Bryan Chan talked about Vispedia, their followup paper to last year’s “Visualization of Heterogenous Data”. There’s nothing particularly earth-shattering about the work, but the results are really nice – I played around with Vispedia last night, and it’s not hard to build visualizations.
I was expecting to play around with it some more during the talk, but the authors decided to hide it to make sure the live demo went well. Hopefully it will come back up soon. For comparison, Martin Wattenberg is giving a word tree demo on the live website. ManyEyes is much simpler, but the demo is much more impressive when I can play around with it too.
A few features I liked about Vispedia: it’s possible to check the final result of the integration, and see the paths that were used to make the table. Since they’re doing ad-hoc integration, it’s important to tell where things went wrong, and it is really easy to find those. I’m not sure how this scales – it took a few seconds for the system to build the visualization, and if more than a handful of users were trying this at the same time, I can expect it will be annoying.
Still, it’s great that people are trying visualization with data in the wild. That gives an incentive for people to make the Vispedia infoboxes more structured, and that in turn makes visualizations easier. Hopefully it’s a trend in the making.
Martin Wattenberg is talking about word trees, which have been in ManyEyes for a while now. They’re very simple, but what amazes me is that the graphics actually look great for Java applets – drawing on the web has come a long way. The “I am married, but” example was pretty striking. Also, he pointed out they choose font sizes so that word area is proportional to frequency (so point size is proportional to square root of frequency). Robert Kosara had a discussion just a few days about this in his site, and Martin just mentioned he had no experimental evidence about the effectiveness of the technique in word trees. That’s a paper waiting to happen.