Investigating the Manual View Specification and Visualization by Demonstration Paradigms for Visualization Construction

Abstract

Interactivity plays an important role in data visualization. Therefore, understanding how people create visualizations given different interaction paradigms provides empirical evidence to inform interaction design. We present a two‐phase study comparing people's visualization construction processes using two visualization tools: one implementing the manual view specification paradigm (Polestar) and another implementing visualization by demonstration (VisExemplar). Findings of our study indicate that the choice of interaction paradigm influences the visualization construction in terms of: 1) the overall effectiveness, 2) how participants phrase their goals, and 3) their perceived control and engagement. Based on our findings, we discuss trade‐offs and open challenges with these interaction paradigms.

DIVI: Dynamically Interactive Visualization

“…Snap-Together Visualization [19] applies PBD to multiple coordinated views, allowing end users to dynamically combine visualizations and configure interactions such as details-on-demand and linked brushing. Falx [36] and Saket et al's Visualization by Demonstration [27][28][29] explore how interactive demonstrations (e.g., re-positioning points in a scatter plot) can be used to infer and refine global transforms to the visual design (e.g., bin points in close proximity together), removing the need to provide manual or textual design specifications. These systems focus on chart specification, whereas DIVI concerns subsequent selection and view transformation.…”

Lyra 2: Designing Interactive Visualizations by Demonstration

“…Through systems like Gold [39] and a recent thread by Saket et al [47][48][49], researchers have shown that PBD is also a viable approach for designing visualizations. With PBD, rather than explicitly binding data fields to encoding channels, users implicitly specify these mappings by performing demonstrations -for instance, when the user drags two points together, the system infers that they intend to create a scatterplot and suggests several x-y axis pairs.…”

Understanding Visualization Authoring Techniques for Genomics Data in the Context of Personas and Tasks

“…Visualization authoring is a non-trivial task [23]. In visualization research, various techniques have been studied and proposed for interactively authoring visualizations, which all present trade-offs related to authoring criteria [3] and task effectiveness [52,53]. For example, template editors, such as MS Excel [44], enable easy and efficient construction of visualizations but result in constrained results and thus are at odds with expressiveness ("Can I build it?"…”

“…On the other side of the spectrum are programmatic approaches, such as Vega-Lite [57] and D3 [9], which are highly expressive but might present a steep learning curve [56]. The techniques also differ in suitability for specific lower-level authoring tasks [42,52]. For example, template editing is good for quickly creating a basic initial scaffold for the visualization, but further customizations are likely cumbersome or even infeasible.…”