Skip navigation

NOTE: Parts [2], [3] & [4] are also now up.

Inspired by a post by @bfist who created the following word cloud in Ruby from VZ RISK INTSUM posts (visit the link or select the visualization to go to the post):

intsumwordcld-copy-2

I ♥ word clouds as much as anyone and usually run Presidential proclamations & SOTU addresses through a word cloud generator just to see what the current year’s foci are.

However, word clouds rarely provide what the creator of the visualization intends. Without performing more strict corpus analysis, one is really just getting a font-based frequency counter. While pretty, it’s not a good idea to derive too much meaning from a simple frequency count since there are phrase & sentence structure components to consider as well as concepts such as stemming (e.g. “risks” & “risk” are most likely the same thing, one is just plural…that’s a simplistic definition/example, though).

I really liked Jim Vallandingham’s Building a Bubble Cloud walk through on how he made a version of @mbostock’s NYTimes convention word counts and decided to both run a rudimentary stem on the VZ RISK INTSUM corpus along with a principal component analysis [PDF] to find the core text influencers and feed the output to a modified version of the bubble cloud:

Screenshot_3_6_13_10_55_PM

You can select the graphic to go to the “interactive”/larger version of it. I had intended to make selecting a circle bring up the relevant documents from the post corpus, but that will have to be a task for another day.

It’s noteworthy that both @bfist’s work and this modified version share many of the same core “important” words. With some stemming refinement and further stopword removal (e.g. “week” was in the original run of this visualization and is of no value for this risk-oriented visualization, so I made it part of the stopword list), this could be a really good way to get an overview of what the risky year was all about.

I won’t promise anything, but I’ll try to get the R code cleaned up enough to post. It’s really basic tm & PCA work, so no rocket-science degree is required. Fork @vlandham’s github repo & follow the aforelinked tutorial for the crunchy D3-goodness bits.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.