adventures in text mining : "programming styles are like assholes: everybody's got one"
data source: http://lanic.utexas.edu/project/castro/db/
purpose: use lsa/lda/knn/svm/hierarchical clustering on various translations of same collection of texts into the same target language for purposes of identifying what George Lakoff calls frames
more on frames: http://truth-out.org/progressivepicks/item/12401-george-lakoff-progressives-need-to-use-language-that-reflects-moral-values
and a bit more: http://fora.tv/2007/11/07/Deceiving_Image_The_Science_of_Manipulation
finally, this gem: http://www.youtube.com/watch?v=f55ZsrXQ8Pg
TO DO:
- add comprehensive bibliography
- compose procedure pseudocode that would identify volatile spots between multiple instances of same text collection
- implement n-gram tokenization using RTextTools instead of tm, as it doesn't require RWeka's rJava dependency with its memory limits
- also use unique function in RTextTools package to obtain k for topic model #: http://www.rtexttools.com/1/post/2011/08/getting-started-with-latent-dirichlet-allocation-using-rtexttools-topicmodels.html
- add multilingual examples
"Fatigued" Bibliography:
Comprehensive Manifesto Project: https://manifesto-project.wzb.eu/
Drew Conway's excellent text mining in comparative politics work: http://drewconway.com/zia/2013/12/10/talk-at-kickstarter
Classic work on matrix math and pattern recognition: http://econ109.econ.bbk.ac.uk/brad/datamining/research_surveys/elden.pdf
SOUNDTRACK: http://www.leftbusinessobserver.com/Radio.html