October 26, 2004
HCI Comments VIII
Readings in Information Visualization: Using Vision to Think, Chapter 1, Stuart K. Card, Jock D. Mackinlay, Ben Shneiderman, Morgan Kaufmann Publishers, pp. 1-34
The grounding of visualization techniques in the biology of the human perceptual system is essential. The authors make this point but do not have enough space to elaborate sufficiently on the specific properties of the human eye and visual pathway. An accessible psychology text that goes into further detail describing vision and other senses is: E. Bruce Goldstein, Sensation and Perception. Wadsworth, Sixth Edition, 2002. ISBN 0534539645.
While concepts relating to static presentation of data are discussed at length ("spatial substrate", "marks", and their graphical properties) the dynamic, temporal properties are only glossed over briefly. I believe more interesting work can be done here. Animation itself can be used to show physical or abstract phenomena, just as static properties of data marks. In sonification for example, all conveyed information is temporal in nature. Interestingly, different frequency ranges are perceived as different phenomena - can similar characteristic ranges be found for visual frequencies?
What about aesthetics and production value? The article explicitly states that many of the demo systems mentioned were not concerned with creating beautiful graphics. But maybe they should be. In "Emotional Design" I believe Don Norman argues that "attractive things work better" (haven't read the book yet- it's sitting on my desk). Can established graphic design practices and guidelines inform information visualization? Can they be captured in a set of heuristics?
Other odds and ends:
- The "Cost-of-Knowledge Characteristic Function" maybe more significantly altered by non-visual tools such as Google. Structure isn't everything.
- Cell animators have been using selective distortion and exaggeration - such as squash and stretch - to create believable characters. How can exaggeration be employed in info vis to guide user attention towards salient/unusual data points or patterns?
- The "overview+detail" technique could be generalized to show n different viewpoints of the same frame of complex data sets to overcome human difficulty in understanding higher-dimensional spaces.
Comments on the demo videos linked from the course page:
FILMFINDER: User interface designers should be sensitive to the importance of production values (cf. in-class discussion of Nielsen's bad graph in the Heuristic Evaluation paper). This demo has an incredibly overdriven audio channel that makes it nearly impossible to listen to with headphones.
TREEMAP: Shneiderman's group places emphasis on the concept of "dynamic queries" - yet their examples in the Treemap demo are all based on Visualization of Excel files that I assume are then manipulated internally in Treemap. Integrating their work with a relational database system that can answer the changing queries directly would make their approach more powerful.
In the example, users can specify their own color gradients. Which gradients maximize perceptible differences between elements is not intuitive. Maybe the color picker could be based on perceptual distances to aid the user?
The Table Lens: Merging Graphical and Symbolic Representations in an Interactive Focus+Context Visualization for Tabular Information, Ramana Rao and Stuart K. Card, CHI 1994: ACM Conference on Human Factors in Computing Systems, pp. 318-22
The paper's abstract tone made it difficult for me to imagine the operation of Table Lens in practice - the demo was much more instructive as to how their technique works. The principal contribution seems to be that much larger data sets can be shown at once than in standard spreadsheet applications. Also, the direct visual comparison affords an intuitive understanding of the characteristics of the data sets. Simple comparisons can be read off and don't have to be calculated. Grouping and complex relationships between a larger set of different variables are still hard to comprehend though since display is restricted to the 2D constraints of a flat area display. The paper was short on evaluation.
Posted by Bjoern Hartmann at October 26, 2004 11:37 PM