all_cooccurrences: Get all cooccurrences in corpus/partition.

In polmineR: Verbs and Nouns for Corpus Analysis

Get all cooccurrences in corpus/partition.

Description

Obtain all cooccurrences in a corpus, or a partition. The result is a Cooccurrences-class object which includes a data.table with counts of cooccurrences. See the documentation entry for the Cooccurrences-class for methods to process Cooccurrences-class objects.

Usage

## S4 method for signature 'corpus'
Cooccurrences(
  .Object,
  p_attribute,
  left,
  right,
  stoplist = NULL,
  mc = getOption("polmineR.mc"),
  verbose = FALSE,
  progress = FALSE
)

## S4 method for signature 'character'
Cooccurrences(
  .Object,
  p_attribute,
  left,
  right,
  stoplist = NULL,
  mc = getOption("polmineR.mc"),
  verbose = FALSE,
  progress = FALSE
)

## S4 method for signature 'slice'
Cooccurrences(
  .Object,
  p_attribute,
  left,
  right,
  stoplist = NULL,
  mc = getOption("polmineR.mc"),
  verbose = FALSE,
  progress = FALSE
)

## S4 method for signature 'partition'
Cooccurrences(
  .Object,
  p_attribute,
  left,
  right,
  stoplist = NULL,
  mc = getOption("polmineR.mc"),
  verbose = FALSE,
  progress = FALSE
)

## S4 method for signature 'subcorpus'
Cooccurrences(
  .Object,
  p_attribute,
  left,
  right,
  stoplist = NULL,
  mc = getOption("polmineR.mc"),
  verbose = FALSE,
  progress = FALSE
)

Arguments

.Object	A length-one character vector indicating a corpus, or a partition object.
p_attribute	Positional attributes to evaluate.
left	A scalar integer value, size of left context.
right	A scalar integer value, size of right context.
stoplist	Tokens to exclude from the analysis.
mc	Logical value, whether to use multiple cores.
verbose	Logical value, whether to output messages.
progress	Logical value, whether to display a progress bar.

Details

The implementation uses a data.table to store information and makes heavy use of the reference logic of the data.table package, to avoid copying potentially large objects, and to be parsimonious with limited memory. The behaviour resulting from in-place changes may be uncommon, see examples.

See Also

To learn about methods available for the object that is returned, see the documentation of the Cooccurrences-class. See the cooccurrences-method (starting with a lower case c) to get the cooccurrences for the match for a query, which may also be a CQP query.

Examples

## Not run: 
# In a first scenario, we get all cooccurrences for the REUTERS corpus,
# excluding stopwords

stopwords <- unname(unlist(
  noise(
    terms("REUTERS", p_attribute = "word"),
    stopwordsLanguage = "en"
    )
  ))
r <- Cooccurrences(
  .Object = "REUTERS", p_attribute = "word",
  left = 5L, right = 5L, stoplist = stopwords
)
ll(r) # note that the table in the stat slot is augmented in-place
decode(r) # in-place modification, again
r <- subset(r, ll > 11.83 & ab_count >= 5)
data.table::setorderv(r@stat, cols = "ll", order = -1L)
head(r, 25)

if (requireNamespace("igraph", quietly = TRUE)){
  r@partition <- enrich(r@partition, p_attribute = "word")
  g <- as_igraph(r, as.undirected = TRUE)
  plot(g)
}

# The next scenario is a cross-check that extracting cooccurrences from
# from a Cooccurrences-class object with all cooccurrences and the result
# for getting cooccurrences for a single object are identical

a <- cooccurrences(r, query = "oil")
a <- data.table::as.data.table(a)

b <- cooccurrences("REUTERS", query = "oil", left = 5, right = 5, p_attribute = "word")
b <- data.table::as.data.table(b)
b <- b[!word %in% stopwords]

all(b[["word"]][1:5] == a[["word"]][1:5]) # needs to be identical!


stopwords <- unlist(noise(
  terms("GERMAPARLMINI", p_attribute = "word"),
  stopwordsLanguage = "german"
  )
)

# We now filter cooccurrences by keeping only the statistically 
# significant cooccurrens, identified by comparison with cooccurrences
# derived from a reference corpus

plpr_partition <- partition(
  "GERMAPARLMINI", date = "2009-11-10", interjection = "speech",
  p_attribute = "word"
)
plpr_cooc <- Cooccurrences(
  plpr_partition, p_attribute = "word",
  left = 3L, right = 3L,
  stoplist = stopwords,
  verbose = TRUE
)
decode(plpr_cooc)
ll(plpr_cooc)

merkel <- partition(
  "GERMAPARLMINI", speaker = "Merkel", date = "2009-11-10", interjection = "speech",
  regex = TRUE,
  p_attribute = "word"
)
merkel_cooc <- Cooccurrences(
  merkel, p_attribute = "word",
  left = 3L, right = 3L,
  stoplist = stopwords, 
  verbose = TRUE
)
decode(merkel_cooc)
ll(merkel_cooc)

merkel_min <- subset(
  merkel_cooc,
  by = subset(features(merkel_cooc, plpr_cooc), rank_ll <= 50)
  )
  
# Esentially the same procedure as in the previous example, but with 
# two positional attributes, so that part-of-speech annotation is 
# used for additional filtering.
   
         
protocol <- partition(
  "GERMAPARLMINI",
  date = "2009-11-10",
  p_attribute = c("word", "pos"),
  interjection = "speech"
)
protocol_cooc <- Cooccurrences(
  protocol,
  p_attribute = c("word", "pos"),
  left = 3L, right = 3L
  )
ll(protocol_cooc)
decode(protocol_cooc)

merkel <- partition(
  "GERMAPARLMINI",
  speaker = "Merkel",
  date = "2009-11-10",
  interjection = "speech",
  regex = TRUE,
  p_attribute = c("word", "pos")
)
merkel_cooc <- Cooccurrences(
  merkel,
  p_attribute = c("word", "pos"),
  left = 3L, right = 3L,
  verbose = TRUE
)
ll(merkel_cooc)
decode(merkel_cooc)

f <- features(merkel_cooc, protocol_cooc)
f <- subset(f, a_pos %in% c("NN", "ADJA"))
f <- subset(f, b_pos %in% c("NN", "ADJA"))
f <- subset(f, c(rep(TRUE, times = 50), rep(FALSE, times = nrow(f) - 50)))

merkel_min <- subset(merkel_cooc, by = f)

if (requireNamespace("igraph", quietly = TRUE)){
  g <- as_igraph(merkel_min, as.undirected = TRUE)
  plot(g)
}


## End(Not run)

polmineR documentation built on Nov. 2, 2023, 5:52 p.m.