polarity: Polarity Score (Sentiment Analysis)

Description

polarity - Approximate the sentiment (polarity) of text by grouping variable(s). polarity_frame - Generate a polarity lookup environment or data.frame for use with the polarity.frame argument in the polarity function.

Usage

polarity(text.var, grouping.var = NULL,
  polarity.frame = qdapDictionaries::env.pol,
  negators = qdapDictionaries::negation.words,
  amplifiers = qdapDictionaries::amplification.words,
  deamplifiers = qdapDictionaries::deamplification.words,
  question.weight = 0, amplifier.weight = 0.8, n.before = 4,
  n.after = 2, rm.incomplete = FALSE, digits = 3, ...)

polarity_frame(positives, negatives, pos.weights = 1, neg.weights = -1,
  envir = TRUE)

Arguments

text.var

The text variable.

grouping.var

The grouping variables. Default NULL generates one word list for all text. Also takes a single grouping variable or a list of 1 or more grouping variables.

polarity.frame

A dataframe or environment containing a dataframe of positive/negative words and weights.

negators

A character vector of terms reversing the intent of a positive or negative word.

amplifiers

A character vector of terms that increase the intensity of a positive or negative word.

deamplifiers

A character vector of terms that decrease the intensity of a positive or negative word.

question.weight

The weighting of questions (values from 0 to 1). Default 0 corresponds with the belief that questions (pure questions) are not polarized. A weight may be applied based on the evidence that the questions function with polarity.

amplifier.weight

The weight to apply to amplifiers/deamplifiers (values from 0 to 1). This value will multiply the polarized terms by 1 + this value.

n.before

The number of words to consider as valence shifters before the polarized word.

n.after

The number of words to consider as valence shifters after the polarized word.

rm.incomplete

logical. If TRUE text rows ending with qdap's incomplete sentence end mark (|) will be removed from the analysis.

digits

Integer; number of decimal places to round when printing.

...

Other arguments supplied to strip.

positives

A character vector of positive words.

negatives

A character vector of negative words.

pos.weights

A vector of weights to weight each positive word by. Length must be equal to length of postives or length 1 (if 1 weight will be recycled).

neg.weights

A vector of weights to weight each negative word by. Length must be equal to length of negatives or length 1 (if 1 weight will be recycled).

envir

logical. If TRUE a lookup table (a dataframe within an environment) is produced rather than a data.frame.

Value

Returns a list of:
all
A dataframe of scores per row with:
- group.var - the grouping variable
- wc - word count
- polarity - sentence polarity score
- pos.words - words considered positive
- neg.words - words considered negative
- text.var - the text variable
group
A dataframe with the average polarity score by grouping variable:
- group.var - the grouping variable
- total.sentences - Total sentences spoken.
- total.words - Total words used.
- ave.polarity - The sum of all polarity scores for that group divided by number of sentences spoken.
- sd.polarity - The standard deviation of that group's sentence level polarity scores.
- stan.mean.polarity - A standardized polarity score calculated by taking the average polarity score for a group divided by the standard deviation.
digitsinteger value od number of digits to display; mostly internal use

Details

The equation used by the algorithm to assign value to polarity of each sentence fist utilizes the sentiment dictionary (Hu and Liu, 2004) to tag polarized words. A context cluster ($x_i^{T}$) of words is pulled from around this polarized word (default 4 words before and two words after) to be considered as valence shifters. The words in this context cluster are tagged as neutral ($x_i^{0}$), negator ($x_i^{N}$), amplifier ($x_i^{a}$), or de-amplifier ($x_i^{d}$). Neutral words hold no value in the equation but do affect word count ($n$). Each polarized word is then weighted $w$ based on the weights from the polarity.frame argument and then further weighted by the number and position of the valence shifters directly surrounding the positive or negative word. The researcher may provide a weight $c$ to be utilized with amplifiers/de-amplifiers (default is .8; deamplifier weight is constrained to -1 lower bound). Last, these context cluster ($x_i^{T}$) are summed and divided by the square root of the word count ($\sqrt{n}$) yielding an unbounded polarity score ($\delta$). Note that context clusters containing a comma before the polarized word will only consider words found after the comma. $$\delta=\frac{x_i^T}{\sqrt{n}}$$ Where: $$x_i^T=\sum{((1 + c(x_i^{A} - x_i^{D}))\cdot w(-1)^{\sum{x_i^{N}}})}$$ $$x_i^{A}=\sum{(w_{neg}\cdot x_i^{a})}$$ $$x_{i}^D=\left{\begin{array}{cc} x_{i}^D & x_{i}^D \geq -1 \ -1 & x_{i}^D < -1 \end{array}\right.$$ $$x_i^{D}=\sum{(- w_{neg}\cdot x_i^{a} + x_i^{d})}$$ $$w_{neg}=\left{\begin{array}{cc} 1 & \sum{x_i^{N}} \bmod {2} >0 \ 0 & \sum{x_i^{N}} \bmod {2} =0 \end{array}\right.$$

References

Hu, M., & Liu, B. (2004). Mining opinion features in customer reviews. National Conference on Artificial Intelligence. http://www.slideshare.net/jeffreybreen/r-by-example-mining-twitter-for

Examples

Run this code

with(DATA, polarity(state, list(sex, adult)))
(poldat <- with(sentSplit(DATA, 4), polarity(state, person)))
counts(poldat)
scores(poldat)
plot(poldat)

poldat2 <- with(mraja1spl, polarity(dialogue,
    list(sex, fam.aff, died)))
colsplit2df(scores(poldat2))
plot(poldat2)
plot(scores(poldat2))

poldat3 <- with(rajSPLIT, polarity(dialogue, person))
poldat3[["group"]][, "OL"] <- outlier_labeler(scores(poldat3)[,
    "ave.polarity"])
poldat3[["all"]][, "OL"] <- outlier_labeler(counts(poldat3)[,
    "polarity"])
htruncdf(scores(poldat3), 10)
htruncdf(counts(poldat3), 15, 8)
plot(poldat3)
plot(poldat3, nrow=4)
qheat(scores(poldat3)[, -7], high="red", order.b="ave.polarity")

## Create researcher defined polarity.frame
POLENV <- polarity_frame(positive.words, negative.words)
POLENV
ls(POLENV)[1:20]

## ANIMATION
#===========
(deb2 <- with(subset(pres_debates2012, time=="time 2"),
    polarity(dialogue, person)))

bg_black <- Animate(deb2, neutral="white", current.speaker.color="grey70")
print(bg_black, pause=.75)

bgb <- vertex_apply(bg_black, label.color="grey80", size=20, color="grey40")
bgb <- edge_apply(bgb, label.color="yellow")
print(bgb, bg="black", pause=.75)

## Save it
library(animation)
library(igraph)
library(plotrix)

loc <- folder(animation_polarity)

## Set up the plotting function
oopt <- animation::ani.options(interval = 0.1)

FUN <- function() {
    Title <- "Animated Polarity: 2012 Presidential Debate 2"
    Legend <- c(-1.1, -1.25, -.2, -1.2)
    Legend.cex <- 1
    lapply(seq_along(bgb), function(i) {
        par(mar=c(2, 0, 1, 0), bg="black")
        set.seed(10)
        plot.igraph(bgb[[i]], edge.curved=TRUE)
        mtext(Title, side=3, col="white")
        color.legend(Legend[1], Legend[2], Legend[3], Legend[4],
              c("Negative", "Neutral", "Positive"), attributes(bgb)[["legend"]],
              cex = Legend.cex, col="white")
        animation::ani.pause()
    })
}

FUN()

## Detect OS
type <- if(.Platform$OS.type == "windows") shell else system

saveHTML(FUN(), autoplay = FALSE, loop = TRUE, verbose = FALSE,
    ani.height = 500, ani.width=500,
    outdir = file.path(loc, "new"), single.opts =
    "'controls': ['first', 'play', 'loop', 'speed'], 'delayMin': 0")

## Detect OS
type <- if(.Platform$OS.type == "windows") shell else system

saveHTML(FUN2(), autoplay = FALSE, loop = TRUE, verbose = FALSE,
    ani.height = 1000, ani.width=650,
    outdir = loc2, single.opts =
    "'controls': ['first', 'play', 'loop', 'speed'], 'delayMin': 0")

FUN2(TRUE)

#=====================#
## Complex Animation ##
#=====================#
library(animation)
library(grid)
library(gridBase)
library(qdap)
library(reports)
library(igraph)
library(plotrix)

deb2dat <- subset(pres_debates2012, time=="time 2")
deb2dat[, "person"] <- factor(deb2dat[, "person"])
(deb2 <- with(deb2dat, polarity(dialogue, person)))

## Set up the network version
bg_black <- Animate(deb2, neutral="white", current.speaker.color="grey70")
bgb <- vertex_apply(bg_black, label.color="grey80", size=30, label.size=22,
    color="grey40")
bgb <- edge_apply(bgb, label.color="yellow")

## Set up the bar version
deb2_bar <- Animate(deb2, as.network=FALSE)

## Generate a folder
loc2 <- folder(animation_polarity2)

## Set up the plotting function
oopt <- animation::ani.options(interval = 0.1)


FUN2 <- function(follow=FALSE, theseq = seq_along(bgb)) {

    Title <- "Animated Polarity: 2012 Presidential Debate 2"
    Legend <- c(.2, -1.075, 1.5, -1.005)
    Legend.cex <- 1

    lapply(theseq, function(i) {
        if (follow) {
            png(file=sprintf("%s/images/Rplot%s.png", loc2, i),
                width=650, height=725)
        }
        ## Set up the layout
        layout(matrix(c(rep(1, 9), rep(2, 4)), 13, 1, byrow = TRUE))

        ## Plot 1
        par(mar=c(2, 0, 2, 0), bg="black")
        #par(mar=c(2, 0, 2, 0))
        set.seed(20)
        plot.igraph(bgb[[i]], edge.curved=TRUE)
        mtext(Title, side=3, col="white")
        color.legend(Legend[1], Legend[2], Legend[3], Legend[4],
              c("Negative", "Neutral", "Positive"), attributes(bgb)[["legend"]],
              cex = Legend.cex, col="white")

        ## Plot2
        plot.new()
        vps <- baseViewports()

        uns <- unit(c(-1.3,.5,-.75,.25), "cm")
        p <- deb2_bar[[i]] +
            theme(plot.margin = uns,
                text=element_text(color="white"),
                plot.background = element_rect(fill = "black",
                    color="black"))
        print(p,vp = vpStack(vps$figure,vps$plot))
        animation::ani.pause()

        if (follow) {
            dev.off()
        }
    })

}

FUN2()

## Detect OS
type <- if(.Platform$OS.type == "windows") shell else system

saveHTML(FUN2(), autoplay = FALSE, loop = TRUE, verbose = FALSE,
    ani.height = 1000, ani.width=650,
    outdir = loc2, single.opts =
    "'controls': ['first', 'play', 'loop', 'speed'], 'delayMin': 0")

FUN2(TRUE)

Run the code above in your browser using DataLab