word_cor: Find Correlated Words

Description

Find associated words within grouping variable(s).

Usage

word_cor(
  text.var,
  grouping.var = qdapTools::id(text.var),
  word,
  r = 0.7,
  values = TRUE,
  method = "pearson",
  ...
)

Value

Returns a vector of associated words or correlation matrix if r = NULL.

Arguments

text.var: The text variable (or frequency matrix).
grouping.var: The grouping variables. Default uses each row as a group. Also takes a single grouping variable or a list of 1 or more grouping variables. Unlike other qdap functions, this cannot be NULL.
word: The word(s) vector to find associated words for.
r: The correlation level find associated words for. If positive this is the minimum value, if negative this is the maximum value.
values: logical. If TRUE returns the named correlates (names are the words). If FALSE only the associated words are returned.
method: A character string indicating which correlation coefficient is to be computed ("pearson", "kendall", or "spearman").
...: Other arguments passed to wfm.

References

The plotting method for the list output was inspired by Ben Marwick; see https://stackoverflow.com/a/19925445/1000343 for more.

Examples

Run this code

if (FALSE) {
x <- factor(with(rajSPLIT, paste(act, pad(TOT(tot)), sep = "|")))
word_cor(rajSPLIT$dialogue, x, "romeo", .45)
word_cor(rajSPLIT$dialogue, x, "love", .5)  

## Negative correlation
word_cor(rajSPLIT$dialogue, x, "you", -.1)
with(rajSPLIT, word_cor(dialogue, list(person, act), "hate"))

words <- c("hate", "i", "love", "ghost")
with(rajSPLIT, word_cor(dialogue, x, words, r = .5))
with(rajSPLIT, word_cor(dialogue, x, words, r = .4))

## Set `r = NULL` to get matrix between words
with(rajSPLIT, word_cor(dialogue, x, words, r = NULL))

## Plotting 
library(tm)
data("crude")
oil_cor1 <- apply_as_df(crude, word_cor, word = "oil", r=.7)
plot(oil_cor1)

oil_cor2 <- apply_as_df(crude, word_cor, word = qcv(texas, oil, money), r=.7)
plot(oil_cor2)
plot(oil_cor2, ncol=2)

oil_cor3 <- apply_as_df(crude, word_cor, word = qcv(texas, oil, money), r=NULL)
plot(oil_cor3)

## Run on multiple times/person/nested
## Split and apply to data sets
## Suggested use of stemming
DATA3 <- split(DATA2, DATA2$person)

## Find correlations between words per turn of talk by person
## Throws multiple warning because small data set
library(qdapTools)
lapply(DATA3, function(x) {
    word_cor(x[, "state"], qdapTools::id(x), qcv(computer, i, no, good), r = NULL)
})

## Find words correlated per turn of talk by person
## Throws multiple warning because small data set
lapply(DATA3, function(x) {
    word_cor(x[, "state"], qdapTools::id(x), qcv(computer, i, no, good))
})


## A real example
dat <- pres_debates2012 
dat$TOT <- factor(with(dat, paste(time, pad(TOT(tot)), sep = "|")))
dat <- dat[dat$person %in% qcv(OBAMA, ROMNEY), ]
dat$person <- factor(dat$person)
dat.split <- with(dat, split(dat, list(person, time)))

wrds <- qcv(america, debt, dollar, people, tax, health)
lapply(dat.split, function(x) {
    word_cor(x[, "dialogue"], x[, "TOT"], wrds, r=NULL)
})

## Supply a matrix (make sure to use `t` on a `wfm` matrix)
worlis <- list(
    pronouns = c("you", "it", "it's", "we", "i'm", "i"),
    negative = qcv(no, dumb, distrust, not, stinks),
    literacy = qcv(computer, talking, telling)
)
y <- wfdf(DATA$state, qdapTools::id(DATA, prefix = TRUE))
z <- wfm_combine(y, worlis)

out <- word_cor(t(z), word = c(names(worlis), "else.words"), r = NULL)
out
plot(out)

## Additional plotting/viewing
require(tm)
data("crude")

out1 <- word_cor(t(as.wfm(crude)), word = "oil", r=.7)
vect2df(out1[[1]], "word", "cor")

plot(out1)
qheat(vect2df(out1[[1]], "word", "cor"), values=TRUE, high="red", 
    digits=2, order.by ="cor", plot=FALSE) + coord_flip()


out2 <- word_cor(t(as.wfm(crude)), word = c("oil", "country"), r=.7)
plot(out2)
}