=Paper=
{{Paper
|id=Vol-2482/paper50
|storemode=property
|title=The Challenge of Personal Attribute Preferences in Recommending
Diverse, Reliable News Sources
|pdfUrl=https://ceur-ws.org/Vol-2482/paper50.pdf
|volume=Vol-2482
|authors=Brooke Auxier,Jennifer Golbeck
|dblpUrl=https://dblp.org/rec/conf/cikm/AuxierG18
}}
==The Challenge of Personal Attribute Preferences in Recommending
Diverse, Reliable News Sources==
The challenge of personal attribute preferences in recommending diverse, reliable news sources Brooke Auxier Jennifer Golbeck Philip Merrill College of Journalism School of Information Studies University of Maryland University of Maryland College Park, Maryland, USA College Park, Maryland, USA bauxier@umd.edu jgolbeck@umd.edu ABSTRACT The affordances of social media and the internet allow users to Online news audiences have a vast amount of content constantly encounter and engage with diverse and novel news content. available at their fingertips. In order to manage this amount of However, user preference and bias may limit the content consumed in these spaces. Using signals of reliability, which have been information and avoid information overload, audiences make studied as they relate to content and information sharers in social decisions about which content and sources to trust, read, follow and media environments, a recommender system could be built to share. However, research suggests that audiences may do this in a suggest news content to users. However, as stated, assessments of way that limits their access to a diverse range of sources and trustworthiness and reliability come with some user bias and an viewpoints [3], [4], as audiences are likely to actively engage with algorithm that uses these preferences could be extremely limiting. content that confirms their existing beliefs [5], [6], [7]. We propose the following solutions: (1) using trusted social connections to surface content; (2) using bots to broaden a user’s This is exacerbated when personalization algorithms are at work. information ecosystem. Algorithms are designed to recommend content that users will like and engage with. If users are personally driven toward a small set of perspectives, “accurate” algorithms may drive them deeper into that space. Beyond just the issue of viewpoint, user preferences - and biases - with respect to the people sharing content also matters. On many social media platforms, algorithms highlight news and other content posted by user accounts, not just organizational accounts. Thus, we must consider how users will respond to the KEYWORDS people sharing the content as well. news recommendations, Reliability, News sources, Social media ACM Reference format: 2 Related work Research suggests then when social media users encounter information sharers on the platform who are unfamiliar to them, they use multiple heuristic cues to gauge the trustworthiness and credibility of the source, as well as the message. Earlier work by [8] found that tweet content with non-standard grammar and 1 Introduction punctuation were viewed as having low credibility, whereas tweets that included links to high quality sources were seen as more Billions of people across the globe use social media sites. In 2016, credible. 2.28 billion people were on social media and that number is expected to increase to 3.02 billion by 2021 [1]. Though there are Audiences also judge the information sharers themselves. Users many reasons for using social media, news consumption is common that had a default image, cartoon or avatar as their profile photos activity on the sites. In the U.S, 67 percent of adults report getting were rated as having low credibility, whereas users with high at least some of their news on social media sites [2]. Though these follower counts who had a Twitter bio were seen as more credible figures are already substantial, the numbers undoubtedly increase [8]. The same study found that the sharer’s @handle or screen name when news usage metrics from mobile apps, messaging platforms also influenced perceived credibility. Topically relevant screen like WhatsApp, WeChat and GroupMe, and news websites are also names were seen as more credible than ‘internet style names.’ considered. Similar, more recent research we conducted suggests that users find certain features of social media profiles more reliable than others [9]. For example, the study (N=261) found that when exposed to Copyright © CIKM 2018 for the individual papers by the papers' authors. Copyright © CIKM 2018 for the volume as a collection by its editors. This volume and its papers are published under the Creative Commons License Attribution 4.0 International (CC BY 4.0). neutral news-oriented Twitter-like content from unknown by incorporating profile attribute preferences. This would need information sharers, users with Western names, gender-neutral experimental validation. However, any improved performance may names and female avatars were perceived as most reliable. come at the cost of bias in the type of sources and in the diversity Respondents were also most likely to share content from of news a person sees. Countering user bias through information sharers with Western names, gender-neutral names and recommendation and leveraging the perceived reliability of a who had non-human avatars (e.g. logos, non-human objects). profile to bring more diverse news to users are both interesting research challenges going forward. They highlight the complexity of social interaction and the role it plays in recommendation, but also the opportunities that arise from a deeper understanding of how 3 Proposed solutions, recommendations users assess people they encounter online. These signals of reliability could be integrated into a recommender system to suggest content to users. If we know they are likely to engage more with posts from people with certain profile features, REFERENCES an algorithm could be designed to bring more content from those [1] eMarketer, Number of social media users worldwide from 2010 to 2021 (in billions), https://www.statista.com/statistics/278414/number-of-worldwide- types of accounts. social-network-users/ [2] Elisa Shearer and Jeffrey Gottfried. News Use Across Social Media Platforms, However, one can see that there is some bias in these evaluations http://www.journalism.org/2017/09/07/news-use-across-social-media- platforms-2017/ of trustworthiness and credibility. There are plenty of reliable, [3] Seth Flaxman, Sharad Goel and Justin M. Rao. (2016). Filter Bubbles, Echo trustworthy and talented journalists and who do not have Western Chambers, and Online News Consumption. Public Opinion Quarterly, 80(S1), sounding names and who use cartoon avatars in their profile photos 298-320. - though subjects in our study gave these types of users among the [5] Thomas J. Johnson, Shannon Bichard and Weiwu Zhang. (2009). Communication Communities or “CyberGhettos?”: A Path Analysis Model lowest credibility ratings. There are also many nefarious sources Examining Factors that Explain Selective Exposure to Blogs. Journal of with profiles that match the traditional standards of trustworthiness. Computer-Mediated Communication, 15(1), 60-82. [6] Solomon Messing and Sean J. Westwood. (2014). Selective Exposure in the Age of Social Media: Endorsements Trump Partisan Source Affiliation When This raises the question of how to build a recommender system that Selecting News Online. Communication Research 41(8), 1042-1063. (1) incorporates information about profile attribute credibility [4] Dimitar Nikolov, Diego F. M. Oliveira, Alessandro Flammini and Filippo assessments, (2) suggests useful, reliable content that users will Menczer. (2015). Measuring Online Social Bubbles. PeerJ Computer Science, perceive as such and (3) does not reinforce unfair biases. 1(38), 12 pages. [7] Stephan Winter, Miriam J. Metzger and Andrew J. Flanagin. (2016). Selective Use of News Cues: A Multiple-Motive Perspective on Information Selection in First, recent results from our work [9] and others [8] indicate a Social Media Environments. Journal of Communication, 66(4), 669-693. preference for news shared from accounts with certain personal [8] Meredith Ringel Morris, Scott Counts, Asta Roseway, Aaron Hoff and Julia Schwarz. (2012). Tweeting is Believing? Understanding Microblog Credibility attributes. These preferences may vary among users. Leveraging Perceptions. In Proceedings of CSCW 2012. ACM, Seattle, Washington, 1-10. profile attribute preferences has the potential to improve perceived [9] Brooke Auxier and Jennifer Golbeck. (2018). Factors influencing perceived recommendation quality, but this requires empirical analysis in reliability of information-sharers in social media spaces. University of Maryland, addition to experimental analysis. College Park, MD. If these preferences do improve recommendation accuracy, they will negatively impact the diversity of sources that are recommended and potentially the diversity of news itself. To counter this, we see research paths in countering these preferences and in playing to them. To counter such preferences, we may look to strong, trusted existing social connections and highlight engagements (e.g. likes, shares) from a user’s friend who shares content from a less-preferred profile type. This may weaken the bias toward certain sources. When those biases may be difficult to overcome, we are interested in the impact of bots. Recommender systems or news organizations may consider creating automated accounts with profiles that play to a user’s reliability bias, and have those accounts share news that will broaden the user’s information ecosystem. 4 Conclusion Recommender systems help people find reliable news in an ecosystem full of sources and perspectives. Previous results that show users have preferences for a range of personal attributes of news-sharing accounts, including avatar type, name, screen name, and gender. This raises several interesting research questions. Recommender systems may be able to improve their performance