polmineR (version 0.7.4)

features,partition-method: Get features by comparison.

Description

The features of two objects, usually a partition defining a corpus of interest, and a partition defining a reference corpus are compared. The most important purpose is term extraction.

#' @rdname features-method setMethod("features", "cooccurrences", function(x, y, included = FALSE, method = "ll", mc = TRUE, verbose = TRUE) newObject <- new( 'compCooccurrences', encoding = x@encoding, included = included, corpus = x@corpus, sizeCoi = x@partitionSize, sizeRef = ifelse(included == FALSE, y@partitionSize, y@partitionSize - x@partitionSize), stat = data.table() ) if (identical(x@pAttribute, y@pAttribute) == FALSE) warning("BEWARE: cooccurrences objects are not based on the same pAttribute!") else newObject@pAttribute <- x@pAttribute

if (verbose == TRUE) message("... preparing tabs for matching") keys <- unlist(lapply(c("a", "b"), function(ab) paste(ab, x@pAttribute, sep="_"))) setkeyv(x@stat, keys) setkeyv(y@stat, keys) MATCH <- y@stat[x@stat]

# remove columns not needed colsToDrop <- c( "ll", "i.ll", "exp_window", "i.exp_window", "rank_ll", "i.rank_ll", "size_window", "i.size_window", "count_a", "i.count_a", "count_b", "i.count_b", "exp_partition", "i.exp_partition" ) for (drop in colsToDrop) MATCH[, eval(drop) := NULL, with=TRUE] setnames(MATCH, old=c("count_ab", "i.count_ab"), new=c("count_ref", "count_coi"))

if (included == TRUE) MATCH[, "count_ref" := MATCH[["count_ref"]] - MATCH[["count_coi"]] ]

newObject@stat <- MATCH for (how in method) if (verbose == TRUE) message("... statistical test: ", how) newObject <- do.call(how, args = list(.Object = newObject))

newObject )

Usage

# S4 method for partition
features(x, y, included = FALSE, method = "chisquare",
  verbose = FALSE)
# S4 method for partitionBundle
features(x, y, included = FALSE,
  method = "chisquare", verbose = TRUE, mc = getOption("polmineR.mc"),
  progress = FALSE)
# S4 method for ngrams
features(x, y, included = FALSE, method = "chisquare",
  verbose = TRUE, ...)

Arguments

x

a partition or partitionBundle object

y

a partition object, it is assumed that the coi is a subcorpus of ref

included

TRUE if coi is part of ref, defaults to FALSE

method

the statistical test to apply (chisquare or log likelihood)

verbose

logical, defaults to TRUE

mc

logical, whether to use multicore

progress

logical

...

further parameters

Value

The function returns a data frame with the following structure: - absolute frequencies in the first row - ...

References

Manning / Schuetze ...

Examples

# NOT RUN {
  use("polmineR.sampleCorpus")
  kauder <- partition("PLPRBTTXT", text_name="Volker Kauder", pAttribute="word")
  all <- partition("PLPRBTTXT", text_date=".*", regex=TRUE, pAttribute="word")
  terms_kauder <- features(x = kauder, y = all, included = TRUE)
  top100 <- subset(terms_kauder, rank_chisquare <= 100)
# }
# NOT RUN {
  use("polmineR.sampleCorpus")
  byName <- partitionBundle("PLPRBTTXT", sAttribute="text_name")
  byName <- enrich(byName, pAttribute="word")
  all <- partition("PLPRBTTXT", text_date=".*", regex=TRUE, pAttribute="word")
  result <- features(byName, all, included=TRUE, progress=TRUE)
  dtm <- as.DocumentTermMatrix(result, col="chisquare")
# }