Visual Information-Seeking Mantra

This question is related to the guideline: Overview first, zoom and filter, details on demand
Source: B Shneiderman, The Eyes Have It: A Task by Data Type Taxonomy for Information Visualizations. Proc. IEEE Symposium on Visual Languages, pp.336-343, 1996. http://citeseer.ist.psu.edu/409647.html

When working with domain scientists in CFD, we have very high-resolution data to be visualized. Domain scientists seem to prefer details first. I am wondering what was the reason. Someone told me that some biologists have the same preference in dealing with large datasets.

Correct, CFD domain scientists, some biologists, journalists, and no doubt other domain scientists operate under a Details-first model (not Overview-first). The rationale is more complex than we’d think (it’s not necessarily the large size of the dataset), and may vary with the domain.

The first factor is the imprecise definition of “details”: in some applications, the details are ill-defined and need repeated refinement. To the point, in CFD the details are spatial features of interest. Per Luciani et al. 2019 (http://ieeexplore.ieee.org/abstract/document/8440850), a number of these features do not have concrete/mathematical definitions, and as such typically require a skilled user in order to be visually identified, separated, and investigated: ```[the feature is] hard to define, but if you see it, you recognize it immediately’’; conversely, "it becomes difficult to draw conclusions from an overview first, when they ``do not know exactly what is present in the data’’`. Since a CFD specialist needs to first identify the details/features, that’s what their visual analysis tends to start with.

A second factor is the domain expert’s familiarity with the “overview”. Chen et al. (“Information Theory Tools for Visualization”, AK Peters) were the first to remark, in 2016, that in “many scenarios, we often observe that an experienced viewer may find [overview first and details on demand] frustrating, as the viewer knows exactly where the interesting part of a detailed representation is.” In CFD, domain scientists often work on the same problem for months, and have a good mental overview of the underlying data.

A third factor is the potentially large size of the overview, as first argued by van Ham and Perer in 2009 (http://ieeexplore.ieee.org/abstract/document/5290699) and then by Luciani et al. in 2019; creating an overview may not be feasible, in particular when a large dataset is being maintained at a centralized location, and transferring it to multiple client machines is not an option.

In a nutshell, Overview-first is not the only mantra we can and should use in visualization design. Alternative mantras are Search-first (van Ham and Perer, see above) and Details-first (Luciani et al., see above), and quite possibly many others. Overall, adoption of a particular design mantra should take into account the benefits, limitations, and possible co-existence of each approach, with careful consideration of data, user knowledge and interests, and user workflows.

In terms of what may be happening outside CFD, some biologists are also first and foremost interested in specific details, such as a specific bonding site (see Fig.2 in Chapman et al. “Initial genome sequencing and analysis of multiple myeloma”). In the “Data-driven Storytelling” book edited by Riche et al. 2018, AK Peters, one journalist also describes their approach to design as “Details-first”. From a wider perspective, “the Details-first model applies to domain expert workflows that start with an in-depth exploration of one model or simulation, then seek to extrapolate or generalize the findings to a collection of models. In such workflows, users may wish to start with the features of interest, in particular when those features are ill-defined and need repeated refinement” (Luciani et al, 2019).

Hope this helps. (I tried to provide links to all the refs above, but visguides only allows two links per post).

Thank you, Lizmarai, for your answer that is really thorough. I learned a lot. The decision as to using which of the three approaches, overview first, detail first and search first, seems to relate largely to the users’ knowledge. It sounds like:

  • If a user knows a lot, detail first,
  • If a user knows some, search first,
  • If a user knows little, overview first.

Am I right?

Sorry for the delayed reply, I meant to answer for a while.

I think your extended interpretation makes sense in some settings, but may not be a good fit for other settings.

My understanding is that all the three mantras documented so far (the “overview first” mantra, the “search-first” mantra, and the “details-first” mantra) can apply to situations where “the user knows little”, and also to situations where “the user knows a lot”.

For example, journalists would argue that casual readers (“the user knows little”) are drawn to stories that start with the details-first. As another example, Google Maps would argue that a casual maps user (“the user knows little”) would start with their current location, whence search-first. As another example, neither an expert astronomer nor a novice citizen scientist (e.g., a school teacher) would be necessarily interested in an overview of the known universe, despite the fact that one knows a lot, and one knows little. I think a better framing is “What is the intended user activity?”, instead of “What does the user know?”.

From a more formal perspective, I think the “user knowledge” interpretation is aligned with the Human-Centered-Design paradigm (the user is central). In contrast, the mantras are better aligned with the Activity-Centered-Design paradigm (the user activity is central). Don Norman has a nice exposition of ACD (which he fully endorses, over HCD) in the third edition of “The Design of Everyday Things”, Chapter 6. Hope this helps!