My implementation (https://github.com/brianquinlan/word-cloud-valentine) is a lot less sophisticated than stylecloud but I think that I had a few interesting ideas about text extraction.
I used nltk to extract only nouns and to do word stemming (e.g. so that "time", "times" and "timing" are only counted as one word).
I also experimented a lot with various method of determining word size i.e. size proportional to frequency, size proportional to log(frequency), size proportional to sqrt(frequency).
I concluded that they do have some utility as "semantic cartograms" in certain contexts, but beyond that they're mostly just decoration. Especially the ones created by this tool. =)
A while ago I tried to write some wordcloud code, but couldn't quite figure out how to do the layout. Does anyone know where one might find a good writeup of the algorithm that tends to be used for this?
It does not cover white-space with words, as Wordle et al will do. Also it's in German. But I guess the code itself is quite clear.
What could "word clouds pertaining to style" possibly mean? Maybe if you gave it a text and it spit out a cloud with things like: "using passive voice," "flowery," "strident," "long-winded," "plain."