Together with my colleagues in Seattle and Umea, I modified the “change over time” visualization to tell a specific story: The formation of neuroscience as a field of its own right over the last decade. Originally scattered across related disciplines (such as medicine, molecular and cell biology or neurology), the neuroscientific journals start to define a niche of their own, reflected in the dense cluster emerging in 2005.

Download a larger version with full explanatory text here: png (1MB) pdf (4MB)

And here is some more in depth info: First, almost 8000 scientific journals are clustered into groups, based on their citation patterns, and using the map equation (demo, paper). In short, for a network partitioned into groups, the map equation specifies the theoretical limit of how concisely we can describe a trajectory of a random walker on the network. Therefore, minimizing the map equation over all possible network partitions reveals regularities of information flow across directed and weighted networks or, in our case, the structure of how citations flow through science.

Second, using the Eigenfactor™ Score, the journals are assigned a measure of importance – much as Google’s PageRank algorithm ranks the importance of web pages. The Eigenfactor™ Score measures the percentage of time that researchers would spend with the respective journal, if they were to move through the network by randomly following citations in the journals.

This process is repeated in two-year chunks from 1999–2007, in order to capture changes in clustering and shifts in importance over the years. For this diagram, we picked only the clusters relevant to the formation of neuroscience.

In the visualization, each cluster occupies a vertical column block in the respective year’s column, further subdivided into a block for each journal. Each journal is connected with a horizontal band over the years. The height of each journal reflects the Eigenfactor Score. All journals in the cluster that corresponds to the field of neuroscience in year 2007 are highlighted to tell the story of the formation of this field of science. The coloring is based on the cluster assignments in the first year, 1999.

We use a subset of the citation data from Thomson Reuters’ Journal Citation Reports 1999–2007. The complete data aggregate, at the journal level, approximately 35,000,000 citations from almost 8000 journals over the past decade, but here we only display journals relevant to the formation of neuroscience.

Finally, the results of a cooperation with the guys from eigenfactor are online!

For the impatient: here’s the direct link: http://well-formed.eigenfactor.org

The site features 4 different visualizations, trying different approaches to mapping information flow and citation structure in the sciences.

A radial visualization based on hierarchical edge bundling does a great job of displaying the overall citation and clustering structure. I love the organic feel the bundled splines bring in, and multiplying the line colors added that deep extra color twist. Aesthetically, my clear favorite. Its downside is that it does not display the direction of information flow very well. Also, I am not 100% happy with the very heavy display when you click a whole cluster, but I guess you cannot have all at the same time.

A stacked column visualization shows how cluster structure and journal importance change over time. It does not display any inter-journal citation information, however. Reminiscent – on first sight – of history flow, or other stacked charts, we were, in fact, inspired by Sankey diagrams, and tried to map change starting from the clustering structure. The truth is, 95% of the data does not change much :) but data-wise, there is a distilled version of, for instance, the very interesting development neuroscience has taken over the years. It is documented already in a paper; we might want to publish a special interactive version for this story alone…

The treemap is probably the most functional of the visualizations. It transports the cluster structure quite well, additionally nicely tells the story how eigenfactor scores sum up to 1, and allows, on click, to get some pretty precise idea about the relation of the journal to others. It is a good counterpart to the radial visualization, with their complementary advantages.

Finally, a map. This is probably the most obvious approach for network visualization, but I couldn’t resist :) especially since it was a 1 day action. I used a spring embedding algorithm based on connection strength to calculate the map coordinates with cytoscape. I was quite pleased with its import, layout and export capabilities. I simply exported as gml, grep’ed the output and voila I had some coordinates to import. Great tool! The distortion lens is custom coded, and just uses two different linear distance scales. No fisheye, since I find these harder to control.

Here is the heart of the distance scaling code, if anyone is interested: var xx:Number = lens.x; var yy:Number = lens.y; var radius:Number = 50; var scale:Number = 5; var scaledRadius:Number = radius * scale; var dist:Number, diffX:Number, diffY:Number, diffUniX:Number, diffUniY:Number; for each (var n:Node in data.group("leaves")) { diffX = (n.props.x - xx); diffY = (n.props.y - yy); dist = Math.sqrt(Math.pow(diffX, 2) + Math.pow(diffY, 2)); if(dist < radius) { n.x = xx + scale * diffX; n.y = yy + scale * diffY; } else { diffUniX = diffX / dist; diffUniY = diffY / dist; n.x = xx + diffUniX * (scaledRadius + dist - radius); n.y = yy + diffUniY * (scaledRadius + dist - radius); } }

All visualizations implemented using flare, my favorite visualization framework. Amazing work, Mr Heer!

For all visualizations, the data basis came from the eigenfactor team, calculating both importance values for individual journals, as well as grouping them hierarchically according to their citation flow “neighborhoods”. You can find lots more information on the eigenfactor site.

The cooperation started when the eigenfactor team used a customized version of the good old relation browser; later, they got in touch with me to produce some more visualizations. We started work in summer 08, and after lots of scribbles, experiments, prototypes, adjustments and a little visit in Seattle autumn, we can finally present the results. I quite enjoyed the cooperation, science and design can be an explosive mix :) So, thanks to Carl Bergstrom, Martin Rosvall, Ben Althouse, and Jevin West.

Anyways — feedback welcome! Btw if you want to blog about it — here are some screenshots.