gistfile1.txt

#' Title: Exploratory data analysis - Agha Shahid Ali
#' Author: Taylor Arnold (taylor.arnold@acm.org)
#' Created: 2015-07-29 18:37
#' Description: Basic exploratory data analysis of the
#'              poems of Agha Shahid Ali.

library(syuzhet)

# Read in the lines and clean some basic characters
poems <- scan("~/Desktop/CMl.txt", what="character", sep="\n")
poems <- gsub("\"", "", poems)
poems <- gsub("(", "", poems, fixed=TRUE)
poems <- gsub(")", "", poems, fixed=TRUE)

# For now, we'll just take the first 100 lines as the
# input has some encoding issues (to be fixed later)
poems <- poems[1:100]

# How positive or negative is each line?
out <- get_sentiment(poems)
table(out)
poems[out == -5]

# Now look at the 8 category sentiment types:
out <- get_nrc_sentiment(poems)
poems[out$anticipation > 0]
poems[out$fear > 0]
poems[out$anticipation > 0 & out$sadness > 0]

# A simple plot of the sentiment of each line
# (did not turn out to be very helpful, but
#  could be later)
plot(0,0,xlim=c(1,nrow(out)),ylim=c(0,max(out)),col="white")
for (i in 1:ncol(out)) {
  lines(1:nrow(out), out[,i], col=rainbow(8)[i], lwd=2)
}
legend(80,2.5,col=rainbow(8),colnames(out),pch=19,cex=1)


# An altenative method -
# Analysis using coreNLP; package needs to downloaded via:
#
#   install.packages("coreNLP")
#   coreNLP::downloadCoreNLP()

library(coreNLP)
coreNLP::initCoreNLP()

anno <- annotateString(poems)
tok <- getToken(anno)
ut <- coreNLP::universalTagset(tok$POS)

# Show the most frequence lemmas (normalized words) by
# part of speech.
sort(table(tok$lemm[ut == "NOUN"]),decreasing=TRUE)[1:24]
sort(table(tok$lemm[ut == "VERB"]),decreasing=TRUE)[1:24]
sort(table(tok$lemm[ut == "ADJ"]),decreasing=TRUE)[1:5]