The Most Interesting Papers at the Infovis Conference (VisWeek 2011)

GestaltLines
This post was written by Kim Rees, a partner at Periscopic (@periscopic), an interactive design firm specializing in data visualization and information presentation.

VisWeek has come and gone, but you can still get your fill by finding most of the papers online. I was able to attend most of the InfoVis and some of the TVCG tracks, and was really excited by much of the work.

While many of the research was focused on trying to "do something better," there was one paper that presented a novel, new type of data visualization. GestaltLines (PDF) by Ulrik Brandes and Nick Bobo of the University of Konstanz used balance to visualize dyadic relationships. Even in its most basic form, a 'Gestaltline' shows type, extent, and time of the relationship. Color is left as a degree of freedom to encode other variables. Using a sparkline or multivariate glyph approach, a gestaltline can easily be placed within text as a dataword. The technique seems like a very intuitive way of viewing relationships.

Another talk I found intriguing was called Discursis (or "Conceptual Recurrence Plots" according to the paper title) by Daniel Angus, Andrew Smith and Janet Wile. By using colored squares plotted on the diagonal, this method visualizes the strength of engagement in a dialogue. Using doctor/patient conversations as their case study, Discursis easily showed which meetings were beneficial to the patient. I can see this method being applied to a number of scenarios.

ScreenShot037 copy.jpg There were a number of papers dealing with optimizing edge bundling and improving visual routing. Of the latter, a good method (PDF) was provided by
Markus Steinberger, Manuela Waldner, Marc Streit, Alexander Lex, and Dieter Schmalstie of Graz University of Technology which preserves as much of the context, whether it's text, image, map, etc., while still providing visually clear links between highlighted items. Their paper is a good read for various techniques as well as criteria to follow for implementation. [editor's note: it seems also ideal to start a discussion about 'Best Paper' awards and the role of academic research in data visualization in general, see Stephen Few's comments here]

Juhee Bae and Ben Watson presented a new way to visualize trees called Quilts (PDF). Rather than the typical node-leaf approach, Bae uses a row-column way of looking at the structure. Personally, I found the learning curve to be very high, but then pretty easy to use. In some ways it seems to introduce visual clutter, but it also feels easier to navigate.

Being strongly rooted in the application of visualization, it's often nice to be validated by the academic and research communities. Lyn Bartram of Simon Fraser presented their evaluation of various energy use displays (PDF) from standard graph-based displays to very artistic visualizations. She stated what most of us already know; that people stay more engaged with artistic displays than standard charts. And Yu-Shuen Wan and Ming-Te Chi (PDF) scientifically prove what all good designers know - where to place labels on a m! ap.

While not the most elegant interface, ChronoLenses (PDF) by Jian Zhao, Fanny Chevalier, Emmanuel Pietriga, and Ravin Balakrishna offers a very robust interface for exploring time-series data. There were many interactive features that allow the user to zoom into regions of data, layer data, stretch data, and so on. It has a very deliberate and well considered feature set that I'm sure will be very useful.

While I enjoyed most of the talks, there were definitely some weak ones. LineSets by Microsoft Research topped my charts with a confusing rendition of set visualization (note: they're not showing routes in the above image, just locations). They should have dispensed with the obtrusive lines and just titled their paper "Sets" instead. My biggest disappointment, however, was BirdVis (NYU/Cornell), primarily because I had such high hopes for it after seeing it on the program. It's billed as a visualization of bird environments, but is nothing more than choropleth maps layered with draggable tag clouds that change with location (just when I thought there was no! thing worse than a tag cloud). I really want to like Image-based Edge Bundling (PDF) (Univ of Groningen), but it's such an eyesore. Gradients may be scientifically better (are they?), but my eyes avert when presented with one of these visualizations.