Using what is without question the most unusual (innovative) data visualization technique I've ever seen, Postgenomic has created Pubmed Faceoff. The idea is simple: humans have evolved highly sophisticated face recognition abilities; why not exploit this ability in the visualization of multidimensional datasets? The technique, known as Chernoff Faces, has been around for awhile now, although the idea of using photorealistic faces seems to be an innovation of the Postgenomic team.

The service itself is actually quite usable, if a bit slow. Only newer papers in PubMed are covered, but that's still enough to get a feel for what's possible.

One feature I'd like to see: depict scientific subjects using faces. For example, just looking at the pathetic crowd I got by typing the keyword "schizophrenia" into the search box seems to suggest that field has some problems. A few stars, but apparently a lot of junk.

Another idea: click on any face and get a page containing all of the papers with that face. In other words, show me all of the papers getting more than the expected number of citations. Or show me all of the papers published in "crap journals" getting no citations. Average citation count seems much less interesting for some reason.

Credit: Thanks to Scholarship 2.0 for the lead.

With the ever-increasing volume and heterogeneity of data that we need to cope with, data visualization is a skill that every effective scientist needs to master. Like most essential scientific skills, data visualization is one of those things that is not so much learned in classes, but rather absorbed from one's surroundings. So if you want to improve your data visualization skills, where can you turn?

By way of Guy Kawasaki's blog, I came across a beautiful and informative summary of data visualization approaches published in Smashing Magazine. The article contains dozens of images and summaries of visualization techniques, tools, and services, grouped by category. The next time you find yourself trying to show yet another 30x30 table on a PowerPoint slide, why not experiment with a more compelling and effective way?