Database Reference
In-Depth Information
design requirements, it is very hard to convince a reader that your model will
solve the problem. In particular, you should consider how to make the case that
the structure you are visually showing actually benefits the target end user. For
example, many authors new to information visualization simply assert, without
justification, that showing the hyperlink structure of the web will benefit end
users who are searching for information. One of my own early papers fell prey
to this very pitfall [26]. However, after a more careful task analysis, I concluded
that most searchers do not need to build a mental model of the structure of the
search space, so showing them that structure adds cognitive load rather than
reduces it. In a later paper [25], I argued that a visual representation of that
hyperlink structure could indeed benefit a specific target community, that of
webmasters and content creators responsible for a particular site.
The foundation of information visualization is the characterization of how
known facts about human perception should guide visual encoding of abstract
datasets. The effectiveness of perceptual channels such as spatial position, color,
size, shape, and so on depends on whether the data to encode is categorical,
ordered, or quantitative [24]. Many individual perceptual channels are preatten-
tively processed in parallel, yet most combinations of these channels must be
serially searched [12]. Some perceptual channels are easily separable, but other
combinations are not [41, Chapter 5]. These principles, and many others, are
a critical part of infovis theory. The last three pitfalls in this section are a few
particularly egregious examples of ignoring this body of knowledge.
Hammer in Search of Nail:
If you simply propose a nifty new technique with
no discussion of who might ever need it, it's dicult to judge its worth. I am
not arguing that all new techniques need to be motivated by specific domain
problems: infovis research that begins from a technique-driven starting place
can be interesting and stimulating. Moreover, it may be necessary to build an
interactive prototype and use it for dataset exploration before it's possible to
understand the capabilities of a proposed technique.
However, before you write up the paper about that hammer, I urge you to
construct an understanding what kind of nails it can handle. Characterize, at
least with some high-level arguments, the kinds of problems where your new
technique shines as opposed to those where it performs poorly.
2D Good, 3D Better:
The use of 3D rather than 2D for the spatial layout of
an abstract dataset requires careful justification that the benefits outweigh the
costs [36]. The use of 3D is easy to justify when a meaningful 3D representation is
implicit in the dataset, as in airflow over an airplane wing in flow visualization or
skeletal structure in medical visualization. The benefit of providing the familiar
view is clear, because it matches the mental model of the user. However, when the
spatial layout is chosen rather than given, as in the abstract datasets addressed
through infovis, there is an explicit choice about which variables to map to
spatial position. It is unacceptable, but all too common with naive approaches
to infovis, to simply assert that using an extra dimension must be a good idea.
The most serious problem with a 3D layout is occlusion. The ability to in-
teractively change the point of view with navigational controls does not solve
Search WWH ::
Custom Search