User Interface Personalization in News Apps Marios Constantinides John Dowell Department of Computer Science Department of Computer Science University College London University College London NW1 2FD, UK NW1 2FD, UK m.constantinides@cs.ucl.ac.uk j.dowell@cs.ucl.ac.uk ABSTRACT consumption. News services are now able to help people find News is increasingly being accessed on smartphones and tablets, news that is relevant to them, to recommend the right news to the establishing mobile news reading as one of the most popular right users when they want it, and to help users abreast of news by activities on mobile devices. News reading is also a very aggregation over multiple sources. Billsus and Pazzani [8] have individual activity with marked differences in the way people read early identified several ways of providing such personalized news and access the news, however, news apps have limited service. The most widely used is through news content personalization. In this paper, we approach news personalization recommendation by pushing filtered articles predicted to match as a two-dimensional problem. We discuss news personalization user’s interests. Alternatives include adaptive news browsing by in terms of ‘what’ content is delivered to the user and ‘how’ that changing the order of articles’ categories, contextual news access content is consumed. We present our approach towards user by offering users access to additional information related to the interface personalization in news apps and we conclude that news news they are reading and news aggregation by automatically content recommendation and user interface personalization should aggregating news stories from multiple sources. A significant co-exist in news apps. amount of work has been conducted in the area of news recommendation. News recommenders have been widely CCS Concepts deployed both in the desktop and mobile environment and •Human-centered computing → HCI design and evaluation techniques from the broader area of recommendation systems methods; User models; •Human-centered computing → User such as content-based or collaborative-filtering have been used [6, interface design; 20]. However, progress in personalizing the choice of news content Keywords has not been matched by progress in personalizing the way that Mobile News Reading; Personalization; User Interfaces content is accessed and read. Mobile news apps are a particular case in point of the need for adaptive personalization in user 1. INTRODUCTION interfaces. News apps frequently adapt to users’ individual news News reading is being changed rapidly due to advances in digital interests through recommendation services and many allow user methods of consumption. App markets are now bursting with customization of news feeds [8]. But, their user interfaces do not prominent apps for accessing news spanning the globe, delivering adapt to how individual users characteristically select and read the completely tailored news recommendations either based on users’ news, as opposed to what news they are interested in reading. interests or location and aggregating news from multiple sources. The use of mobile devices to consume news media is rapidly In this paper we present news personalization as a two- growing and is seen as the future for the news industry as recent dimensional problem, defined by the ‘what’ and the ‘how’ numbers show [2]. aspects. The ‘what’ is characterized by the content itself, i.e. the content that is delivered to match user’s interests and preferences Reading the news is now amongst the most popular activities through news recommendation engines. The ‘how’ is defined by people perform on a daily basis with their handheld personal the user interface in terms of the presentation of news stories and devices [3]. Additionally, it is a very individual activity in which the interaction that is used to access and interact with news people follow idiosyncratic patterns for accessing and reading the stories. Finding articles to match user interests is different to how news. For example, people are likely to have distinctive ways in users selecting and reading them. We highlight in this paper the browsing news headlines or choosing how much of an article to need of user interface personalization and we conclude that new read and how they do it [18]. This implied diversity in users’ news content recommendation and user interface personalization should reading behaviour is reflected by the wide choice of available exist side-by-side in a news personalized service. news apps, by the personalization features of some apps, and by the development of recommendation engines to filter news. The 2. RELATED RESEARCH proliferation of smartphones and tablets along with the indispensable nature that play in peoples’ everyday life indicate 2.1 Adaptivity in News Apps Much of the news personalization literature is encapsulated in the their significant role as platforms for cross-media news areas of adaptable and adaptive systems [25, 27]. Adaptable systems allow users to manually tailor the interface or the system to fit their particular needs and demands to complete specific tasks. Although these systems may require additional effort and time by the end user to learn how to customize and use them [22], they are widely deployed. The majority of news apps allow users to manually create a personalized experience mainly through configurations and customizations, for example, by explicitly Copyright is held by the owner/author(s). selecting topics of interest or specifying system’s parameters on INRA 2016 Workshop, Halifax, Canada how they want the visual presentation of a story. We reviewed news apps from Apple’s and Google’s marketplaces Although, the larger portion of news apps adopt the adaptable way that provide personalization mostly in an adaptable manner; first, of providing personalization, the alternative is the use of adaptive to get a better understanding of how they achieve personalization, principles. Adaptive systems attempt to overcome some of the and, second, to identify possible gaps that would inform the limitations of adaptable systems by dismissing the user from design of more personal user interfaces. Our review is based on manually personalizing the system or the interface. Adaptive online tech blogs including the DigitalTrends, the Wired, the systems mainly leverage prior knowledge about the user or exploit BusinessInsider and the SimplyZesty. user’s content to infer their goals and needs to automatically alter Leading news organizations such as BBC and CNN have already the system’s behaviour. However, despite these potential benefits realized the need of personalization in their own news apps. For of adaptive systems, news apps tend to adopt the adaptable example, BBC news app provides a more personal news reading principles by letting users manually customize the content or the experience through customizations of the interface and other interface themselves. Today, however, users are more system’s parameters related to the content. Example features of sophisticated interacting with user interfaces than a decade ago. the revamped app include the most read stories, an option to add a Adaptive principles could possibly work better for today’s list of news stories user follows, presentation settings of smartphone user interfaces. Smartphones have much more displaying and categorising the stories such as a compact layout or advanced capabilities such as 3G connectivity, high-resolution carousels and many others. screens, sophisticated interactions with the user interface (swipe, flick, scroll) and others. The priority in pre-smartphones era was A new breed of news apps, the news aggregators, has drawn the to deliver content but we believe users now expect more than that. attention lately. This type of service mainly focuses on the aggregation and the classification of news content from multiple 2.2 News Reading Behaviour sources. With more news sources emerging and a tremendous An investigation of news reading behaviour is the necessary amount of stories spanning all over the world, news aggregators starting point to creating personalized news services. The news help users to identify news topics of interest easily and access domain is characterized by a number of particular challenges such specific topics from different news providers. Flipboard, for as finding the right stories for the right people at the right time or example, uses the metaphor of a ‘personal magazine’ by making presenting news stories in a way that match the particular needs of the entire reading process stylish. It gives the sense of flipping a the reader. Many studies and reports focused in analysing human magazine page while navigating through news. Users curate and behaviour to identify patterns of news consumption, users’ share their own mini-magazines with the app, drawing in stories preferences, and others. We believe news personalization needs a on their preferred topics. Zite is an intelligent magazine-like news clear understanding of how people consume news, especially on app that recommends stories based on user’s interests and reading mobile devices. Addressing questions such as how news readers habits. The app learns user’s preferences through a thumbs-up or select stories to read, what reading patterns follow while reading, thumbs down button on each story. Inside.com – Breaking News is an essential prerequisite to effectively create personalized allows users to select news topics to follow and then provides systems. 300-character summaries of relevant stories along with links to Liu [21] identified a new reading behaviour (named as screen- the original sources. Newsbeat is another aggregator but one that based reading), mainly emerged with the advent of digital news creates ‘personalised radio news bulletins’. Users select their services, which is characterised by more time spent on browsing preferred text news sources from which stories are pulled each and scanning, keyword spotting, one-time reading, non-linear day, summaries created, then news podcasts created using text-to- reading, and less time on in-depth reading. Recently mobile news voice technology. Feedly aggregates news items, longer articles, readers, particularly younger audiences, are exposed in a more blog posts, and quick videos into a single spot in an elegant way. snackable format of consuming digital news. A recent study from Further, instead of providing a massive list of articles it breaks the BBC R&D [1] identified the need of delivering news stories in a content feed up into manageable chunks. News360 differentiates quick, snappy format while also providing readers the opportunity from other aggregators by incorporating two swipeable screens, in for in-depth reading when needed. Other data from Reuters [5] which the top part shows the most popular stories while the and Pew Research Centre [3] revealed interesting insights about bottom displays the current article you are reading. news consumption on mobile devices. One to every five mobile Social networking platforms such as Facebook and Twitter are news readers tend to read in-depth news articles a study from Pew becoming distribution channels for news stories. Recently, it Research Centre identified. Likewise, a Reuters Institute report appears to be a huge interest in such services with more people showed that more than a third of online news users across all getting their news stories and updates from social networks as countries use two or more digital devices to access the news and a numbers indicate [4]. Therefore, this kind of service could used to fifth uses a mobile phone as their primary access point. develop apps that pull or leverage knowledge from users’ social Summarizing these findings, it is evident that news reading, networks activities. For example, Pulse, developed by LinkedIn, especially on mobile devices, is a very individual activity. There delivers personalized news from a user’s professional network. are characteristic differences amongst people in the way they Further, Phelan et al. [24] proposed a system, which recommends consume news, especially in younger audiences. We believe there and ranks news articles by analysing real-time Twitter data. is potential and this diversity in news reading behaviour should be Apart from news apps, web portals such as Google News and taken into account when we design and deploy news personalized Reddit aggregate news sources and/or recommend news articles to services. assist desktop end-users to find and read news more efficiently. Although these studies have presented interesting insights about These systems gather information about their users either mobile news reading behaviour, to the best of our knowledge, no explicitly, i.e. users give rates to articles, in the case of Reddit, or study has attempted to categorise news readers based on particular implicitly by observing user behaviour, i.e. track user’s activity, news reading characteristics. Categorizing users as the basis of reading preferences, etc., in the case of Google News [20]. adaptation has been demonstrated in other domains such as the AVANTI project for people with disabilities [26], the work conducted by Carberry et al. [10] for natural dialogues, the interface and interaction in response to user’s stereotypical visiting styles/categories in a museum scenario [29], and others. behaviour, i.e. offering the user to choose a variant interface that This idea dates well back when Elaine Rich (1979) introduced the would suit better their idiosyncratic news reading behaviour. We idea of using stereotypes to model users and concluded that the propose user interface personalization through variant interfaces use of stereotypes in conjunction with the ability to record explicit for each news reader category as opposed to feature-level user statements about himself may provide a powerful mechanism adaptation. for creating computer systems that can react differently to Much of the work in the literature has been focused on feature- different users. level adaptation [13, 23], i.e. modification of specific user In our previous research [12] we surveyed people who read the interface features in response to interactions with specific news on their mobile phones with the aim of identifying functions. An alternative approach would be to determine a stereotypical patterns of news consumption. We identified three category for a user and provide a matching user interface variant. distinct kinds of news reader (we label Trackers, Reviewers and In our previous work [12], we developed an early prototype of the Dippers) distinguished by five ‘reader factors’. The five variant interfaces matching the three news reader categories. The characteristic factors included how often they read the news, how interface consists of two levels: the navigation and the reading. much time they spend on reading, how they browse to find stories, The former consists of features related to how users interact with what strategy they use to read and where they read the news. For the news app interface, i.e. how they browse to find stories, what example, Trackers are news readers who read the news many strategies they use to select stories and how frequent they access times a day and in short bursts, they tend to skim read articles the news. The latter includes features related to how users perform rather read them word for word, and they read the news in the reading task, specifically, how stories are presented to better different locations. We believe this news reader categorization match user’s reading style. could be useful for any news personalization service either focused on recommending articles or personalizing the user 3. DISCUSSION interface. News personalization is a very specific domain that has unaddressed challenges in providing personalized service. Much 2.3 Learning the User of the work in this domain has been focused on recommending Given that we believe there are three kinds of news readers, how news content, whereas personalizing the user interface has would we identify an individual reader as belonging to one of received less attention. three kinds is the next step in creating personalized user interfaces for news apps. We believe that personalization of news access needs to broaden its scope to also include not only what content users access but Undoubtedly, any kind of personalization relies on the system also how they access and interact with that content. Mobile news having an effective user model wherein “unobservable access makes the need for personalized interaction with news apps information about a user is inferred from observable information much more apparent. This is not simply because the displays and from that user” [14]. Such unobservable information may include input are still relatively limiting, but because of the different ways the user’s interests, knowledge, background and skills, goals and in which different users read the news with mobiles and the tasks [9]. Observable information is collected either explicitly different settings in which those individual users read the news. through direct user intervention and/or implicitly through Adaptive news navigation through reordering menus of headlines monitoring user activity [17]; the latter is often preferred by users. is clearly one area of personalization of how a user accesses the User models vary in the methods used for inferring unobservable news. However, we propose that adaptation could be far more information from the observable. Rudimentary methods used for extensive and multi-dimensional. For example, users are likely to re-ordering menus include recency and frequency scores of have idiosyncratic patterns for browsing news headlines to which command usage [16]; more sophisticated methods of user a headline display could respond dynamically and adaptively. modelling involve supervised learning techniques for inferring Interaction with mobile news is likely to vary significantly preferences from interaction data [8, 19]. Some user models infer between users in the way they browse news headlines and the way group level user stereotypes and categories, as previously they read news articles, for example, how much a person chooses mentioned, particularly in relation to natural dialogues [10] and to read of a news article [28]. Those variations are also likely to accessible systems for users with disabilities [26]. User modelling conform with particular profiles in the sense of stereotypical has also been demonstrated with mobile devices that log patterns for accessing and reading news. Therefore, the idea of interaction data, including interactions with search engines [7] and recommending variant user interfaces for particular news reader with web pages [11], and using function usage histories to refine types would be suitable for this kind of personalization. Although, menu displays [15, 23]. Further, Billsus and Pazzani [8] have one can argue about the effectiveness of this approach due to the demonstrated the use of supervised learning methods to develop a limitation of categorizing news readers as one of three kinds. news recommendation system. They proposed NewsDude, which Human behaviour is more complex and people can fall in between uses a combination of algorithms to model short-term and long- categories or behave as a different kind per day. We acknowledge term user’s interests. the limitations, but nevertheless, this approach is a preliminary Therefore, having a successful user modeling component that step to user interface personalization. Future directions may classifies users as one of three kinds is at the core of our approach include a more sophisticated way of identifying news readers to provide user interface personalization. types or a combination of them. This also reflects the variant interfaces design at the point of selecting features to create an 2.4 A News App’s Personalized User Interface individual variant interface. Our approach focuses on the idea of user interface personalization To sum up, user interface personalization is less explored in news in news apps compared to the widely applied news content services compared to news content recommendation. Despite the recommendation. Having an effective way of identifying user’s adaptive features of news apps through customizations, we news reading type, we envision a news app, which alters its believe mobile news apps are in the need of more sophisticated ways of user interface adaptive personalization. We believe this is Recent advances using soft computing techniques. Expert the time that both news content recommendation and user Systems with Applications, 29(2), 320-329. interface personalization should coincide to constitute the ideal [15] Fukazawa, Y., Hara, M., Onogi, M., & Ueno, H. 2009. news personalized app. Automatic mobile menu customization based on user operation history. In Proc. ACM MobileHCI, 50. 4. CONCLUSION In this paper we present news personalization as a two- [16] Gajos, K. Z., Czerwinski, M., Tan, D. S., & Weld, D. S. dimensional problem, characterized by what contents users access 2006. Exploring the design space for adaptive graphical user and how users access and interact with news content. We realize interfaces. In Proc. of the working conference on Advanced the importance of user interface personalization in news apps and visual interfaces, 201-208 we discuss our approach towards this direction. Our previous [17] Gauch, S., Speretta, M., Chandramouli, A., & Micarelli, A. work supports the feasibility of user interface personalization and 2007. User profiles for personalized information access. we conclude that news content recommendation and user interface In The adaptive web. Springer Berlin Heidelberg, 54-89. personalization should co-exist in news apps. [18] Grzeschik, K., Kruppa, Y., Marti, D., & Donner, P. 2011. 5. REFERENCES Reading in 2110-reading behavior and reading devices: a [1] BBC R&D. Elastic News – skimming and digging using case study. The Electronic Library. atomized, http://www.bbc.co.uk/rd/blog/2015/09/elastic- [19] Horvitz, E., Breese, J., Heckerman, D., Hovel, D., & news-on-a-mobile, 2015. Rommelse, K. 1998. The Lumiere project: Bayesian user [2] Ofcom. News consumption in the UK. Ofcom report, 2014. modeling for inferring the goals and needs of software users. In Proc. of the Fourteenth conference on Uncertainty in [3] Pew Research Centre. The Future of Mobile News. Public artificial intelligence. Morgan Kaufmann Publishers Inc, report, 2012. 256-265. [4] Reuters Institute. Digital News Report, [20] Liu, J., Dolan, P. and Pedersen, E.R. 2010. Personalized http://www.digitalnewsreport.org/survey/2015/social- news recommendation based on click behavior. In Proc. IUI, networks-and-their-role-in-news-2015/, 2015. 31-40. [5] Reuters Institute. Tracking the future of news. Public Report, [21] Liu, Z. 2005. Reading behavior in the digital environment: 2014. Changes in reading behavior over the past ten years. Journal [6] Adomavicius, G. and Tuzhilin, A. 2005. Toward the next of documentation 61.6, 700-712. generation of recommender systems: A survey of the state- [22] Mackay, W. E. 1991. Triggers and barriers to customizing of-the-art and possible extensions. Knowledge and Data software. In Proc. ACM SIGCHI, 153-160. Engineering, IEEE Transactions on, 17(6), 734-749. [23] Nivethika, M., Vithiya, I., Anntharshika, S., & Deegalla, S. [7] Bertini, E., Calì, A., Catarci, T., Gabrielli, S., & Kimani, S. 2013. Personalized and adaptive user interface framework for 2005. Interaction-based adaptation for small screen devices. mobile application. In Proc. of the ICACCI, 1913-1918. User Modeling 2005. Springer Berlin Heidelberg, 277-281. [24] Phelan, O., McCarthy, K. and Smyth, B. 2009. Using twitter [8] Billsus, D., & Pazzani, M. J. 2000. User modeling for to recommend real-time topical news. In Proc. ACM adaptive news access. User Modeling and User-Adapted conference on Recommender systems, 385-388. Interaction, 10(2-3), 147-180. [25] Sears,A. &Jacko,J.A. 2009 .Human Computer Interaction [9] Brusilovsky, P., & Millán, E. 2007. User models for adaptive Fundamentals: CRC Press. hypermedia and adaptive educational systems. The adaptive web. Springer-Verlag, 3-53. [26] Stephanidis, C., Paramythis, A., Sfyrakis, M., Stergiou, A., Maou, N., Leventis, A., ... & Karagiannidis, C. 1998. [10] Carberry, S., Carbonell, J. G., Chin, D. N., Cohen, R., Adaptable and adaptive user interfaces for disabled users in Lehman, J. F., Finin, T. W., ... & Wahlster, W. User Models the AVANTI project. In Intelligence in Services and in Dialog Systems. A. Kobsa, & W. Wahlster (Eds.) 2012. Networks: Technology for Ubiquitous Telecom Services, Springer Science & Business Media. Springer Berlin Heidelberg,153-166. [11] Carreira, R., Crato, J. M., Gonçalves, D., & Jorge, J. A. [27] Weld, D. S., Anderson, C., Domingos, P., Etzioni, O., Gajos, 2004. Evaluating adaptive user profiles for news K., Lau, T., & Wolf man, S. 2003. Automatically classification. In Proceedings of the ACM IUI, 206-212. Personalizing User Interfaces. IJCAI, 1613-1619. [12] Constantinides, M., Dowell, J., Johnson, D., & Malacria, S. [28] Westlund,O. 2008. From mobile phone to mobile device: 2015. Exploring mobile news reading interactions for news News consumption on the go. Canadian Journal of app personalisation. In Proceedings of the ACM Communication, 33(3). MobileHCI, 457-462. [29] Zancanaro, M., Kuflik, T., Boger, Z., Goren-Bar, D. and [13] Findlater, L., & Gajos, K. Z. 2009. Design space and Goldwasser, D. 2007. Analyzing museum visitors’ behavior evaluation challenges of adaptive graphical user patterns. In User Modeling, Springer Berlin Heidelberg, 238- interfaces. AI Magazine, 30(4), 68. 246. [14] Frias-Martinez, E., Magoulas, G., Chen, S., & Macredie, R. 2005. Modeling human behavior in user-adaptive systems: