [89], and riots [82]. 26 https://www.forbes.com/sites/quora/2017/01/10/why-twitter-is-still-the- best-place-for-breaking-news-despite-its-many-challenges. 25 https://www.mturk.com. 27 www.feedzilla.com. 84 C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94

In sports news, using a combination of in-depth interviews and and do not necessarily reflect that a particular person is being content analysis of print and online articles, English [80] exam- referenced, but rather that the tweet refers to a topic that involves ined when and why sports journalists adopt Twitter. This work this person in some way. Replies are always in the beginning of found that promoting articles and engaging readers are reasons the tweet and reflect an intent to engage the referenced individual why some sports journalists start their accounts in the first place. in conversation. Mentions and replies are similar in that they Indeed, several journalists reported that having a large number are both explicit signals by a tweeter to engage journalists (or of followers led to more reader interaction that was helpful in other readers). Notably, it has been shown that their use varies sourcing ideas and opinions before deciding if a piece is suitable across news categories [34]. In [31], Orellana-Rodriguez et al. used for publication. In a similar way, financial news on Twitter has been tweet corpora and audience responses from two national news increasingly used as a direct line of communication with other ecosystems, namely Ireland (1.7 M news-tweets) and the UK (1.2 M journalists, governments, banks, politicians, and investors sharing news-tweets), to develop predictive models for the features of interests, analyses and priorities; thus, increasing attention from news-tweets that impact audience attention. This work found that readers, who consider Twitter a newswire for financial news [85]. tweets, with mentions or replies, significantly impacted public at- Although, it also seems that different news providers act somewhat tention independently of the news category; however, for business differently in certain news categories. In countries such as the audiences in Ireland, and sports audiences in the UK, news-tweets Netherlands and the United Kingdom, elite newspapers turn to with mentions/replies received more attention (as measured by Twitter to source ‘‘hard news’’ (e.g., politics, business and economy, retweets). Intuitively, tweets that include one or more mentions and health care), as well as ‘‘soft news’’ (e.g., human-interest and of known people, places, or organizations, attract more attention sports topics), to a lesser degree. For non-elite counterparts, the from news readers [92]. However, the relationship between the opposite occurs, as they pay more attention to soft news [92]. In presence of known entities and tweet popularity is not easily contrast, the tweeting of ‘‘political news’’ has been changed less as evidenced; researchers claim that for engagement-prediction tasks both journalists’ awareness and organizational norms bound the it is better to use this feature (i.e., mentions of entities) in combina- changes so they do not cause a major shift in traditional journalists tion with other features such as article source, news category, and practices [84]. Interestingly, in general, it has been shown in a subjectivity of language [72]. This multivariate aspect of popularity recent qualitative analysis of Norwegian news outlets, that soft and sharing has been a hallmark of more recent predictive models news items tend to be shared more than hard news items [117]. (see [118,119]).

5.3. Tweet elements: Hashtags, mentions, replies & URLs URLs. The final element used in tweets that impacts attention is the use of URLs; as clickable entities they provide a method for We have seen that the content of the tweet clearly impacts at- sharing access to the actual news article published by a news tention, just as in traditional newspapers. However, there are also organization or journalist. Click data associated with these URLs specific elements in tweets that impact attention to news-tweets, can be used to understand user interests, to measure influence promoting diffusion; namely, hashtags, mentions, and replies, as of the journalist/news-outlet involved, or to predict the number well as the URLs that users can click to get more details on news of clicks similar URLs (aka similar articles) will obtain in the fu- articles of interest. These elements are regularly used by many ture [68]. Furthermore, studying audience attention to URLs on journalists to gain greater attention for their news-tweets. Twitter, reveals that different news events attract groups of people Hashtags. User-defined tags, so-called hashtags – such as #Char- with different characteristics that can be exploited to identify re- lieHebdo – are added to news-tweets to make them easier for lated news events [95]. Click-Per-Follower (CPF; i.e. proportion of readers to find. In theory, if a hashtag clearly discriminates one followers that click on a URL/link in a tweet sent by a user) has been news-item from others, then readers are more likely to encounter used as a measure to distinguish differences in engagement for a news-tweet containing it, creating the opportunity for more en- different Twitter accounts. CPF shows that official accounts might gagement and subsequent diffusion. In general, the use of hashtags receive less audience attention, than those curated and shared on in all tweets is relatively low (typically, ∼ 20% of tweets). However, Twitter by other users (possibly unofficial) that attract significant in breaking news-events, tweets use hashtags more often (∼ 42% attention [96]. of tweets) [71]. As such, it is clear that, under certain conditions, hashtags are used by readers and journalists to promote engage- 5.4. Textual elements: Stand-out phrases, commentaries, quotes & ment with stories. While, in general, there is evidence that hash- memes tags promote attention and engagement, we have found few tests of their specific impact on news-tweets. However, Shi et al. [94] Apart from Twitter-specific elements, there are other iden- have shown that tweets with no hashtags or with the hashtag tifiable textual-elements in news-tweets that can be separately #news, attract equivalent numbers of impressions, engagement assessed for their impact on attention: namely, stand-out phrases, and URL clicks; while tweets with tailored hashtags generate more commentaries, quotes and memes. impressions, engagement and clicks and a higher engagement rate (0.86% vs. 0.47%). There is also a noted tendency for individual Standout phrases. Some phrases can act to set one news- journalistic accounts to use hashtags more in news-tweets than tweet apart from others, affecting their memorability. Danescu- corporate accounts; often leading to increased attention and en- Niculescu-Mizil et al. [97] found that memorable phrases are (i) gagement [34]. lexically distinctive (i.e., they include words that are less common but use a common part-of-speech structure), and (ii) general Mentions & replies. Like hashtags, mentions and replies are other (i.e., they can be easily applied in other contexts). These phrasal content elements of a tweet that are used to make news-tweets effects are akin to ones found in the psychological studies of text easier for ‘‘selected’’ readers to find. Mentions/replies direct a processing, where quite superficial characteristics of a text have tweet to specific people and/or create an awareness of their in- been shown to influence its assessment (e.g., [120]). volvement in an event (e.g., using a journalist’s user-name @Mary- Fitzger). Mentions and replies look very similar but they differ in Commentaries. As a social-media platform, Twitter bridges the their physical location within the tweet and, possibly, in the intent communication gap between journalists and news readers, and a behind them. Mentions can be inserted anywhere in the tweet growing number of tweets contain commentaries and interactions C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 85

[13,38]. However, the use of commentaries is not uniform across all accounts tend to be used to reach out to a specific journalist, to journalistic accounts. Focusing on two daily regional newspapers make personal contact with them, to converse with them and to in the UK, Canter [90] analyzed journalist’s tweets reflecting com- participate directly in the news cycle [81,93]. mentaries or interactions with the readers, finding that journalists Indeed, the way corporate and individual accounts attract at- gradually built relationships by engaging in conversations with tention can change depending on where a story is in the news their audience. However, some news organizations do not promote cycle [103]. In breaking news, before a given story is confirmed these types of exchange, and differences exist between individual by the majority of media outlets, people attend more to individual and corporate Twitter accounts; in general, corporate accounts journalists than to corporate accounts (e.g., especially when the manifest less interaction than individual journalists accounts. journalist may have been first to tweet this news item); but as soon as a news item is officially verified, people engage more Quotes & memes. For news to spread, its content should prove with corporate accounts, possibly seeking to add credibility to accessible, timely and relatable to news readers. Quotes may be their own tweets, when spreading the news to their network of excerpts from political speeches, declarations, or interviews, that followers [100]. can be inserted into tweets to give users a snippet of news. When News organizations with a strong Twitter presence appear to those quotes take on a variable quality they may become memes, have an authority, in themselves, that people rely on [101]. It has spreading further by virtue of their shareability. In some cases, been shown that the more social recommendations people get for a professional news providers use memes (or mutations of memes) particular news organization, the more likely they are to trust and to attract attention by including them in their news-tweets [92,98]. develop an interest in its news [33]. For certain news organizations Quotes or memes might also be specifically used in newspapers – such as the BBC, Mashable, and The New York Times – the URLs when reporting on news events that break on Twitter (e.g., declara- they tweet have a longer life, reflecting a greater engagement with tions from celebrities, athletes, the public and politicians; see [99]). their news-tweets; indeed, these tweets can spread for several hours (or even days) after the URLs first appear on Twitter. The 6. User features of news-tweets tweeted URLs of other organizations – such as Bloomberg, Wired, and Forbes – often have shorter lifespans, attracting lower levels of As outlined earlier, there are three main categories of features engagement [102]. that can be explored in understanding attention to news-tweets – Furthermore, some journalists leverage people’s recognition of content, user and context features – into which we can map the the reputation of news organizations by referring to them in their existing literature (see Section4). Beyond the features that make tweets [82], perhaps adding greater credibility while exposing up a news-tweet’s content, there are also user features that impact their posts to the larger audiences attracted to these organizations. attention and engagement, those that depend on who posts the Indeed, the tweeting behavior of journalists is, to some extent, tweet; so-called user features. User features’ center on the owner of shaped by the organization that employs them; journalists work- the Twitter account (i.e., whether it is owned by a corporate entity ing for elite news organizations tend to be more careful and adhere or individual journalist), the self-brand of the individual journalist, more to journalistic norms than their colleagues working for non- the tweeting style of the account and a variety of demographic elite organizations [93]. In fact, elite news organizations, such as factors (e.g., gender, nationality, ethnicity). The New York Times, periodically update policies for their journal- In considering user features, a fundamental division arises be- ists, to regulate the subjectivity of their Twitter interactions.29 tween corporate and individual journalists’ accounts; indeed, the news organization behind a specific corporate account is important 6.2. Branding & reputational features too (see Section 6.1). With respect to individual journalists’ accounts, there are a host of user features to do with how journalists We have already seen that Twitter, arguably, undermines jour- project themselves to readers and how they create a ‘‘brand’’ that nalism by making it appear more subjective (see Section 2.2). This attracts attention to their news-tweets (see Section 6.2). Simi- growing subjectivization of the news has emerged as journalists larly, the tweeting style adopted in an account may also impact present themselves in ways that are known to attract more atten- attention; for instance, in interacting with followers/followees, in tion and engagement. This behavior can be viewed as a form of self- the volume of news-tweets generated and in the way tweets are branding, where journalists are projecting an image of themselves posted (see Section 6.3). Finally, and more generally, some user to their readers, one that will attract greater attention to their features reflect demographic characteristics – such as gender, na- news-tweets. For example, journalists that interact with readers tionality and ethnicity – that have been shown to impact attention and that make self-disclosures attract more attention and engage- (see Section 6.4). ment, especially from younger audiences [13]. Having said this, some of the traditional virtues of journalism still matter. A journal- 6.1. Account ownership: Corporate versus individual accounts ist’s expertise and interest in the topic area is a good predictor of readers’ positive perceptions of them and of engagement with their Corporate Twitter accounts have distinct user-related features news-tweets, especially if the reader and journalist share common that distinguish them from individual journalist accounts, leading beliefs and interests [13]. to different patterns of attention and engagement that may some- Three specific tactics have been shown to be important in self- times depend on the organization behind the account. branding; self-reference, increased subjectivity and individualized Corporate accounts present the corporate-brand of the news curation. Self-reference in a tweet is one method used by journalists organization as a whole (e.g., @IrishTimes, or @nytimes), whereas to promote themselves. One distinct category of such tweets has individual journalist accounts present the personal-brand of a spe- been noted: meformer tweets [71]. Meformer tweets are those that cific journalist (e.g., @jstallman, @maryfitz). On average, corporate include users’ comments, opinions, and thoughts on a highly per- accounts have 39% more followers than individual accounts and sonal level [45]. Obviously, journalists using meformer tweets are 28 also tend to be added to more lists ; presumably, reflecting a belief taking a much more subjective approach to the news. However, the amongst users that following a corporate account will alert them to attentional and engagement benefits for a journalist of being highly all the news published by that source [81,93] In contrast, individual

29 https://www.nytimes.com/2017/10/13/reader-center/social-media- 28 A list is a curated group of Twitter accounts. guidelines.html. 86 C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 subjective remain to be proven. Indeed, the measured subjectivity Revealing the journalistic process. Journalists that adopt a tweet- of individual journalists has not been verified as a good predictor of ing style that reveals aspects of the journalistic process impact engagement [72]. Finally, the gatekeeping actions of the individual attention to their news-tweets; this is often achieved by including journalist may also help to shape their brand. Traditionally, news eyewitness evidence in news-tweets (e.g., pictures and videos) organizations have acted as gatekeepers, performing a curation or by revealing discussions with other journalists. This style of process for readers, as editors select stories and promote certain tweeting has been more apparent in news-tweets covering fast- views of events; though, digital gatekeeping is now performed moving breaking news; for example, in the Tunisian and Egyp- by individuals (journalists and non-journalists) and, even, plat- tian uprisings, journalists included user-generated multimedia or forms [104]. In Twitter, individual journalists perform a similar links to images in their tweets, to help readers follow developing gatekeeping, curation role that, again, may attract greater attention events [121]. The posting of tweets that are viewed as transparent to their news-tweets [74,105]. For example, readers may follow a and informative (e.g., with verified information) can improve a given journalist because they value their expert commentary on journalist’s standing and attract more attention to their work. events (e.g., following a Nobel prize economist’s commentary on This transparency is often conveyed by including ‘‘job talk’’ in the U.S. economy). Finally, journalists may take a more objective tweets (i.e., professional, job-related jargon), by discussions be- approach to the news, using so-called informer tweets, that include tween journalists/public-figures, and by providing links to relevant actual information about events of public interest [71]. external resources [45]. In certain contexts – such as events involving medical emergencies, disasters, fatalities, or floods – people are more prone to engage with tweets when these are easier to 6.3. User features conveyed by tweeting style find and the information comes from a trusted source, which does not necessarily have to be the author of the original tweet [122]. Apart from directly projecting their Twitter personality, jour- Indeed, increasingly, journalistic tools are being developed to sup- nalists may also project themselves through the way in which they port the verification of sources based on a tweets’ geographical tweet, their tweeting style. This set of user features is manifested in location, time of creation, and users social network [10]. (i) how journalists manage their followers and followees, (ii) the frequency and volume of their tweeting behavior, and in (iii) how 6.4. Features reflecting demographics: Gender, nationality & ethnicity much they reveal the journalistic process in their tweets.

Followers and followees. Many journalistic accounts build up sig- Apart from the individual features of a journalist that impact nificant follower and followee lists, presumably based on the sim- attention and engagement, research has shown that certain demo- ple assumption that the larger these groups, the higher the level graphic characteristics of groups of journalists are also important; of attention and engagement; that there will be a rich get richer notably, gender, nationality and ethnicity. effect as more followers and followees will see a news-tweet and Gender. News articles by female and male journalists posted on respond to it [95]. However, the dynamics of following are not Twitter receive different responses from audiences, responses that this simple; that is, higher levels of engagement do not necessarily can vary across news category. With respect to this feature, it is occur from larger groups of followers/followees, one must have important to appreciate that gender representation in different followers that become active spreaders of one’s tweets [68]. Indeed, categories of news, may not be balanced from the outset; that is, there is evidence to suggest that one can sometimes have too many many major news organizations, to begin with, have more male followers, leading to a decrease of awareness, engagement, and than female writers (e.g., from 143 K opinion articles published participation due to several factors, including cognitive overload by The Guardian between July 2011 and June 2012, only 21% were and higher noise (e.g., followers feel that interactions become less written by women). But, even when female journalists get to personal) which can demotivate the audience from spreading a tweet their news, they often receive lower levels of attention in news-tweet [71]. There are also significant biases around gender social media (in Facebook and Google Plus, as well as Twitter) For in the following behaviors that male and female journalists at- instance, in the Daily Mail, an article written by a female journalist tract [74]. in the sports section receives 33% of the impressions received by an Volume of tweeting. The volume of tweets delivered by a jour- article written by a male journalist [73]. Furthermore, even though nalistic account (e.g., number of tweets published in a given female journalists are more open and transparent in their news- time period) clearly has some impact on the levels of attention tweets than their male counterparts [36], characteristics that are and engagement that arise, though the causal relationship is not valued by audiences, these audiences still engage more with male- straightforward. Corporate accounts tweet at very high rates by authored tweets. Female journalists also suffer from follower bias, comparison to individual accounts [93]; but, as we have seen, these though systems to remediate this have been proposed [74]. Indeed, accounts play a special role in spreading news. It is important for in general, news items tend to be shared more by males (that are white) [123]. journalists to find a balanced volume of tweeting. On the one hand, tweeting too often may decrease the likelihood of a news-tweet Nationality & ethnicity. Other demographic factors have also being seen, as these will remain at the top of followers’ timelines been identified as being important in news tweeting and re- for less time [108]. On the other hand, being an active tweeter tweeting, notably nationality and ethnicity. There are differences raises awareness among the journalists’ audience, which may lead in journalistic-tweeting across geographical boundaries, perhaps to more engagement [34]. reflecting different cultural norms in different parts of the world. There also appears to be culture differences in tweeting styles We have already seen that Arab journalists tend to broadcast by journalists in different countries, presumably, based on some tweets and their audience seems to respond positively to that style assessment of what works. For example, Arab journalists tend to of tweeting; whereas English journalists tend to target tweets, broadcast their tweets (i.e., high tweeting volume and frequency, which gets positive attention from their audience [93]. In Australia, use of hashtags and links), and this behavior seems to attract a journalists tend to be more gregarious in their tweeting, being favorable response from their audience. In contrast, British jour- approachable, willing to interact, and to create ties to other people nalists are more likely to target their tweets, directly engaging with of public interest [81]. In Ireland, journalists tend to interact with their readers, to attract significant audience attention [93]. users as well, although this behavior varies depending on the news C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 87 category, as is the public engagement with their tweets [34]. Over- such analysis, it is clear that the geographical sources of tweets all, in promoting their news, journalists in different countries differ provide important information to news providers about reader in their use of sensationalism on Twitter [77,113]. In addition, in interests in different geographical locations [106,107]. Having said general, it has recently been shown that white, male users tend this, it should also be noted that, users’ interests can go beyond to be more active in terms of sharing news, suggesting that there their location, as they often interact with tweets involving news may be a propagation bias to news dealing with the interests of this occurring in places far away from where the users are located [71] demographic group [123]. (e.g., where people become involved in fund-raising for natural disasters in other countries). 7. Context features of news-tweets

We have argued that the three main categories of features that 7.3. The visibility of news tweets can be explored in understanding attention to news-tweets are content, user and context features (see Section4). We have seen All news-tweets appear somewhere in the Twitter timeline how the contents of a news-tweet and the features of the jour- over the 24-h of any day on which they are posted. In theory, nalistic user posting that tweet can impact attention and engage- all these tweets are visible. However, readers are operating under ment. Beyond these types of features, the final category captures the conditions or context in which that tweet is posted. Context attentional-limitations as they search for posts of interest [108]. features deal with the factors surrounding the posted news-tweet; This means that all the posts that are physically displayed may namely, the temporal aspects of the post, locational aspects of it, never become psychologically visible depending on the time, effort and the visibility of the post in the tweet stream. The same piece and attention of readers. Once a journalist has posted a news- of tweeted news sent by a given journalist may be responded to tweet, its visibility is affected by two opposing forces. On the very differently depending on whether it was read in the morning one hand, if a user (e.g., journalist’s follower) has many followees or evening by an audience; for example, in the morning it could be (i.e., more than 250) then she/he is increasingly exposed to stimuli a ‘‘breaking-news first’’, whereas by the afternoon it could be ‘‘old and thus new tweets remain visible for a shorter period of time and news’’. are harder to find. On the other hand, if many of the user’s followees tweet or retweet the same post, the probability of that tweet being 7.1. Temporal aspects of news tweeting seen, read and subsequently shared, increases [108].

The way an audience responds to different categories of news can change at different times of the day or, indeed, at different 8. Studying the interactions between features times of the week. Orellana-Rodriguez et al. [31], found that Mon- day, Tuesday and Thursday mornings are the best times to attract In the preceding sections, we have teased out the effects of retweets for political news-tweets, whereas sports-tweets receive three distinct categories of features – content, user and context greater levels of engagement on the weekends. Within a given day, features – on attention to and engagement with news-tweets. This the timing of tweets may also be crucial. For instance, breaking partitioning of features was adopted to develop a simple and clear news needs to be timely; posting a news-tweet a few minutes picture of these effects. However, in everyday news distribution on later than competitors may result in the loss of an audience for Twitter, all of these features will be interacting together sometimes that news item [66]. In the case of Osama Bin Laden’s death, three working in an additive way, sometimes in a subtractive way and journalists (i.e., @jacksonjk, @keithurban and @brianstelter), received significant attention to their tweets when they reported the other times in a neutral fashion. We have seen that several studies possible death of Osama; indeed, their breaking tweet convinced have been referenced under different feature-categories, when the audience of this event, before any official announcement was features interact with each other; however, there are significant aired [100]. In contrast, posting a news-tweet with a link to a long- issues in surfacing such interactions. read, in-depth article should probably be targeted at times of the There are two major issues that arise in attempting to under- day when readers are known to have the time to read such articles stand such feature interactions. First, there are very few controlled (e.g., during commuting times or at the weekends). studies that have specifically examined the interactions between Audience reactions to news-tweets at different times can reveal these variables. So, to provide a template for this type of work, how interest levels wax and wane for different news events in a we summarize one exemplary study on feature interaction (see population of readers. Such data can be used to fashion similar Section 8.1). Second, a very real problem that arises in consider- articles with a view to eliciting similar engagement patterns. Using ing feature interactions, is the diversity of tasks examined in the Twitter data around news stories published by the BBC and Al literature. Part of the fragmentation in the literature arises from Jazeera, Lehmann et al. [95] discovered that after a news-tweet is the very different questions that researchers have attempted to posted, people’s subsequent response (i.e., tweets or retweets) are answer; for instance, some have tried to predict the popularity correlated during a brief period of time, before dispersing again, of news on Twitter, others try to identify bias in online sources, and that similar news articles attracted similar crowds. others attempt to understand the factors that affect people’s engagement with real-world events, and to find the key groups in 7.2. Locational aspects of news tweeting advancing news stories on Twitter. Indeed, to complicate matters When a journalist is in situ at a news event and is tweeting further, these studies often use different data sources (e.g., news about it, more attention and engagement with that news-tweet articles, metadata, online traffic logs, distributions of social media will occur. The classic case of this phenomenon occurs when a interactions), different output measures (e.g., attention, types of tweeting journalist is an eyewitness at the event and is playing engagement, diffusion), and examine different platforms in their a key informational role for first responders and/or community work (e.g., Twitter, Facebook, Reddit). To understand this diversity, volunteers. In [89], the authors report on such a case during Hur- we show how some representative studies can be cast in terms of ricane Irene, when professional journalists in a rural community the features discussed here (see Table 3), to get a sense of how to successfully led an online volunteer community effort. On foot of understand feature-interaction in different tasks (see Section 8.2). 88 C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94

Table 3 Features explored in a sample of research papers showing the tasks used, findings made, and data analyzed. Title Task & findings Data Features (following our typology) The pulse of news in Predict popularity of News articles, metadata Content: news category, named social media: forecasting News on Twitter: and number of link entities; popularity [72] Attention can be predicted shares on Twitter; user: news source, subjectivity from news source, news category, language subjectivity, entities Transient news crowds Help journalists rapidly News articles plus Content: URLs, hashtags, topics; in social media [95] detect follow-up stories Tweets including those user: ratio followers/followees, to their articles: articles URLs number of followers/followees Users re-tweeting an article Tweets per day, favorite tweets can reveal ‘‘transient crowds’’, sources of follow-on news events Social media news commu- Identify bias in online RSS feeds and corres- Content: mentions, URLs; nities: gatekeeping, covera- News sources and social ponding corporate user: partisanship, number of ge and statement bias [88] media communities: Twitter accounts of Tweets; Selection, coverage & statement prominent news sources context: geographical region biases found in news sources, more geographical than political Newsworthiness and network Study the role of social Tweets with links to Content: deviant/non-deviant gatekeeping on Twitter: the deviance and the resharing news articles Tweets; role of social deviance [75] of news headlines by net- user: news organization work gatekeepers on Twitter: Twitter users show same preference for social deviant News items as journalists Predicting user engagement Understand the factors that Tweets and users related Content: directed and broadcast on Twitter with real-world affect people’s engagement to a list of real world, Tweets, hashtags, topics; events: [71] with real world events; events plus all tweets of user: total tweets, tweets/hour, Users’ prior Twitter activity those users in the last followers/followees, topical interests, geolocation, six-months meformer/informer tweets; social network structures context: geographical proximity correlate with engagement What journalists retweet: Explore what journalists Tweets and retweets from Content: topics; opinion, humor and brand retweet: Tweeting journalists 8 American journalists user: number of retweets, name, development on Twitter [45] often violate norms of News organization, description objectivity and independence Journalists and Twitter: Compare Twitter usage Tweets from journalists Content: URLs, hashtags, mentions, a multidimensional quan- by journalists, news and news orgs. replies; titative description of organizations and news user: number of followers; usage patterns [93] consumers: context: tweet source (mobile, Arab, English & Irish desktop, app) journalists have different tweeting strategies for news Sharing news articles using Examine how news Tweets with news Content: mentions, replies, URLs; 140 characters: a diffu- articles diffuse on articles’ URLs user: number of followers, sion analysis on Twitter; [102] Twitter: followers participation Network & temporal analyses show propagation patterns determining life-spans of news Breaking news on Identify the key Tweets covering the User: mentions, URLs, Twitter [100] groups in advancing first rumors of the News category the story of Osama News and president Bin Laden’s death: Obama’s speech Twitter broke death story then spread by a few ‘‘opinion formers’’ Characterizing the life Study the lifecycle News articles with Content: tweets’ entropy; cycle of online news of news articles over 3.6M visits and user: tweets and shares/min, stories using social posted online: at least 235 K social corporate accounts media reactions [66] Social media reactions can followers/followees predict traffic of article (Facebook + Twitter) within first 10 mins Modeling gender dis- Analyze differences in News articles plus Content: news category; crimination by online for online liking, sharing, number of likes, context: day of week News audiences [73] re-sharing of articles by male authors’ gender; & female journalists: for each article in Differences in engagements for Facebook, Twitter articles by men vs. women, and Google Plus women 1/3 rate of men’s rate C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 89

In predicting audience engagement, this study separates individual versus corporate Twitter accounts, and using regression analyses shows how a variety of content, user and context features operate in both of them (e.g., temporal, mentions, popularity). Fig. 7 shows, for the individual accounts, the relative importance of content, user and context features as they interact with the key content feature of news category (e.g., lifestyle versus sports). Fig. 8 shows, for corporate accounts, the relative importance of the same content, user and context features (n.b., as content categories are mixed up in corporate accounts, this variable cannot be easily disassembled). In the case of individual accounts, this analysis shows that, for example, sports news audiences tend to engage more with tweets containing mentions, URLs and hashtags, possibly in that order. Contextual features (i.e., time and date when the tweets are posted) are also important, perhaps reflecting an audience engagement that depends on the coverage of relevant sports events in a timely fashion. Also, sports news readers in Ireland value the popularity of the journalist who posts the tweets, engaging more with well-known journalists, that have built up a following and a solid reputation. In the case of corporate accounts, the analysis Fig. 7. Feature importance for five feature groups (columns) per news category shows that people do not tend to base their engagement primarily (rows) for individual journalist accounts in Ireland. Intensity of color indicates on conversational hints, such as mentions, but rather on the pop- higher importance. (For interpretation of the references to color in this figure ularity of the source, confirming that organization-level accounts legend, the reader is referred to the web version of this article.) serve mostly as a source of news, rather than having the social Source: Figure found in [34]. aspect seen in individual journalists’ accounts. Overall, this work shows that some of the interactions between different categories of features can be teased apart in a single study, perhaps providing a template for how such studies might be designed in future.

8.2. Structuring the diversity of research

Much of the fragmentation in the literature, arises from different researchers analyzing different tasks and different collections of features. In this section, we try to show how this diversity can be structured by dividing studies into their focal task, the datasets they use and the category-features they examine in their work. Fig. 8. Feature importance for five feature groups (columns) for corporate accounts A number of studies in the literature were sampled to show how in Ireland. Intensity of color indicates higher importance. (For interpretation of the they can be parsed into these different dimensions, allowing us references to color in this figure legend, the reader is referred to the web version of this article.) to align them in terms of their content, user and context features Source: Figure found in [34]. (see Table 3). In sampling these papers, we have selected ones that represent the diversity that exists in the literature. Hence, some works are relatively highly-cited [72,66] while others are not [93,73], some papers are relatively recent [93] whereas others 8.1. A sample study on feature interactions are older [72], and some examine very specific, narrow phenomena (e.g., social deviance [75]) whereas others examine very broad Few studies have systematically examined content, user and phenomena (e.g., life-cycle of news articles, predicting news pop- contextual features interacting together in a single task situation. ularity on Twitter). Overall, while the sample is small, we consider However, [34] is an example of one such study that explored all it represents some of the diversity that exists in the literature three feature-categories to examine attention to news-tweets. providing a good test of whether the proposed framework is useful. The authors found that news-tweets posted by Irish journalists A key differentiator between these studies is the task adopted; and news organizations, attracted different levels of engagement this plays a key role in selecting the dataset to be used and, perhaps, based on an interplay of content, user, and context features. This the features that are deemed to be important in the analyses work separated news-tweets by user features – e.g. corporate carried out. For example, in [102] and [34], both research groups (i.e., posted by corporate accounts (e.g., @IrishTimes) versus indi- explore attention to news-tweets but with a different task focus vidual accounts (posted by individual journalists) – and classified and, hence, use different features. In [102], attention is operational- news-tweets using the content feature of news category (using six ized as secondary attention with diffusion and social-network news categories, namely lifestyle, politics, sports, science and tech- features are examined (i.e., how tweets diffuse from one user to nology, breaking news, and business). Contextual features were the next in Twitter’s followers structure), while in [34] attention is also examined (such as, time and day of tweet posting). In total, operationalized as primary attention measured by retweeting, so 41 different features of the news-tweet were explored, including social-network aspects play less of a role. those mentioned above (e.g., the use of hashtags, mentions and One interesting observation that arises from this organization URLs, the gender of the journalist, etc.). The authors then examined of studies, is how researchers customize the features they use in how features combine to predict audience engagement (as mea- examining a focal task. For example, in order to identify how sured by number of retweets). journalists share news on Twitter [75], the authors concentrate on 90 C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 the news organization that posts the tweets, and on tweet-content The expression of opinions. Most news organizations explicitly in- classified as deviant or non-deviant. For other tasks, namely, pre- struct their employees on how to express opinions in their posts, dicting user engagement with real-world events [71], describing because of the impact this can have on attention and engagement. the usage patterns of journalists [93], identifying transient news AP remind their staff that ‘‘opinions they express may damage the crowds [95], and modeling how news articles diffuse on Twit- AP’s reputation as an unbiased source of news’’ and that they ‘‘must ter [102], a more extensive set of features is used (e.g., max. number refrain from declaring their views on contentious public issues in of tweets per hour or minute, number of followers/followees, total any public forum and must not take part in organized action in mentions in retweets, number of hashtags, whether the tweet is a support of causes or movements’’ [8]. AFP remind employees that meformer or an informer, or the entropy of the tweet vocabulary). their ‘‘words will be public and will be archived and referenced by search engines’’ thus they should ‘‘refrain from reacting in the heat of the moment; take time to write a considered cool-headed 9. Guidelines for news providers post’’ [125]. Reuters journalists are urged ‘‘to be responsible, fair and impartial’’ [124] when using social media accounts. In the previous sections, we have tried to surface the key factors in the literature that address what is broadly called ‘‘attention and The use of retweets. Agencies also give direction on retweeting as engagement’’ for news-tweets. A notable feature of this literature it may be interpreted as agreement, endorsement, or support for is that it mainly addresses ‘‘active attention’’ or ‘‘engagement’’ a post, even though they are not opinions per se. Indeed, the AFP (where users actively respond to the news-tweet), rather than advice is that ‘‘a retweet or link is often considered a sympathetic recommendation’’ and thus journalists should ‘‘only retweet alerts ‘‘passive attention’’ (where users read or scan the news-tweet). that have been tweeted from the company’s @AFP account or It is also clear that most of this literature is very much directed published by an AFP client’’ and should ‘‘avoid retweeting any in- at understanding/analyzing/predicting journalist and reader be- formation that could turn out to be problematic’’ [125]. AP propose havior on Twitter rather than generating guidelines for journalists that ‘‘a retweet with no comment of your own can easily be seen as about how to use Twitter. Granted, most of these studies have an a sign of approval of what you’re relaying’’ and provide examples ‘‘implicit’’ rider to journalistic practice that any identified factors of possibly problematic and acceptable tweets [8], as follows: could be exploited but, for the most part, the literature does not explicitly articulate these discoveries as guidelines. We believe • Possible problematic: ‘‘RT @dailyeuropean: At last, a euro plan that this is a significant blind spot in the literature and also a missed that works’’ – Retweet without commentary. opportunity. • Acceptable: ‘‘Big European paper praises euro plan. RT @dai- In this section, for completeness, we survey the general guide- lyeuropean: At last, a euro plan that works.’’ – clarify that it is lines for journalistic practice that have emerged in the media a simple report. industry (see Section 9.1); presumably informed, in some way, by the reviewed literature and the experiences of journalists ‘‘on the Reuters state that in retweeting ‘‘[Reuters’] journalists should be ground’’. As an illustrative case, we also show how data-analytic mindful of the impact their publicly expressed opinions can have studies can produce specific guidelines for journalistic practice.30 on their work and on Reuters’’ [124]. (see Section 9.2) Curation of following. Twitter’s social dynamics involve users following and being followed by other users. Journalists are public 9.1. Social media guidelines for journalists figures whose posts are followed to keep track of the latests news and, at the same time, they can follow other people themselves. The three largest news agencies in the world – AP (Associated Obviously, the latter could be viewed as a form of endorsement. Press), Reuters and AFP (Agence France-Presse) – provide their jour- Reuters advise that ‘‘by friending or following someone, they may nalists with guidelines for the use of social media. These guidelines, be giving out the identity of a source. Everything depends on our which are periodically revised and updated in order to keep up keeping trust’’ [124]. Because following a user may be seen as a with the changing dynamics of social media, typically address, at form of bias, AP express that ‘‘friending and ‘liking’ political candi- least, four aspects in the use of social media: (i) account creation, dates or causes may create a perception among people unfamiliar (ii) expression of opinions, (iii) use of retweets, and (iv) curation of with the protocol of social networks that AP staffers are advocates. following. Therefore, staffers should try to make this kind of contact with figures on both sides of controversial issues.’’; they, however, also Account creation. News agencies encourage journalists to open encourage employees to ‘‘feel free to ask their followers on social social media accounts, to identify themselves as journalists and networks for their opinions on news stories, or to put out a call for to clarify that the opinions posited in tweets are personal and witnesses and other sources, including people who have captured not those of the agencies. Reuters suggest that their employees photos or video that AP might want to authenticate and use’’ [8]. should ‘‘identify as Reuters journalists and declare that they speak Other general guidelines. Besides the above mentioned news agen- for themselves, not for Thomson Reuters’’ [124]. The AP instruct cies, several other news organizations, such as the BBC and The employees to ‘‘identify themselves as being from AP if they are Washington Post, also provide their employees with guidelines for using their accounts for work in any way... use a personal image social media usage. The BBC’s guidelines are summarized by two (not an AP logo) for the profile photo, and... identify as an AP main rules: (1) ‘‘whatever is published – on Twitter, Facebook staffer’’ [8]. AFP suggest journalists to ‘‘include a disclaimer on their or anywhere else – must have a second pair of eyes prior to Twitter profiles such as: ‘The views expressed here are my own. publication’’, and (2) ‘‘avoid you saying or linking to something Links and re-tweets are not endorsements’ or ‘On Twitter I speak unwise which could land you, or the BBC, in trouble.’’ [126]. The only for myself.’ ’’ [125]. Washington Post focus on (i) maintaining credibility: ‘‘refrain from writing, tweeting or posting anything – including photographs or video – that could be perceived as reflecting political, racial, 30 These proposals happen to be based mainly on own work, a key novelty of which has been an attempt to closely link data-analytics research to journalistic sexist, religious or other bias or favoritism.’’, (ii) avoiding con- practice. We have not found other studies that attempt to make such explicit links flicts, for example, ‘‘Post journalists should not accept or place in this way. tokens, badges or virtual gifts from political or partisan causes on C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 91 pages or sites’’, (iii) maintaining professionalism, (iv) promoting in engaging the audience. Temporal features are important for transparency, e.g., ‘‘clearly distinguish news from opinion when Ireland and the UK, possibly reflecting an audience engagement promoting or aggregating content.’’, (v) link curation, (vi) real-time that depends on the coverage of relevant sports events. The popu- reacting to news, e.g., ‘‘We should be conscientious of timing by larity of the journalists is of higher importance for Ireland than for reacting quickly to breaking stories. If news breaks on social net- the UK, in this category. Interaction with users through mentions works, notify your editor immediately.’’, and (vii) medium mindful is useful for gaining retweets; however, for news in the sports behavior, e.g., ‘‘credibility and influence in social media are tied to category, tweets with mentions are particularly valued. Monday your ability to actively participate’’ [127]. and weekends are the key days to gain attention, but in general, anytime seems good for sports. Posting tweets outside office hours 9.2. Specific guidelines on audience engagement may get more readers, but the difference is not clearly significant.

As we have seen above, most of the guidelines issued by news 10. Future directions agencies to their journalists are quite general in their recommendations. However, data-analytic studies can also produce quite The past decade has seen radical disruptions in the news busi- specific guidelines for journalistic practice based on the interpre- ness, in how news is produced, distributed and handled in journal- tation of the results from analysis and predictive modeling efforts. istic practices. One of the main drivers for these changes has been However, such proposals are not at all common in the literature. the emergence of social media and, in particular, the use of Twitter For example, [34] and [31] used regression analyses to predict as a distribution platform for news. the combination of content, user, and context features that leads The focus of this survey has been on reviewing and discussing to higher audience engagement, amongst Irish and UK readers, the distribution of news on Twitter, particularly, our interest lies across six news categories. These analyses surface some quite in understanding what makes news readers actively interact with general rules, along with specific rules, that could help journalists news tweets and pay attention to them. To this end, we presented make better use of Twitter. The general guidelines are relevant to different types of news posts found in Twitter, and discussed all journalists irrespective of the topic of their tweets, while the how computer science researchers dissect such tweets in order to specific guidelines are different depending on the news category. extract features that are later used in a variety of tasks. We present these proposals here as an example of how one might In the last part of this article, we took a detailed view of translate research findings into practical recommendations for three types of features of news tweets, namely content, user, and everyday use by journalists. context features, and examined how these are used together to General guidelines. In general, there appear to be two main facets address research tasks oriented at understanding both journalists of news-tweets that improve the attention they gain from audi- and news readers’ behavior and attention to news on Twitter. ences: Finally, we discussed efforts by both news agencies and computer scientists, directed at helping journalists and news organizations • Getting personal. Tweets that include mentions reflecting to take advantage of the possibilities that Twitter offers as a news direct interactions are well received by news audiences in distribution platform. Ireland and the UK. As journalists continue to embrace new platforms on which to • Rich content. Enriching tweets with hashtag annotations post their news, and audiences continue seeking these news in and URLs to media content helps to increase engagement. alternative media, we foresee two main directions for research: Adding URLs, however, shows lower impact in Ireland than • For journalism researchers, to continue studying whether including hashtags, which is noteworthy because at present and how traditional and modern, more social journalism many corporate account of news providers include URLs as can co-exist in social media, and to direct efforts toward links to articles they are publishing, whereas this finding using social networking platforms, such as Twitter, to their suggests that it does not in itself promote better audience. In advantage and, furthermore, that of the news consumers. the UK, the scenario is different, and URLs are slightly more important than hashtags. • For computer scientists, to aim at bridging the gap between outcomes of predictive or optimization models and journal- These guidelines emerge from collections of features, whereas istic practices. the more specific guidelines really address the effects of specific features. Interdisciplinary collaboration, where both journalists and computer scientists work together and communicate their needs, Specific guidelines. Particular aspects of a tweet impact audience expectations and learning outcomes, is necessary to advancing our engagement differently according to the news category. For exam- understanding of the factors that impact audience attention to ple, audiences of political and sports news, show different patterns news. of engagement with tweets. For political tweets, audiences are strongly influenced by con- Acknowledgments tent features, especially by mentions, but who posts the tweet is also important, particularly in the UK. In Ireland, the day/time of This work is supported by Science Foundation Ireland through posting the tweets is similar in importance as is the journalist’s the Insight Centre for Data Analytics under grant number popularity. For political news tweets, analyses of audience interac- SFI/12/RC/2289. The authors also acknowledge the advice received tions reveal that early in the morning (before 9:00 a.m.) is the best from The Irish Times during this project. time to gain retweets, and that having an audience of more than 50 unique retweeters attracts more readers; suggesting a ‘‘rich References gets richer effect’’ where journalists in the political arena need to develop a reputation that people follow or be identified as the [1] P.M. Napoli, Measuring media impact: An overview of the field, 2014. URL expert in a particular topic area. https://learcenter.org/pdf/measuringmedia.pdf. For sports tweets, audiences are strongly influenced by men- [2] A. Schiffrin, E. Zuckerman, Can we measure media impact? Surveying the tions, with URLs and hashtags playing a relatively important role field, Stanford Social Innovation Review, 2015. 92 C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94

[3] M. Sill, Journalists engage readers by learning who they are, 2016. URL http: [29] M. Barthel, E. Shearer, Jeffrey Gottfried, A. Mitchell, The evolving role of news //niemanreports.org/articles/journalists-engage-readers-by-learning-who- on Twitter and Facebook. URL http://www.journalism.org/2015/07/14/the- they-are/. evolving-role-of-news-on-Twitter-and-Facebook/. [4] Pew Research Center, News use across social media platforms 2016, 2016. [30] J. Lehmann, C. Castillo, Mounia Lalmas, R. Baeza-Yates, Story-focused reading URLhttp://www.journalism.org/2016/05/26/news-use-across-social-media in online news and its potential for user engagement, J. Assoc. Inf. Sci. -platforms-2016/. Technol. 68 (4) (2017) 869–883. [5] A. Hermida, Journalism practice, chap. Twittering the News: The Emergence [31] C. Orellana-Rodriguez, D. Greene, M.T. Keane, Spreading one’s tweets: How of Ambient Journalism, 2010.http://dx.doi.org/10.1080/1751278100364070 can journalists gain attention for their tweeted news? J. Web Sci. 4 (1) (2017) 3. 1–16. [6] Reuters Institute Digital News Report,Tech. Rep., 2015. URL http://www. [32] P.C. Álvarez, Journalism and social media: How spanish journalists are using digitalnewsreport.org/. Twitter/periodismo y social media: Cómo están usando Twitter los periodis- [7] Pew Research Center, In changing news landscape, even television is vul- tas españoles, Estudios sobre el mensaje periodístico 18 (1) (2012) 31–53. nerable, 2012. URL http://www.people-press.org/2012/09/27/in-changing- [33] A.M. Lee, H.I. Chyi, Motivational consumption model, J. Mass Commun. Q. news-landscape-even-television-is-vulnerable/. 91 (4) (2014) 706–724.http://dx.doi.org/10.1177/1077699014550088. [8] Associated Press, Social media guidelines for AP employees, 2012.http:// [34] C. Orellana-Rodriguez, D. Greene, M.T. Keane, Spreading the news: How can corp-live-bypass-ap.cphostaccess.com/assets/documents/social-media-guid journalists gain more engagement for their tweets? in: Proceedings of the elines_tcm28-9832.pdf, [Online]. 8th ACM Conference on Web Science, WebSci ’16, ACM, New York, NY, [9] B. Stelter, Journalism rules are bent in news coverage from Iran. URL http: USA, ISBN: 978-1-4503-4208-7, 2016, pp. 107–116. URL http://doi.acm.org/ //www.nytimes.com/2009/06/29/business/media/29coverage.html. 10.1145/2908131.2908154. [10] N. Diakopoulos, M.D. Choudhury, M. Naaman, Finding and assessing social [35] M. Brautović, I. Milanović-Litre, R. John, Journalism and Twitter: Between media information sources in the context of journalism, in: CHI Conference journalistic norms and new routines, MediAnali: International scientific jour- on Human Factors in Computing Systems, CHI ’12, ACM, 2012, pp. 2451–2460. nal of media, journalism, mass communication, public relations, culture and URL http://doi.acm.org/10.1145/2207676.2208409. society 7 (13) (2013) 19–36. URL http://hrcak.srce.hr/111224. [36] D.L. Lasorsa, Transparency and other journalistic norms on Twitter, J. Stud. [11] D. Bhattacharya, S. Ram, Understanding the competitive landscape of news 13 (3) (2012) 402–417.http://dx.doi.org/10.1080/1461670X.2012.657909. providers on social media, in: Proceedings of the 25th International Con- [37] W Dean, The journalist as a ‘committed observer’, American Press Institute, ference Companion on World Wide Web, International World Wide Web 2013. URL https://www.americanpressinstitute.org/journalism-essentials/ Conferences Steering Committee, 2016, pp. 719–724. what-is-journalism/journalist-committed-observer/. [12] H.G. de Zúñiga, T. Diehl, A. Ardèvol-Abreu, When citizens and journalists [38] D.L. Lasorsa, S.C. Lewis, A.E. Holton, Journalism studies, chap. Normalizing interact on Twitter, J. Stud. 19 (2) (2018) 227–246. Twitter, 2011.http://dx.doi.org/10.1080/1461670X.2011.571825. [13] J. Lee, The double-edged sword: The effects of journalists’ social media [39] M. Sullivan, How social media affects journalistic objectivity, 2013.https:/ activities on audience perceptions of journalists and their news products., /publiceditor.blogs.nytimes.com/2013/02/13/how-social-media-affects-jour J. Comput. Mediat. Commun. 20 (3) (2015) 312–329. URL http://dblp.uni- nalistic-objectivity/?_r=0. trier.de/db/journals/jcmc/jcmc20.html#Lee15. [40] Harvard Kennedy School, Shorenstein Center. NYT public editor sees social [14] S. Cohen, J.T. Hamilton, F. Turner, Computational Journalism, 2011. URLhttp media as ’Double-edged sword’ that is changing objective Journalism, 2013. s://cacm.acm.org/magazines/2011/10/131400-computational-journalism/fu URL https://shorensteincenter.org/speaker-series-with-margaret-sullivan/. lltext. [41] T. Rosenstiel, J. Sonderman, K. Loker, M. Ivancin, N. Kjarval, Twitter and [15] N. Diakopoulos, A functional roadmap for innovation in computational jour- the News: How People Use the Social Network to Learn About the World, nalism. URLhttp://www.nickdiakopoulos.com/2011/04/22/a-functional-roa American Press Institute, 2015. URLhttp://www.americanpressinstitute.or dmap-for-innovation-in-computational-journalism/. g/wp-content/uploads/2015/09/Twitter-and-News-How-people-use-Twitte [16] N. Kayser-Bril, A. Valeeva, I. Radchenko, Transformation of communica- r-to-get-news-American-Press-Institute.pdf. tion processes: Data journalism, CoRR (2016). URL http://arxiv.org/abs/1605. [42] E.-Ju Lee, S.Y. Shin, Are they talking to me? Cognitive and affective effects 01956. of interactivity in politicians’ Twitter communication, Cyberpsy., Behavior, [17] P.N. Howard, A. Duffy, D. Freelon, M.M. Hussain, W. Mari, M. Maziad, and Soc. Networking 15 (10) (2012) 515–520. URL http://dblp.uni-trier.de/ Opening closed regimes: What was the role of social media during the Arab db/journals/cbsn/cbsn15.html#LeeS12. spring? SSRN (2011).http://dx.doi.org/10.2139/ssrn.2595096. [43] A. Hermida, #JOURNALISM: Reconfiguring journalism research about Twit- [18] M. Thakur, D. Patel, S. Kumar, J. Barua, NewsInstaMiner: Enriching news ter, one tweet at a time, Digit. J. 1 (3) (2013) 295–313. (ISSN: 2167-0811, article using instagram, in: Srinath Srinivasa, Sameep Mehta (Eds.), Big Data 2167-082X). URLhttp://www.tandfonline.com/doi/abs/10.1080/21670811.2 Analytics: Third International Conference, BDA 2014, New Delhi, India, De- 013.808456. cember 20-23, 2014. Proceedings, Springer International Publishing, Cham, [44] R.G. Lawrence, L. Molyneux, M. Coddington, A. Holton, Tweeting conventions, ISBN: 978-3-319-13820-6, 2014, pp. 174–188.http://dx.doi.org/10.1007/97 J. Stud. 15 (6) (2014) 789–806.http://dx.doi.org/10.1080/1461670X.2013. 8-3-319-13820-6_16. 836378. [19] Facebook Media, Facebook journalism project, 2015. URL https://www. [45] L. Molyneux, What journalists retweet: Opinion, humor, and brand devel- Facebook.com/facebookmedia/. opment on Twitter, Journalism 16 (7) (2015) 920–935.http://dx.doi.org/10. [20] Google, Google news lab, 2015. URL https://newslab.withgoogle.com/. 1177/1464884914550135. [21] B. Ritholtz, Twitter is becoming the first and quickest source of invest- [46] E. Elejalde, L. Ferres, E. Herder, On the nature of real and perceived bias in the ment news, 2013. URL https://www.theguardian.com/technology/2013/apr/ mainstream media, PLoS One 13 (3) (2018) e0193765. 23/Twitter-first-source-investment-news. [47] D. Nikolov, D.F.M. Oliveira, A. Flammini, F. Menczer, Measuring online so- [22] E.C. Tandoc, T.P. Vos, The journalist is marketing the news, J. Pract. 10 (8) cial bubbles., CoRR abs/1502.07162 (2015). URL http://dblp.uni-trier.de/db/ (2016) 950–966.http://dx.doi.org/10.1080/17512786.2015.1087811. journals/corr/corr1502.html#NikolovOFM15. [23] E. Diaz-Aviles, A. Stewart, E. Velasco, K. Denecke, W. Nejdl, Epidemic intelli- [48] C. Grevet, L.G. Terveen, E. Gilbert, Managing political differences in social gence for the crowd, by the crowd, in: J.G. Breslin, N.B. Ellison, J.G. Shanahan, media., in: S.R. Fussell, W.G. Lutters, M.R. Morris, M. Reddy (Eds.), CSCW, Z. Tufekci (Eds.), Proceedings of the Sixth International Conference on We- ACM, ISBN: 978-1-4503-2540-0, 2014, pp. 1400–1408. URL http://dblp.uni- blogs and Social Media, The AAAI Press, 2012. URL http://www.aaai.org/ocs/ trier.de/db/conf/cscw/cscw2014.html#GrevetTG14. index.php/ICWSM/ICWSM12/paper/view/4594. [49] M. Del Vicario, S. Gaito, W. Quattrociocchi, M. Zignani, F. Zollo, News con- [24] M. Broersma, T. Graham, Social media as beat, J. Pract. 6 (3) (2012) 403–419. sumption during the Italian referendum: A cross-platform analysis on Face- http://dx.doi.org/10.1080/17512786.2012.663626. book and Twitter, in: Data Science and Advanced Analytics, DSAA, 2017 IEEE [25] M. Hasanain, M. Bagdouri, Tamer Elsayed, D.W. Oard, What questions do International Conference on, IEEE, 2017, pp. 648–657. journalists ask on Twitter?sep, in: Social Media in the Newsroom, Papers [50] J. Stray, Are we stuck in filter bubbles? Here are five potential paths from the 2016 ICWSM Workshop, Cologne, Germany, May 17, 2016, 2016. out, 2012. URL http://www.niemanlab.org/2012/07/are-we-stuck-in-filter- URL http://aaai.org/ocs/index.php/ICWSM/ICWSM16/paper/view/13221. bubbles-here-are-five-potential-paths-out/. [26] S. Owens, How the atlantic uses Twitter to drive traffic to Content, 2015. [51] S.R. Flaxman, S. Goel, J.M. Rao, Ideological segregation and the effects of social URL https://medium.com/thoughts-on-journalism/how-the-atlantic-uses- media on news consumption, in: SSRN Scholarly Paper ID 2363701, Social Twitter-to-drive-traffic-to-content-5593bc02a414, [Online]. Science Research Network, Rochester, NY, 2013. URL https://bfi.uchicago. [27] S. Hong, Online news on Twitter: Newspapers’ social media adoption and edu/sites/default/files/research/flaxman_goel_rao_onlinenews.pdf. their online readership, Inf. Econ. Policy (ISSN: 0167-6245) 24 (1) (2012) 69– [52] H. Allcott, M. Gentzkow, Social media and fake news in the 2016 election, J. 74. The Economics of Digital Media Markets. URL http://www.sciencedirect. Econ. Perspect. 31 (2) (2017) 211–236. URL http://www.aeaweb.org/articles? com/science/article/pii/S0167624512000054. id=10.1257/jep.31.2.211. [28] Reuters institute digital news report, Tech. Rep., 2016. URL http://www. [53] K. Shu, A. Sliva, S. Wang, J. Tang, H. Liu, Fake news detection on social media: digitalnewsreport.org/. A data mining perspective, ACM SIGKDD Explor. Newsl. 19 (1) (2017) 22–36. C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94 93

[54] S. Vosoughi, D. Roy, S. Aral, The spread of true and false news online, Science [76] J.J. Ratey, A.M. Galaburda, A user’s guide to the brain: Perception, atten- 359 (6380) (2018) 1146–1151. tion, and the four theaters of the brain, Vintage Series, Vintage Books, [55] P.N. Howard, S. Bradshaw, B. Kollanyi, C. Desigaud, G. Bolsover, Junk news ISBN: 9780375701078, 2002. URLhttps://books.google.ie/books?id=-_nYPm and bots during the french presidential election: What are french voters C5bp0C. sharing over Twitter? Data Memo 2017.3. Oxford, UK: Project on Compu- [77] D.K. Kilgo, S. Harlow, V. García-Perdomo, Ramón Salaverría, A new sensation? tational Propaganda, 2017. URL https://www.oii.ox.AC.uk/blog/junk-news- an international exploration of sensationalism and social media recommen- and-bots-during-the-french-presidential-election-round-i/. dations in online news publications, Journalism (2016). [56] L.-M. Neudert, B. Kollanyi, P.N. Howard, Junk news and bots during the [78] A. Piotrkowicz, V. Dimitrova, J. Otterbacher, K. Markert, Headlines mat- German federal presidency election: What Were German voters sharing ter: Using headlines to predict the popularity of news articles on Twit- over Twitter?, 2017. URL http://comprop.oii.ox.AC.uk/wp-content/uploads/ ter and Facebook., in: ICWSM, AAAI Press, ISBN: 978-1-57735-788-9, 2017, sites/89/2017/03/What-Were-German-Voters-Sharing-Over-Twitter-v6-1. pp. 656–659. URL http://dblp.uni-trier.de/db/conf/icwsm/icwsm2017.html# pdf, [Online; 2017]. PiotrkowiczDOM17. [57] C. Shao, G.L. Ciampaglia, O. Varol, A. Flammini, F. Menczer, The spread of [79] A. Piotrkowicz, V.G. Dimitrova, J. Otterbacher, K. Markert, The impact of news fake news by social bots, CoRR abs/1707.07592 (2017). URL http://dblp.uni- values and linguistic style on the popularity of headlines on Twitter and trier.de/db/journals/corr/corr1707.html#ShaoCVFM17. Facebook, in: Proceedings of the Second International Workshop on News [58] F. Rangel, P. Rosso, Irina Chugur, M. Potthast, M. Trenkmann, B. Stein, B. and Public Opinion, ICWSM NECO 2017, Association for the Advancement of Verhoeven, Walter Daelemans, Overview of the 2nd author profiling task Artificial Intelligence, 2017. at pan 2014, in: CLEF 2014 Evaluation Labs and Workshop Working Notes [80] P. English, Twitter’s diffusion in sports journalism: Role models, laggards and Papers, Sheffield, UK, 2014, 2014, pp. 1–30. followers of the social media innovation, New Media Soc. 18 (3) (2016) 484– [59] F. Rangel, P. Rosso, M. Potthast, B. Stein, Overview of the 5th author profiling 501.http://dx.doi.org/10.1177/1461444814544886. task at pan 2017: Gender and language variety identification in Twitter, [81] A. Bruns, Journalists and Twitter: How australian news organisations adapt Working Notes Papers of the CLEF (2017). to a new medium, Media Int. Aust. 144 (1) (2012) 97–107.http://dx.doi.org/ [60] C. Castillo, M. Mendoza, B. Poblete, Predicting information credibility in time- 10.1177/1329878X1214400114. sensitive social media, Tech. Rep., Internet Research 23 (5) (2013) 560–588. [82] F. Vis, Twitter as a reporting tool for breaking news, Digit. J. 1 (1) (2013) 27– 47.http://dx.doi.org/10.1080/21670811.2012.741316. [61] K. Starbird, J. Maddock, M. Orand, P. Achterman, R.M. Mason, Rumors, false [83] M. Osborne, M. Dredze, Facebook, Twitter and google plus for breaking news: flags, and digital vigilantes: Misinformation on Twitter after the 2013 boston Is there a winner? in: Proceedings of the Eighth International Conference on marathon bombing, in: IConference 2014 Proceedings, 2014. Weblogs and Social Media, ICWSM 2014, The AAAI Press, 2014. URL http: [62] A. Zubiaga, M. Liakata, R. Procter, G.W.S. Hoi, P. Tolmie, Analysing how people //www.aaai.org/ocs/index.php/ICWSM/ICWSM14/paper/view/8072. orient to and spread rumours in social media by looking at conversational [84] J.H. Parmelee, Political journalists and Twitter: Influences on norms and threads, PLoS One 11 (3) (2016) e0150989. practices, J. Media Pract. (ISSN: 1468-2753) 14 (4) (2013) 291–305. URLht [63] A. Zubiaga, A. Aker, K. Bontcheva, M. Liakata, R. Procter, Detection and tp://www.ingentaconnect.com/content/routledg/jmp/2013/00000014/0000 resolution of rumours in social media: A survey, ACM Comput. Surv. 51 (2) 0004/art00003. (2018) 32. [85] M. Dredze, P. Kambadur, Gary Kazantsev, G. Mann, M. Osborne, How Twit- [64] A. Arif, J.J. Robinson, S.A. Stanek, E.S. Fichet, P. Townsend, Z. Worku, K. ter is changing the nature of financial news discovery, in: Proceedings of Starbird, A closer look at the self-correcting crowd: Examining corrections the Second International Workshop on Data Science for Macro-Modeling, in online rumors, in: Proceedings of the 2017 ACM Conference on Computer in: DSMM’16, ACM, New York, NY, USA, ISBN: 978-1-4503-4407-4, 2016, Supported Cooperative Work and Social Computing, ACM, 2017, pp. 155–168. pp. 2:1–2:5. URLhttp://doi.acm.org/10.1145/2951894.2951903. [65] K. Starbird, D. Dailey, O. Mohamed, G. Lee, E.S. Spiro, Engage early, correct [86] N. Keane, C. Yee, L. Zhou, Using topic modeling and similarity thresholds more: How journalists participate in false rumors online during crisis events, to detect events, in: Proceedings of the The 3rd Workshop on EVENTS: in: Proceedings of the 2018 CHI Conference on Human Factors in Computing Definition, Detection, Coreference, and Representation, 2015, pp. 34–42. Systems, ACM, 2018, p. 105. [87] A. Olteanu, C. Castillo, N. Diakopoulos, K. Aberer, Comparing events coverage [66] C. Castillo, M. El-Haddad, Jürgen Pfeffer, M. Stempeck, Characterizing the life in online news and social media: The case of climate change, in: Proceedings cycle of online news stories using social media reactions, in: Proceedings of of the Ninth International Conference on Web and Social Media, ICWSM 2015, the 17th ACM Conference on Computer Supported Cooperative Work & Social AAAI Press, 2015, pp. 288–297. URLhttp://www.aaai.org/ocs/index.php/ICW Computing, CSCW ’14, ACM, New York, NY, USA, ISBN: 978-1-4503-2540-0, SM/ICWSM15/paper/view/10583. 2014, pp. 211–223. URL http://doi.acm.org/10.1145/2531602.2531623. [88] D. Saez-Trumper, C. Castillo, M. Lalmas, Social media news communities: [67] J. Yang, J. Leskovec, Patterns of temporal variation in online media, in: Gatekeeping, coverage, and statement bias, in: Proceedings of the 22nd ACM Proceedings of the fourth ACM international conference on Web search and iNternational Conference on Conference on Information & Knowledge Man- data mining, ACM, 2011, pp. 177–186. URLhttp://scholar.google.com/schol agement, in: CIKM ’13, ACM, New York, NY, USA, ISBN: 978-1-4503-2263-8, ar.bib?q=info:lcsi9DT12o0J:scholar.google.com/&output=citation&hl=en&as 2013, pp. 1679–1684. URLhttp://doi.acm.org/10.1145/2505515.2505623. _sdt=0,5&as_vis=1&ct=citation&cd=0. [89] D. Dailey, K. Starbird, Journalists as crowdsourcerers: Responding to crisis by [68] D.M. Romero, W. Galuba, S. Asur, B.A. Huberman, Influence and passivity in reporting with a crowd, Comput. Support. Coop. Work (CSCW) (ISSN: 1573- social media, in: Proceedings of the 20th International Conference Compan- 7551) 23 (4) (2014) 445–481.http://dx.doi.org/10.1007/s10606-014-9208-z. ion on World Wide Web, WWW ’11, ACM, New York, NY, USA, ISBN: 978-1- [90] L. Canter, The interactive spectrum, Convergence 19 (4) (2013) 472–495. 4503-0637-9, 2011, pp. 113–114. URL http://doi.acm.org/10.1145/1963192. http://dx.doi.org/10.1177/1354856513493698. 1963250. [91] L. McCay-Peet, M. Lalmas, V. Navalpakkam, On saliency, affect and focused [69] L. Zhu, K. Lerman, Attention inequality in social media, CoRR abs/1601.07200 attention, in: Proceedings of the SIGCHI Conference on Human Factors in (2016). URL http://arxiv.org/abs/1601.07200. Computing Systems, in: CHI ’12, ACM, New York, NY, USA, ISBN: 978-1-4503- [70] M. Borger, A. van Hoof, José Sanders, Expecting reciprocity: Towards a model 1015-4, 2012, pp. 541–550. URLhttp://doi.acm.org/10.1145/2207676.22077 of the participants’ perspective on participatory journalism, New Media Soc. 51. 18 (5) (2016) 708–725.http://dx.doi.org/10.1177/1461444814545842. [92] M. Broersma, T. Graham, Twitter as a news source, J. Pract. 7 (4) (2013) 446– [71] Y. Hu, S. Farnham, Kartik Talamadupula, Predicting user engagement on 464.http://dx.doi.org/10.1080/17512786.2013.802481. Twitter with real-world events, in: Proceedings of the Ninth International [93] M. Bagdouri, Journalists and Twitter: A multidimensional quantitative de- Conference on Web and Social Media, ICWSM 2015, University of Oxford, scription of usage patterns, in: ICWSM, AAAI Press, ISBN: 978-1-57735-758- Oxford, UK, May 26–29, 2015, 2015, pp. 168–178. URL http://www.aaai.org/ 2, 2016, pp. 22–31. URLhttp://dblp.uni-trier.de/db/conf/icwsm/icwsm2016. ocs/index.php/ICWSM/ICWSM15/paper/view/10670. html#Bagdouri16. [72] R. Bandari, S. Asur, B.A. Huberman, The pulse of news in social media: Fore- [94] B. Shi, G. Ifrim, N.J. Hurley, Learning-to-rank for real-time high-precision casting popularity, CoRR abs/1202.0332 (2012). URL http://dblp.uni-trier.de/ hashtag recommendation for streaming news, in: J. Bourdeau, J. Hendler, R. db/journals/corr/corr1202.html#abs-1202-0332. Nkambou, I. Horrocks, B.Y. Zhao (Eds.), WWW, ACM, ISBN: 978-1-4503-4143- [73] J.N. Matias, H. Wallach, Working paper: Modeling gender discrimination by 1, 2016, pp. 1191–1202. URLhttp://dblp.uni-trier.de/db/conf/www/www20 online news audiences, in: Computation + Journalism Symposium, Columbia 16.html#ShiIH16. University, 2015. [95] J. Lehmann, C. Castillo, M. Lalmas, E. Zuckerman, Transient news crowds in [74] J.N. Matias, S. Szalavitz, E. Zuckerman, FollowBias: Supporting behavior social media, in: ICWSM, 2013. change toward gender equality by networked gatekeepers on social media, [96] M. Gabielkov, A. Ramachandran, A. Chaintreau, A. Legout, Social clicks: What in: CSCW, 2017. pp. 1082–1095. and who gets read on Twitter? ACM SIGMETRICS Perform. Eval. Rev. 44 (1) [75] N. Diakopoulos, A. Zubiaga, Newsworthiness and network gatekeeping on (2016) 179–192. Twitter: The role of social deviance, in: ICWSM, The AAAI Press, ISBN: 978-1- [97] C. Danescu-Niculescu-Mizil, J. Cheng, J.M. Kleinberg, L. Lee, You had me at 57735-659-2, 2014. URLhttp://dblp.uni-trier.de/db/conf/icwsm/icwsm2014 hello: How phrasing affects memorability, in: ACL (1), The Association for .html#DiakopoulosZ14. Computer Linguistics, ISBN: 978-1-937284-24-4, 2012, pp. 892–901. URLh 94 C. Orellana-Rodriguez, M.T. Keane / Computer Science Review 29 (2018) 74–94

ttp://dblp.uni-trier.de/db/conf/acl/acl2012-1.html#Danescu-Niculescu-Mizi [111] T. Szymanski, C. Orellana-Rodriguez, M.T. Keane, Helping news editors write lCKL12. better headlines: A recommender to improve the keyword contents & share- [98] J. Leskovec, L. Backstrom, Jon Kleinberg, Meme-tracking and the dynamics ability of news headlines, IJCAI-17 Natural Language & Journalism. also arXiv of the news cycle, in: Proceedings of the 15th ACM SIGKDD International preprint arXiv:1705.09656. Conference on Knowledge Discovery and Data Mining, KDD ’09, ACM, New [112] K. Williams, Read All About It!: A History of the British Newspaper, Routledge, York, NY, USA, ISBN: 978-1-60558-495-9, 2009, pp. 497–506. URL http://doi. New York, NY, USA, ISBN: 0203-59689-7, 2010. acm.org/10.1145/1557019.1557077. [113] V. García-Perdomo, R. Salaverría, D.K. Kilgo, S. Harlow, To share or not to [99] C. Suen, S. Huang, C. Eksombatchai, R. Sosic, J. Leskovec, NIFTY: A system share: The influence of news values and topics on popular social media for large scale information flow tracking and clustering, in: Proceedings of content in the united states, brazil, and argentina, J. Stud. (2017) 1–22. the 22nd International Conference on World Wide Web, WWW ’13, ACM, [114] D. Trilling, P. Tolochko, B. Burscher, From newsworthiness to shareworthi- New York, NY, USA, ISBN: 978-1-4503-2035-1, 2013, pp. 1237–1248. URL ness: How to predict news sharing based on article characteristics, J. Mass http://doi.acm.org/10.1145/2488388.2488496. Commun. Q. 94 (1) (2017) 38–60. [100] M. Hu, S. Liu, F. Wei, Y. Wu, J. Stasko, K.-L. Ma, Breaking news on Twitter, [115] J.N. Blom, K.R. Hansen, Click bait: Forward-reference as lure in online news in: Proceedings of the SIGCHI Conference on Human Factors in Computing headlines, J. Pragmat. 76 (2015) 87–100. Systems, in: CHI ’12, ACM, New York, NY, USA, ISBN: 978-1-4503-1015-4, [116] S. Petrovic, M. Osborne, R. McCreadie, C. Macdonald, I. Ounis, L. Shrimpton, 2012, pp. 2751–2754. URL http://doi.acm.org/10.1145/2207676.2208672. Can Twitter replace newswire for breaking news? in: E Kiciman, N.B. Ellison, [101] M.M. Malik, J. Pfeffer, A macroscopic analysis of news content in Twitter, Digit. J. 4 (8) (2016) 955–979. B. Hogan, P. Resnick, I. Soboroff (Eds.), ICWSM, The AAAI Press, ISBN: 978-1- [102] D. Bhattacharya, S. Ram, Sharing news articles using 140 characters: A diffu- 57735-610-3, 2013. URLhttp://dblp.uni-trier.de/db/conf/icwsm/icwsm2013 sion analysis on Twitter, in: Proceedings of the 2012 IEEE/ACM International .html#PetrovicOMMOS13. Conference on Advances in Social Networks Analysis and Mining, ASONAM [117] B. Kalsnes, A.O. Larsson, Understanding news sharing across social media: 2012, ISBN: 9780769547992, 2012, pp. 966–971,. Detailing distribution on Facebook and Twitter, J. Stud. (2017) 1–20. [103] M. De Choudhury, N. Diakopoulos, M. Naaman, Unfolding the event landscape [118] Y. Keneshloo, S. Wang, E.-H. Han, N. Ramakrishnan, Predicting the popularity on Twitter: Classification and exploration of user categories, in: Proceedings of news articles, in: Proceedings of the 2016 SIAM International Conference of the ACM 2012 Conference on Computer Supported Cooperative Work, on Data Mining, SIAM, 2016, pp. 441–449. in: CSCW ’12, ACM, New York, NY, USA, ISBN: 978-1-4503-1086-4, 2012, [119] C. Liu, W. Wang, Y. Zhang, Y. Dong, F. He, C. Wu, Predicting the popularity pp. 241–244. URL http://doi.acm.org/10.1145/2145204.2145242. of online news based on multivariate analysis, in: Computer and Information [104] J. Wallace, Modelling contemporary gatekeeping: The rise of individuals, Technology, CIT, 2017 IEEE International Conference on, IEEE, 2017, pp. 9–15. algorithms and platforms in digital news dissemination, Digit. J. 6 (3) (2018) [120] L. Connell, M.T. Keane, What Plausibly affects plausibility? Concept coherence 274–293. and distributional word coherence as factors influencing plausibility judg- [105] D.L. Lasorsa, S.C. Lewis, A.E. Holton, Normalizing Twitter, J. Stud. 13 (1) (2012) ments, Mem. Cogn. (2004). 19–36.http://dx.doi.org/10.1080/1461670X.2011.571825. [121] A. Hermida, S.C. Lewis, R. Zamith, Sourcing the arab spring: A case study of [106] C.-Kang Hsieh, L. Yang, H. Wei, M. Naaman, D. Estrin, Immersive recom- andy carvin’s sources on Twitter during the tunisian and egyptian revolu- mendation: News and event recommendations using personal digital traces, tions, J. Comput. Mediat. Commun. (ISSN: 1083-6101) 19 (3) (2014) 479–499. in: Proceedings of the 25th International Conference on World Wide Web, http://dx.doi.org/10.1111/jcc4.12074. in: WWW ’16, International World Wide Web Conferences Steering Commit- [122] A. Olteanu, S. Vieweg, C. Castillo, What to expect when the unexpected tee, Republic and Canton of Geneva, Switzerland, ISBN: 978-1-4503-4143-1, happens: Social media communications across crises, in: Proceedings of the 2016, pp. 51–62.http://dx.doi.org/10.1145/2872427.2883006. 18th ACM Conference on Computer Supported Cooperative Work & Social [107] A. Monroy-Hernández, d. boyd, E. Kiciman, M. De Choudhury, S. Counts, The Computing, in: CSCW ’15, ACM, New York, NY, USA, ISBN: 978-1-4503-2922- new war correspondents: The rise of civic media curation in urban warfare, 4, 2015, pp. 994–1009. URL http://doi.acm.org/10.1145/2675133.2675242. in: Proceedings of the 2013 Conference on Computer Supported Cooperative [123] J. Reis, H. Kwak, J. An, J. Messias, F. Benevenuto, Demographics of news Work, in: CSCW ’13, ACM, New York, NY, USA, ISBN: 978-1-4503-1331-5, 2013, pp. 1443–1452. URL http://doi.acm.org/10.1145/2441776.2441938. sharing in the us Twittersphere, in: Proceedings of the 28th ACM Conference [108] N.O. Hodas, K. Lerman, How visibility and divided attention constrain social on Hypertext and Social Media, ACM, 2017, pp. 195–204. contagion, in: Proceedings of the 2012 ASE/IEEE International Conference on [124] Reuters, Reporting from the internet and using social media, 2012. URLhttp Social Computing and 2012 ASE/IEEE International Conference on Privacy, ://handbook.reuters.com/?title=Reporting_From_the_Internet_And_Using_S Security, Risk and Trust, in: SOCIALCOM-PASSAT ’12, IEEE Computer Society, ocial_Media, [Online]. Washington, DC, USA, ISBN: 978-0-7695-4848-7, 2012, pp. 249–257.http:// [125] AFP, Guidelines for using social media, 2013. URL https://www.afp.com/ dx.doi.org/10.1109/SocialCom-PASSAT.2012.129. communication/new-social-media-guidelines.pdf, [Online]. [109] S. Chesney, M. Liakata, M. Poesio, M. Purver, Incongruent headlines: Yet [126] BBC, News: Social media guidance, 2011. URL http://news.bbc.co.uk/2/ another way to mislead your readers, in: Proceedings of the 2017 EMNLP shared/bsp/hi/pdfs/14_07_11_news_social_media_guidance.pdf, Workshop: Natural Language Processing meets Journalism, 2017. pp. 56–61. [Online; 2011]. [110] J.H. Kim, A. Mantrach, A. Jaimes, A. Oh, How to compete online for news [127] Washington Post, Digital publishing guidelines: Social media, 2011. URL http: audience: Modeling words that attract clicks, in: Proceedings of the 22nd //www.washingtonpost.com/wp-srv/guidelines/social-media.html, [Online; ACM SIGKDD International Conference on Knowledge Discovery and Data 2011]. Mining, ACM, 2016, pp. 1645–1654.