pos: Parts of Speech Tagging

In qdap: Bridging the Gap Between Qualitative Data and Quantitative Analysis

Parts of Speech Tagging

Description

pos - Apply part of speech tagger to transcript(s).

pos_by - Apply part of speech tagger to transcript(s) by zero or more grouping variable(s).

pos_tags - Useful for interpreting the parts of speech tags created by pos and pos_by.

Usage

pos(
  text.var,
  parallel = FALSE,
  cores = detectCores()/2,
  progress.bar = TRUE,
  na.omit = FALSE,
  digits = 1,
  percent = TRUE,
  zero.replace = 0,
  gc.rate = 10
)

pos_by(
  text.var,
  grouping.var = NULL,
  digits = 1,
  percent = TRUE,
  zero.replace = 0,
  ...
)

pos_tags(type = "pretty")

Arguments

text.var	The text variable.
parallel	logical. If TRUE attempts to run the function on multiple cores. Note that this may not mean a speed boost if you have one core or if the data set is smaller as the cluster takes time to create.
cores	The number of cores to use if parallel = TRUE. Default is half the number of available cores.
progress.bar	logical. If TRUE attempts to provide a OS appropriate progress bar. If parallel is TRUE this argument is ignored. Note that setting this argument to TRUE may slow down the function.
na.omit	logical. If TRUE missing values (NA) will be omitted.
digits	Integer; number of decimal places to round when printing.
percent	logical. If TRUE output given as percent. If FALSE the output is proportion.
zero.replace	Value to replace 0 values with.
gc.rate	An integer value. This is a necessary argument because of a problem with the garbage collection in the openNLP function that pos wraps. Consider adjusting this argument upward if the error java.lang.OutOfMemoryError occurs.
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.
type	An optional character string giving the output of the pos tags. This must be one of the strings "pretty" (a left justified version of the output optimized for viewing but not good for export), "matrix" (a matrix version of the output), "dataframe"\ "df" (a dataframe version of the output), "all" (a list of all three of the previous output types).
...	Other argument supplied to pos.

Value

pos - returns a list of 4:

text	The original text
POStagged	The original words replaced with parts of speech in context.
POSprop	Dataframe of the proportion of parts of speech by row.
POSfreq	Dataframe of the frequency of parts of speech by row.
POSrnp	Dataframe of the frequency and proportions of parts of speech by row.
percent	The value of percent used for plotting purposes.
zero.replace	The value of zero.replace used for plotting purposes.

pos_by - returns a list of 6:

text	The original text
POStagged	The original words replaced with parts of speech in context.
POSprop	Dataframe of the proportion of parts of speech by row.
POSfreq	Dataframe of the frequency of parts of speech by row.
POSrnp	Dataframe of the frequency and proportions of parts of speech by row.
pos.by.prop	Dataframe of the proportion of parts of speech by grouping variable.
pos.by.freq	Dataframe of the frequency of parts of speech by grouping variable.
pos.by.rnp	Dataframe of the frequency and proportions of parts of speech by grouping variable.
percent	The value of percent used for plotting purposes.
zero.replace	The value of zero.replace used for plotting purposes.

Note

Note that contractions are treated as two words; for example the word count on "what's" is 2 for "what + is". This is not consistent with the word_count treatment of contractions but makes sense in a part of speech framework where a phrase such as "She's cool" is treated as a pronoun, verb and adjective respectively for "She + is + cool".

References

http:/opennlp.apache.org

See Also

Maxent_POS_Tag_Annotator, colcomb2class

Examples

## Not run: 
posdat <- pos(DATA$state)
ltruncdf(posdat, 7, 4)
## str(posdat)
names(posdat)
posdat$text           #original text

## Methods
preprocessed(posdat)  #words replaced with parts of speech
counts(posdat)        #frequency of parts of speech by row
proportions(posdat)   #proportion of parts of speech by row

## Methods Plotting
plot(preprocessed(posdat))
plot(counts(posdat))
plot(proportions(posdat))
plot(posdat)

out1 <- pos(DATA$state, parallel = TRUE) # not always useful
ltruncdf(out1, 7, 4)

#use pos_tags to interpret part of speech tags used by pos & pos_by
pos_tags()[1:10, ]
pos_tags("matrix")[1:10, ]
pos_tags("dataframe")[1:10, ]
pos_tags("df")[1:10, ]
ltruncdf(pos_tags("all"), 3)

posbydat <- with(DATA, pos_by(state, sex))
names(posbydat)

## Methods
scores(posbydat)   
preprocessed(posbydat)
counts(posbydat)     
proportions(posbydat)   

## Methods Plotting
plot(preprocessed(posbydat))
plot(counts(posbydat))
plot(proportions(posbydat))
plot(posbydat)

ltruncdf(posbydat, 7, 4)
truncdf(posbydat$pos.by.prop, 4)

POSby <- with(DATA, pos_by(state, list(adult, sex)))
plot(POSby, values = TRUE, digits = 2)
#or more quickly - reuse the output from before
out2 <- with(DATA, pos_by(posbydat, list(adult, sex)))

## Definite/Indefinite Noun 
## 2 approached compared...
## The later is more efficient but less accurate

## ------------------------##
## Part off speech tagging ##
## ------------------------##
pos_after <- function(text.var, words, pos){

    posses <- strsplit(as.character(text.var[["POStagged"]][["POStagged"]]), "\\s+")
    namespos <- lapply(posses, function(x) {
        y <- unlist(strsplit(x, "/"))
        setNames(y[c(TRUE, FALSE)], y[c(FALSE, TRUE)])
    })

    lapply(namespos, function(x, thewords = words, thepos = pos){
        locs <- which(x %in% thewords)
        locs <- locs[!is.na(locs)]

        if (identical(unclass(locs), integer(0))) return(NA_character_)

        nounlocs <- which(names(x) %in% thepos)

        unname(x[unique(sapply(locs, function(x){ 
            min(nounlocs[nounlocs - x > 0])
        }))])
    })  
}

out2 <- setNames(lapply(list(a=c("a", "an"), the="the"), function(x) {
    o <- pos_after(rajPOS, x, c("NN", "NNS", "NNP", "NNPS"))
    m <- stats::setNames(data.frame(sort(table(unlist(o))), 
        stringsAsFactors = FALSE), c("word", "freq"))
    m[m$freq> 3, ]
}), c("a", "the"))


dat2 <- setNames(Reduce(function(x, y) {
    merge(x, y, by = "word", all = TRUE)}, out2), c("Word", "A", "THE"))

dat2 <- reshape2::melt(dat2, id="Word", variable.name="Article", value.name="freq")

dat2 <- dat2[order(dat2$freq, dat2$Word), ]

ord2 <- aggregate(freq ~ Word, dat2, sum)

dat2$Word <- factor(dat2$Word, levels=ord2[order(ord2[[2]]), 1])
rownames(dat2) <- NULL
ggplot(dat2, aes(x=freq, y=Word)) +
    geom_point()+ facet_grid(~Article) +
    ggtitle("Part Of Speech Parsing Approach")

dev.new()

## --------------------##
## Regular Expressions ##
## --------------------##

library(qdapRegex);library(ggplot2);library(reshape2)

out <- setNames(lapply(c("@after_a", "@after_the"), function(x) {
    o <- rm_default(stringi:::stri_trans_tolower(raj$dialogue),
        pattern = x, extract=TRUE)
    m <- stats::setNames(data.frame(sort(table(unlist(o))), 
        stringsAsFactors = FALSE), c("word", "freq"))
    m[m$freq> 3, ]
}), c("a", "the"))

dat <- setNames(Reduce(function(x, y) {
    merge(x, y, by = "word", all = TRUE)}, out), c("Word", "A", "THE"))

dat <- reshape2::melt(dat, id="Word", variable.name="Article", value.name="freq")

dat <- dat[order(dat$freq, dat$Word), ]

ord <- aggregate(freq ~ Word, dat, sum)

dat$Word <- factor(dat$Word, levels=ord[order(ord[[2]]), 1])
rownames(dat) <- NULL
ggplot(dat, aes(x=freq, y=Word)) + 
    geom_point()+ facet_grid(~Article) + 
    ggtitle("Regex Approach")

## End(Not run)

qdap documentation built on May 31, 2023, 5:20 p.m.