Emerging Tastes: Considering How Preferences Evolve Duane Degler Design for Context Washington, DC, USA duane@designforcontext.com ABSTRACT Our work in cultural heritage has been focused primarily on Experiencing cultural heritage is a voyage of discovery and the user experience and design of applications for museums learning, where emerging insights and serendipity play a and archives, and helping institutions plan to incorporate significant role. The experience also happens in a blended linked data to enhance people’s experience with their personal and social context. At the broader level, cultural assets. We increasingly support institutions that engagement is longitudinal – what we learn from modern want to share and enrich data via federated approaches, cultural experiences (daily life in our surroundings) can allowing information access to expand over time and across provide clues to analogous interests in cultural materials, institutional boundaries. Rich personalization is important. and vice versa. The richness of personal experience poses Many tasks will need a greater level of support as the challenges and opportunities for capturing preferences in volume of digital information – and associations via linked ways that support a user’s experience with cultural heritage data capabilities – grows in the coming years. As we design across institutions and over time, both in the digital realm interfaces, we explore the personal experiences people have and where digital interaction blends with a physical space. with cultural heritage information, whether online or in Author Keywords person, and how their expectations and interactions with Cultural heritage; digital humanities; personalization; cultural heritage evolve over time. The considerations that context; information architecture; user experience; linked arise from that work are the subject of this paper. data; LODLAM. CONSIDERATIONS ACM Classification Keywords Designing personalized interactions that are H.3.3 [Information Storage and Retrieval]: Information companionable, flexible, and not awkward is vital. To Search and Retrieval—Information filtering; H.5.m achieve that aim, personalization models need to provide a [Information Interfaces and Presentation]: Miscellaneous great degree of user transparency [9], control and be open to many outside signals that respond to new experiences INTRODUCTION and changing tastes, and are carefully aligned with a user’s Cultural heritage institutions recognize the need to extend needs and expectations. These aspects are not static – they their digital strategies as they gain an understanding of change in specific contexts and evolve over time, forming emerging technologies and web-scale linked open data. longitudinal patterns that may help inform how This means moving beyond “making things available” and personalization models adapt. allowing users to save “personal collections” to more sophisticated and personalized experiences. This has broad Preferences Interact with User Knowledge implications for data management that goes beyond At any one time, a person’s interaction with cultural publishing collection and image metadata. It prompts a new heritage has a purposeful dimension – whether that is wave of design thinking for cultural sites and applications, entertainment, knowledge-seeking for its own sake, or finding ways to meet the needs of diverse user types and resource-seeking for some outside task. scenarios of use. Personalizing lifelong learning presents To achieve a goal of supporting lifelong learning, it is opportunities to suggest new areas of exploration and important to focus on the general user experience with discovery that enrich experience, particularly when digital cultural heritage interactions. Yet there are spanning multiple institutions, cultures, and subjects. specialized audiences who have their own needs for It is important to recognize that the domain’s data itself is a personalization. Scholars and practitioners in the field of moving target – available data (particularly linked open cultural heritage have a driving motivation for discovering, data) is emerging rapidly. And there are a growing number interpreting, analyzing, synthesizing, publishing, and of initiatives for institutions to share and harmonize their sharing. Their requirements are particularly rich, due to data representations online, which means the potential to their specialized and detailed knowledge in particular integrate information models and user profiles across subject areas. Educators have varying levels of experience cultural institutions will continue to evolve. and knowledge, and they act as surrogates for others who do not share their level of knowledge. An educator’s interaction with cultural heritage can be largely driven by a Copyright held by the author. task that is focused outside themselves, as their role is primarily communication of cultural heritage information to others (although they also have their own personal interests when and how to interpret interaction as interest, and assess to balance with their professional focus areas). their actual longevity. It is useful to identify what elements of personal interest/engagement are relevant to the activities The richer a person’s experiences and knowledge, the richer available at a particular cultural venue (whether this is a and more nuanced their personalized modeling may need to formal cultural institution/site, an informal urban location, be. And the broader the possible goals and tasks, the more or a purely digital online interaction), and identify ways to contextually aware applications need to become. At the interpret what someone “takes away” from the experience. same time, single interactions between the individual and And algorithmic interpretation gets harder as the time gap an institution may be motivated by needs that are separate grows between the experience and the expression in an from preferences and knowledge. For example, Falk [6] electronic form. More distant value judgments (ratings, outlines five ways that individual identity is reflected in suggestions from one user to another, etc.) can have lower their actions within an institution, including rechargers, validity in recommendation algorithms [9]. experience-seekers, and facilitators. Preferences Evolve Through Lifelong Experiences Preferences have Scenarios and May be Transitory Where do personal preference signals come from? Cultural There are often situations where a person is engaging with linked data used in education, media, tourism, gaming will cultural heritage as an aspect of a very specific, directed produce aspects of personal interest that can be reflected task. This could be writing a paper as a school student, or back into cultural heritage preferences. This will require a researching a book as a scholar, or preparing a treatment for broader – and more nuanced – way of modeling a major artwork as a museum conservator. While a person “preference”, and reflects the PATCH1 workshop goal of will often be clear about their particular context when exploring a longitudinal perspective that encompasses interacting with cultural heritage, they may not externalize lifelong learning. Various dimensions to support and that context to a supporting technology. evaluate models have been proposed [10, 2]. When that task is completed, whether in two weeks or two At the broader level, engagement crosses timespans – what years, the intensity of focus decreases. In most cases there we learn from modern cultural experiences (daily life in our is still an interest in a particular subject or area of culture, surroundings) can perhaps translate to analogous interests but the goal that prompted a strong, focused interest may in cultural materials from other time periods [15]. have waned. The corresponding strength of the preference may need to be tuned accordingly. One further aspect to consider for lifelong models is identifying when a person transitions from experience to Preferences Emerge over Time When people encounter something for the first time, they knowing, and as a result is likely to need different kinds of may not immediately sense its significance. Tastes (and recommendation, as they are able to be more personally emotional connections to cultural objects or experiences) directed via their own knowledge and experience. In this way, applications and agents need to consider how much, are emergent and are often recognized only on reflection, and how little, to be involved in an experience. rather than “in the moment.” So something “stays with you” after the experience, or you recognize something as Preferences are Balanced by Serendipity valuable/important only in the context of subsequent Personalization models need to guard against over- experiences with other things (they form a pattern that simplicity or rigidity, and in that way foster discovery and makes a whole, and a preference forms at that higher level). learning. At a basic human level, people seek to engage Strong interests from one interaction may over time have with culture and art in order to find delight in something evolved or been eclipsed by more lasting connections with new as well as to experience known and loved objects. what were, at the time, weaker signals of engagement. Serendipity and discovery are a significant motivation for exploring cultural heritage. This is particularly true among In this way, personalization for cultural heritage may be scholars and curators, whose life work centers around more nuanced and longitudinal than preference and interpretation and discovering new perspectives. It is also recommendation modeling required in other domains. true for people who engage in conservation and built When interacting with art and exhibitions, it is also helpful cultural environments (archaeologists, historians, to recognize that the experience itself may be what is most preservationists, architects) who want to engage in the latest important to the user, not necessarily the specific object of science and interpretation. the experience. In a recent conversation, two museum-goers Some implementations of personalization can be restrictive, described in great detail an electronic exhibition piece even if the intention is to reduce informational “noise” and where they interactively engaged with art. They described increase relevance for a user. The “filter bubble” [13] term deep engagement with the experience but had trouble was coined to describe a concern where a system algorithm recalling the specific art that was the focus. As systems collect data about interactions (whether clicks, views, likes, saves, shares), it becomes important to discern 1 PATCH 2015: https://patch2015.wordpress.com/about-2 manages navigation through a significant glut of “…no matter how enabled by artificial intelligence, such information, but users are not easily able to go beyond the metamaps and compasses tend to become less accurate as boundaries of the algorithm’s filters and may not encounter they try to be smarter and more richly relevant to context. something that is valuable and engaging [12, 5, 4]. The bigger the gap we’re trying to bridge, the more it’s subject to the fog of ambiguity…” (pg.104) The overall aim of personalization needs to be transparent and controllable, so as to avoid becoming restrictive. It is Implications arise from rich multi-dimensionality, uneven vital for applications to open up new, unexpected (and yet interest weighting, and increasing ambiguity. Items that ideally tangentially-related) experiences for users in cultural users select among online artwork or cultural artifacts today heritage. Serendipity is not simply random encounters, could themselves have a different “profile” at a future date, rather it is a process that incorporates and synthesizes new changing the way that automated personalization systems things into experience [1] – and it needs to be fostered. It is then map between profiles of the art and the nature of a important to find patterns that can foster an “aha!” moment person’s interests over time. So not only is a person’s – that moment when they discover a relationship between longitudinal profile evolving, but the object models that are what they know and what they experience. drawn upon also evolve, with uncertain consequences. Preferences are Influenced by Social Interactions INFORMATION ARCHITECTURE AND DESIGN When I go to a museum with other people, we engage with At the information architecture level, how can publishers of things that Group/social interaction in physical space makes cultural heritage information and creators of cultural it harder to know what is persistently preferential for the experiences formulate dynamic information architectures individual rather than a reflection of an immediate social that respond appropriately to personal representations and group dynamic. What do I “keep for later” and transfer departures from those representations? And how can the between contexts, and what is purely “in the moment” for technology community establish models and frameworks my relationship with the people and place? that reflect the inherent dynamism in this data? One research area to consider is exemplified by the At the UI level, how can we use UI frameworks to broaden Epiphany Project [7]. This emerging research aims to and evolve experiences, without losing focus or analyze social media streams to identify how individual overwhelming the user? Is it feasible to craft an “ambient” interests, and institutional influences, are mirrored in what awareness of information and opportunities for an individual publishes via social media. engagement, without at the same time intruding on an individual’s primary experience? In addition, the role of intra-group profiling of personalized dimensions plays a role in weighting recommendations and For the overall experience, how do we establish design interactions in situations where the experiences are social. patterns that make sure a person has an easy way to “turn off” aspects of personalization that become dissonant to Another aspect of social interaction with taste-making is their immediate experience, or where their digital that a person’s knowledge of participants in a community interactions are out of the immediate context? For example, and the mutual alignment of interests can affect their using a mobile device to look something up based on an in- interpretation of how they judge recommendations [17]. progress conversation with a friend that is not related to the When in a social recommendation environment for some surrounding cultural space where they are located at that subject I don’t know well, I expect to use different value time? Their task context is separated from their physical judgments about other people’s preferences in relation to context for some period of interaction. In other words, my interests. And those judgments can grow and change make sure the algorithm is not in charge of the experience. over time as my interaction with those same people grows over time, calling for an evolutionary learning approach to The Role of a Guide or Agent my preferences based on social context. Recommendations in a cultural setting are likely not just focused on single objects, but a sequence of items in a place Preferences are Uneven Across Descriptive Dimensions that help craft an experience flow. We find it helpful to The dimensions of description and interpretation of cultural consider the role of attentive guide as an aspect of material and places are deeply multi-faceted. As we know personalization in cultural heritage. that not all dimensions have equal weight in a person’s engagement with culture [15], finding longitudinal patterns Creating an emerging personal profile could involve is important for discerning relative weighting of interests interacting with an agent that is focused on your derived from personal experience. personalized experience; one that both guides and listens to a person’s expressions of interest [14]. One perspective on As we consider rich dimensions, it is important also to this involves the role of the “information flaneur” [5]. This consider the challenges that arise from such deepening data is an independent, knowledgeable agent who provides a pools. In the recent book Understanding Context [8], perspective on overall information spaces, as well as being Andrew Hinton writes: a guide to more specific information objects. The agent embodies properties of curious explorer, critical spectator, and creative mind to prompt new perspectives. It is useful • Recognize shared and social interactions: Models to consider what such a guide would need to know about could usefully identify social contexts that people are in the individual and the alignment with the cultural space at when preferences are engaged, and have ways of re- hand, as well as motivations for any particular interaction, aligning their weightings accordingly – or prompt the for example as framed by Falk [6]. The flaneur could offer user to take greater control of the experience. a launch point for refining the role of recommendation and • Allow dynamic weighting: Recognize context and user guidance in subjective, interpretive learning settings. expectations, and provide an appropriate level of control Who Controls the Data? for a person to express goals and needs. Then have those Beyond the individual’s experience, is there a role for an expressions reflect back into the preference model. aggregation of experience patterns across many individuals • Provide simple frameworks for permission-giving: and institutions? How might those aggregations be Plan for an emerging ecosystem of information around consumed and used by an institution, ideally in ways that personal interests information and preferences. Identify increase diversity of experience rather than homogenize? how to make the collection and use of information as How might they be shared among institutions, so that they transparent as possible, to foster trust and communication can craft the way their applications respond to personalized among the parties involved (whether humans, institutions, needs in ways that create more seamless experiences for or digital agents) [3]. individuals as they move among cultural sites? Personal profiling must, in this context, move beyond REFERENCES individual applications and institutions. Are there aspects of 1. André, P., schraefel, m.c., Teevan, J., and Dumais, S.T. personal models that should not only be linked data, but Discovery is never by chance: designing for linked open data so that the models can be extended and (un)serendipity. In Proceedings of the seventh ACM built upon? It is interesting and important to consider to conference on Creativity and cognition (C&C '09), pg what extent a profile of personal interests remains under the 305-314. New York, NY, USA, ACM, 2009. control of the person (for example, carried with the person http://doi.acm.org/10.1145/1640233.1640279 in their mobile smart phone [11]), and how much of the 2. Ardissono, L., Kuflik, T., and Petrelli, D. profile needs to be shared with a cultural institution for a Personalization in cultural heritage: the road travelled relevant experience to be crafted. It seems clear that the and the one ahead. In User Modeling and User-Adapted representation of personal interest is best held by the person Interaction 22, 1-2 (April 2012), 73-99. rather than individual institutions or a data aggregator, but http://dx.doi.org/10.1007/s11257-011-9104-x that raises questions that go beyond cultural heritage personalization. 3. Cramer, H., Evers, V., Ramlal, S., van Someren, M., Rutledge, L., Stash, N., Aroyo, L. and Wielinga, B. The Focusing on the institution’s perspective, how does a effects of transparency on trust in and acceptance of a particular cultural environment access an individual’s content-based art recommender. Journal of User personalized model to align the person’s experience with Modeling and User-Adapted Interaction (UMUAI), information or an activity? What permissions might be Volume 18, Number 5, 2008. required, and perhaps how would an automated agent be http://dx.doi.org/10.1007/s11257-008-9051-3 empowered to provide that on a person’s behalf? 4. Cremonesi, P., Garzotto, F., and Turrin, R. Investigating POSSIBLE ATTRIBUTES OF PERSONAL MODELS the Persuasion Potential of Recommender Systems from Reflecting on the above considerations for personalization a Quality Perspective: An Empirical Study. ACM Trans. leads to attributes that could be incorporated into Interact. Intell. Syst. 2, 2, Article 11, 2012. frameworks explored by PATCH participants, as well as http://dx.doi.org/10.1145/2209310.2209314 others in the cognitive computing and HCI communities. 5. Dörk, M., Carpendale, S., and Williamson, C. The • Longitudinal preference building: Both the elements of information flaneur: a fresh look at information seeking. preference (signals of interests expressed by an In Proceedings of the SIGCHI Conference on Human individual) and their strengths may be accrued over time, Factors in Computing Systems, 2011. so the longevity of interest and the context in which it http://doi.acm.org/10.1145/1978942.1979124 arose is evaluated over time. 6. Falk, J. Identity and the museum visit experience. Left • Organic movement and decay: Individuals can remain Coast Press (2009). interested in things long after their focus or tastes have http://www.lcoastpress.com/book.php?id=214 changed. But models need to provide a method of “decay” for interests that are not acted on, as preferences change over time, and outdated preferences can be perceived as noise by users. 7. Gerrard, D., Jackson, T. and O'Brien, A. The Epiphany Part II (ISWC'12). Vol. Part II, pg 391-398. Berlin, Project: Discovering the Intrinsic Value of Museums by Heidelberg: Springer-Verlag, 2012. Analysing Social Media. In Museums and the Web http://dx.doi.org/10.1007/978-3-642-35173-0_28 2014, N. Proctor & R. Cherry (eds). Silver Spring, MD: 13. Pariser, E. The Filter Bubble. What the Internet is hiding Museums and the Web (2014). from you. London: Viking/Penguin Press, 2011. http://mw2014.museumsandtheweb.com/paper/the- epiphany-project-discovering-the-intrinsic-value-of- 14. Richards, D. Agent-based museum and tour guides: museums-by-analysing-social-media/ applying the state of the art. In Proceedings of The 8th Australasian Conference on Interactive Entertainment: 8. Hinton, A. Understanding Context: Environment, Playing the System. Article 15, 2012. Language, and Information Architecture. California, http://doi.acm.org/10.1145/2336727.2336742 USA: O'Reilly Media, 2014. Quote pg 104. 15. Wang, Y., Stash, N., Aroyo, L., Hollink, L. and 9. Kay, J. "Scrutable adaptation: Because we can and Schreiber, G. Using Semantic Relations for Content- must." In Adaptive hypermedia and adaptive web-based based Recommender Systems in Cultural Heritage. systems, pp. 11-19. Springer Berlin Heidelberg, (2006). Workshop on Ontology Patterns (WOP) at International 10. Koren, Y., Bell, R., Volinsky, C. Matrix Factorization Semantic Web Conference (ISWC), 2009. Techniques for Recommender Systems. Computer, http://chip.win.tue.nl/presentation/wangetal-wop.pdf vol.42, no.8, (2009). 16. Wecker, A.J., Kuflik, T., Stock, O. Personalized http://dx.doi.org/10.1109/MC.2009.263 Cultural Heritage Experience outside the Museum: 11. Kuflik, T., Kay, J., and Kummerfeld, B. Lifelong Connecting the outside world to the museum personalized museum experiences. In Pervasive User experience. In PATCH Proceedings, (2013). http://ceur- Modeling and Personalization (PUMP'10) at ws.org/Vol-997/patch2013_paper_3.pdf UMAP2010, (2010). 17. Ziegler, C-N., Golbeck, J. Investigating interactions of 12. Maccotrazzo, V. Burst the filter bubble: using semantic trust and interest similarity, Decision Support Systems, web to enable serendipity. In Proceedings of the 11th 2006. http://dx.doi.org/10.1016/j.dss.2006.11.003 international conference on The Semantic Web - Volume