Humanities Centered Design Features: Emergent Serendipity with HuViz

Introduction
The Digital Humanities (DH) community has made several attempts to foster serendipitous experiences through digital visualization tools (Ridge et al., 2014; Martin et al., 2017). Some tools (Bohemian Bookshelf (Thudt et al., 2012) and StackLife (Innovation Lab, 2012)) visualize bookshelves, hoping to produce the 'aha' moments that frequently occur in libraries and archives. Visualization tools developed specifically for exploring linked data, (De Vocht et al., 2015), (Heim et al., 2010), demonstrate the possibilities for serendipity that arise when information is semantically connected. Recently, the research team behind FERASAT, an interface for exploring linked datasets, identified design elements that enhance the potential for making unexpected connections (Khalili et al., 2018).
These directed attempts at serendipity design reflect the importance of the accidental acquisition of information to humanities scholars. Sometimes, however, the aspects of visualization that lend themselves to serendipity are themselves discovered accidentally. This paper describes HuViz, the Humanities Visualizer, a digital tool for visualizing semantic relationships and ontologies represented using the Resource Description Framework (RDF). Though not originally conceived as an environment to foster serendipitous experiences, user studies conducted over the past two years indicate that it does just that. Moreover, unlike FERASAT, aimed at STEM and the social sciences researchers, HuViz was built with humanistic inquiry in mind. We here demonstrate how HuViz aligns with features identified in library and information science (LIS) literature as fostering serendipity. We conclude by introducing pending features designed to enhance its serendipitous potential.

HuViz: Background and use

Figure 1. The HuViz Shelf

OrlandoVision (OViz), the precursor to HuViz, was prototyped in 2010 as a tool to display extracts from The Orlando Project’s textbase (Brown, Clements, and Grundy, 2006-2018) as a series of interconnected nodes in a graph, allowing for a new method of exploring this large dataset (Holland and Elford, 2016). Initial experiments confirmed that such tools open up new hermeneutical pathways (Rockwell and Sinclair, 2016; Drucker 2014). The next generation of the tool, known as HuViz, grew out of the recognition of its potential for much wider applications beyond Orlando. In its current state, HuViz’s browser-based, interactive interface allows for the exploration of semantic relationships and ontologies represented in RDF. The shift positions HuViz to be compatible with Linked Open Data (LOD) from across the Web.
HuViz users begin by selecting an RDF dataset (in Turtle format) and related ontology. The initial visualization is a circular 'shelf' containing all the nodes found in the dataset (Fig. 1). There are two ways to interact with the data in HuViz. The first method, for users unfamiliar with ontologies, is to simply drag a node into the center of the shelf. Once this original node is released, any nodes that are connected to it by one or more predicates will follow it into the graph, resulting in a small network of interconnected nodes (Fig. 2). Additional nodes can be dragged into the centre of the graph or “stage” to explore more connections, and unwanted nodes can be dragged back to the shelf. The Command Panel at the right of the shelf offers the second set of affordances (Fig. 3). This Panel allows users to select groups of nodes by class, edge/predicate, or set and perform operations on groups. Some familiarity with ontologies is desirable for using this second option, though either method of interaction lends itself to information exploration.

Figure 2. Network appears after one node is dragged into center.

User-testing
User testing defamiliarizes a tool for tool makers, alerting the team to aspects of design and functionality that are only visible to an outsider, making this an important step in its development (see also: Terras et al., 2011; Ruecker et al., 2011). Testing brought our attention to the value researchers place on serendipity as well as to the ways in which HuViz facilitates unplanned discoveries.
We conducted 23 user tests with graduate students, senior scholars, library professionals, and members of CWRC projects. The tests took the form of a combined survey-tutorial, beginning with the simple task of building a network using the drag-and-drop technique, gradually working up to more difficult, directed queries. Participants explored datasets generated from Orlando’s textbase of women writers, including a dataset centred on Ada Byron and another of famous cookbook authors. Participants could generate networks based on the node type, discovering, for instance, various jobs held by Jewish novelists. Finally, by clicking on links between nodes, users could see the texts from which the data was generated, learning for example about the lesser-known connection between Ada Byron and Charles Dickens. In both the survey responses and the post-test interviews, we found that users valued the ability to contextualize relationships, to draw upon their own expertise to customize the dataset, and to follow their own exploratory pathways through the data. This feedback helped to inform the next stage of the tool’s development.

Figure 3. The HuViz Command Box

Serendipitous elements of design
Thudt et al. (2012) highlight several design features that they believe encourage serendipity, drawing on several LIS articles. These include: Multiple visual access points; Highlighting adjacencies; Flexible visual pathways; Enticing curiosity; and Playful exploration. Since 2012 a number of studies in LIS and DH have highlighted features of digital tools that encourage serendipity. Ridge et al (2014) used the concept of play to guide their development of Serendip-o-matic. Khalili et al (2018), in describing FERASAT, created a table of 12 design features related to serendipity. Two features relevant to our work are tools that support background knowledge and user contextualization and supporting convergent and divergent information behaviour.
As HuViz was not designed with serendipity as a primary goal, the research team did not have a similar list of features that support serendipity as objectives during development. However, user tests indicate that several design features actively support serendipitous information encountering as described in the literature. These design features are briefly described below:

Playful interaction and enticing curiosity
HuViz is unusual in providing the ability to switch between two modes of inquiry, each of which lend themselves to serendipitous discovery in different ways and are also connected to the affordance of “having multiple visual access points” (Thudt et al., 2012). The hands-on mode encourages spontaneous, intuitive interaction, which in turn invites users to further explore both the dataset and the capabilities of the tool. Indeed, during testing, participants would frequently deviate from the set tasks and start experimenting on their own. Their tendency to go ‘off script’ suggests that most found working directly in the graph not only intuitive but enjoyable, moving from simple play to curiosity about other affordances of HuViz.

Convergent information seeking
The Command Panel interface is less intuitive, requiring additional practice in order to familiarize oneself with the various functions. In this sense, it lends itself to more focused and purposeful, rather than exploratory investigation. At the same time, the ability to manipulate the dataset in more complex and sophisticated ways, which comes with higher levels of expertise, creates new opportunities for unexpected findings. Martin and Quan-Haase (2016) show how chance discoveries often incorporate “an element of intention.” Similarly, the HuViz Command Panel lets the user narrow their searches through related material, fostering serendipitous experiences through more directed forms of inquiry.

Divergent information seeking
HuViz also allows for multiple ways to diverge from the task at hand, allowing users to come across related information that they were not expecting. Simply dragging a node onto the stage also moves in all nodes connected within one degree in as well - immediately opening up users to a series of relationships. Following the edge from one node to another, users can see the relationship between these two nodes, and can continue dragging in or activating more nodes to expand their inquiry. In addition to this, once they find an edge, or relationship between nodes, that catches their interest, they can click on the edge to display a pop-up box to provide context (see below) and a link to the source of the data.

Contextualization of information
Martin and Quan-Haase (2016) found that historians were reluctant to use digital search strategies because of a lack of contextual information. HuViz provides context by providing 'snippets' of information to users when they click on the edge connecting two nodes, highlighting a portion of the original text from which the relationship was extracted (Fig 4). For further information, users can click through to the Orlando textbase (if they have institutional access) and expand the context further.

Figure 4. Once an edge is clicked, a "snippet box" appears to show context.

Conclusion and future steps
Our user testing has had two main results: inspiring us to push forward with HuViz and motivating us to further foster serendipity, leading to the development of several new features, which we will briefly demonstrate. These include optimizing HuViz to work with the Web Annotation data model (further enhancing the ability to contextualize information), expanding the history function with a “path” option (highlighting adjacencies in the data, and creating a visual pathway for users), and responding to SPARQL queries (allowing users with background knowledge to traverse a web of data relevant to their interests).

Bibliography

Brown, S., Clements, P. and Grundy, I. (2006-2018). The Orlando Project. The Orlando Project: Women’s Writing in the British Isles from the Beginning to the Present.

De Vocht, L., Verborgh, R., Mannens, E. and Seidl, T. (2015). Improving Semantic Relatedness in Paths for Storytelling with Linked Data on the Web. The Semantic Web: ESWC 2015 Satellite Events: Conference Proceedings. Portorǒz, pp. 31-35.

Drucker, J. (2014). Graphesis: Visual Forms of Knowledge Production. Harvard University Press.

Heim, P., Lohmann, S. and Stegemann, T. (2010). Interactive Relationship Discovery via the Semantic Web. In Aroyo, L. et al. (eds), The Semantic Web: Research and Applications. ESWC 2010. Lecture Notes in Computer Science, 6088: 303–17.

Holland, K., and Smith Elford, J. (2016). Textbase as Machine: Graphing Feminism and Modernism with OrlandoVision. In Ross, S. and O’Sullivan, J. (eds) Reading Modernism with Machines: Digital Humanities and Modernist Literature, Springer, pp. 109–34.

Innovation Lab, Harvard Library. (2012). StackLife | About. StackLife. http://stacklife.harvard.edu/about.

Khalili, A., Van Den Besselaar, P. and De Graaf, K. A. (2018). FERASAT: A Serendipity-Fostering Faceted Browser for Linked Data. In Gangemi A. et al. (eds) The Semantic Web. ESWC 2018. Lecture Notes in Computer Science, 10843: 351–66.

Martin, K., Greenspan, B. and Quan-Haase, A. (2017). STAK – Serendipitous Tool for Augmenting Knowledge: A Conceptual Tool for Bridging Digital and Physical Resources. Digital Studies/Le Champ Numerique 9. https://www.digitalstudies.org/articles/10.16995/dscn.265/.

Martin, K., and Quan-Haase, A. (2016). The Role of Agency in Historians’ Experiences of Serendipity in Physical and Digital Information Environments. Journal of Documentation 72 (6).

Ridge, M., Croxall, B., Papaelias, A. and Kleinman, S. (2014). Play as Process and Product: On Making Serendip-O-Matic. Digital Humanities 2014: Conference Abstracts. Lausanne: EPFL and UNIL, Switzerland, pp. 327-329.

Rockwell, G., and Sinclair, S. (2016). Hermeneutica: Computer-Assisted Interpretation in the Humanities. MIT Press.

Ruecker, S., Radzikowska, M. and Sinclair, S. (2011). Visual Interface Design for Digital Cultural Heritage: A Guide to Rich-Prospect Browsing. Ashgate.

Terras, M., Ross, C. and Warwick, C. (2011). Building Useful Virtual Research Environments: The Need for User-Led Design. In University Libraries and Digital Learning Environments., Ashgate, pp. 151–68.

Thudt, A., Hinrichs, U. and Carpendale, S. (2012). The Bohemian Bookshelf Supporting Serendipitous Discoveries through Visualization. CHI ’12: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. Austin, Texas, pp. 1461-1470.