Chronowheel
Euan Adie
Tuesday, 03 June 2008 13:51 UTC
I’ve been experimenting with a new viz idea in Processing. I call it… the ChronoWheel. Unless that’s already trademarked by a clock manufacturer.
It’s for time series data.
Here the dataset is of key words automatically extracted from the Nature archives. It’s a set of vectors, one per word, describing the number of times that word appears in each year between 1970 and 2008.
Basically terms are represented as dots orbiting a central point. Popular terms are larger and further out from the centre. Each dot is anchored back to the core by a curved line, which came about accidentally (I’m crap at maths and mixed up Bezier curve variables) but looks quite cool, I think.
Here’s the MPEG of it in action. It’s just a proof of concept, rotates too quickly and doesn’t label enough terms but you get the idea.
Terms can rotate at different speeds and have different hues. At the moment they’re both randomized for a bit of visual interest but I’m thinking hue could reflect how quickly a term is growing / shrinking in popularity (is it hot / cold) and the speed of rotation could reflect how transient it is (does it appear throughout the archive, just in the 80s, just in two or three particular years?)
Right now there are 200 orbiting terms but there are something like 10k in the archive. I had to pick a random selection. Not sure how to keep things readable but include more data. Also not sure how to make the labels stand out more – maybe outline them somehow?
Updated 03 June 2008 13:55 UTC
-
Replies
-
Lovely — looks like a firework (“Catherine” wheel?) or a galaxy.
Would the labels stand out more if they were white?
-
Neat! – out of curiosity, why do the terms seem to disappear around 1990? If the length of the radius represents popularity, what does the location of the point on the circumference (the degree?) represent?
-
I’m not sure but I think it’s an artifact of how the 200 terms were picked (was oversimplifying when I said it was random, they had to meet minimum number of appearances criteria too).
The degree doesn’t correspond to anything at the moment but it could, good idea! Maybe related terms could be clustered together?
Will have a shot at the clearer labels, too.
-
Sweet, I like this a lot.
Using co-occurrence might be good, as then you can see as time progresses the terms which stay around for a period of time and then how they merge and split into new fields.
Maybe an item should move further out if it is around for longer, certainly that would make the labels more legible.
-
Nice work! It’s a pleasure to look at it, even though I have some trouble following the terms moving around on their orbits.
ThemeRiver is something that you might want to have a look at. It’s been developed to visualize theme changes over time.
Nils
-
Ah! Cool, I like ThemeRiver (hadn’t seen the paper before). I think Alf Eaton who is also here at NPG might have used it to plot the same dataset as above. Maybe we can get him to post the output here.
-
This was that:
– could do with picking more interesting terms though. There are some glitches that aren’t real, too, I think.
-
