test-text-bbc.txt

Função collocations: collocations.r

Help da função: help-collocations.txt

collocations              package:none               R Documentation

Extract collocations for a target word from a given raw text.

Description:

  collocations receives a text and a target word and select the 
  sentences from the text which contain the target word. From 
  those sentences, the co-occurrences between target word and the 
  other words which are above a certain threshold will constitue 
  the set of collocations.

Usage:

  collocates(thetext, targetword, ncollmax)

Arguments:

  thetext		character. Text given by the user in 
  .txt format and UTF-8 encoding. 

  targetword		character. Any word the user has chosen 
from the text. It will the reference for the extraction of the 
collocations.  

  ncollmax		numeric. Maximum number of collocates 
  to be displayed on the graph generated by the function. In case 
  the number of extracted collocates is less than the stipulated 
  maximum, then ncollmax will be ignored.

Details:

  The function may not work well depending on the size of 
  the text file given even though some optimizations were tried 
  such as using environments hash to count faster the words' 
  occurrences. 

Value:

  Instead of returning values, collocates generates one text 
  file and another file for a barplot in png format. Both are 
  saved in the workspace being used to run the function.
	
Warning:

  Depending on the size of the text file, the function may 
  get too slow or not work. As a suggestion, the usar can exeriment 
  the function with different text sizes. See Examples for a simple 
  teste of the function. 

Author:

  Viviane Santos da Silva
  
  viviane.sds90@gmail.com
  viviane.santos.silva@usp.br

References:

  http://en.wikibooks.org/wiki/R_Programming/Text_Processing Last 
access in may 18th 2014.

  About environments and hash argument: 
http://adv-r.had.co.nz/Environments.html (There has been created a hash 
function to optimize the use of hashes, but it only works for later 
versions of R. Read "See Also")

  Download of non-annotaded corpora for testing the function: 
http://corpora.informatik.uni-leipzig.de/download.html Last access 
in may 15th 2014.
  
  To understand a little bit more about collocations in a more 
intuitive way: http://esl.fis.edu/grammar/easy/colloc.htm

See Also:

  For more information on hash usage in R, see: 
http://cran.r-project.org/web/packages/hash/index.html, 
http://cran.r-project.org/web/packages/hash/hash.pdf and 
http://opendatagroup.wordpress.com/2009/07/26/hash-package-for-r/.

Examples:

  # Download the file "teste-texto-bbc.txt" in (http://ecologia.ib.usp.br/bie5782/doku.php?id=bie5782:01_curso_atual:alunos:trabalho_final:
viviane.santos.silva:start) and save it to your R workspace to run this example.
  
  collocates(thetext="test-text-bbc.txt", targetword="fiction", ncollmax=10) 
# generates a barplot for the 10 first collocates which co-occur 
with the target word "film" in the text given.

##### FUNCTION TO EXTRACT COLLOCATIONS FROM RAW TEXTS #####

collocations <- function(thetext, targetword, ncollmax) {

  
## Reading the text ##

text <- scan(thetext, character(0), quote = NULL, sep = "\n", allowEscapes = FALSE, strip.white = TRUE, fileEncoding = "UTF-8") # reads text input in .txt format  
textstring <- tolower(paste(text, collapse = "")) # converts upper-case characters to lower-case
# textstring

textclean1 <- gsub("[^[:alnum:][:space:]'’\\.]", "", textstring) # removes from the textstring all ist characters, except alphabetic, spaces, apostrophes and period 
textclean2 <- gsub("[[:space:]]+", ' ', textclean1) # substitutes multiple spaces that may accidentally appear in the text for simple space character
textclean3 <- gsub("\\. ", '.', textclean2) # substitute "period+space" for period only (to avoid future problems with the extraction of words from the begnings of the paragraphs)

textdotsplit <- strsplit(textclean3, split='\\.') # separate the text by its sentences using the periods to guide this process 

txtw <- list() # creating an empty list
# iterating to split the text to make it possible to access its words individually
for (i in 1:length(textdotsplit[[1]])){
  txtw[i] <- strsplit(textdotsplit[[1]][i], split=' ') # txtw is now a list of lists, each inner list corresponding to a sentence with indexes to access its words
}


## Creating a dictionary to save word occurrences ##

wordcounter <- function(sentencelist){ # function to transform a list of list such as txtw into a dictionary containing words counts
  wordcount <- new.env(hash=TRUE, parent = emptyenv()) # initializing the dictionary 
  
  for (i in seq(1:length(sentencelist))){ 
    sentence <- sentencelist[[i]] # this was made to make the code easier to read, avoiding excessive '[' notation 
    for (j in seq(1:length(sentence))){
      eachword <- sentence[j]  # 'eachword' is accessing each of the words from the text given 
      if (is.null(wordcount[[eachword]])) { # dictionary's new entry 
        wordcount[[eachword]] <- list(value=1)
      } else{wordcount[[eachword]]$value <- wordcount[[eachword]]$value + 1 # updates existing entry
      }
    }
  }
  return(wordcount)
}


## Creating a named list from a dictionary ##

ordnamedlist <- function(dict){ # tranform dictionary into list and sort it by its values
  
  occlist <- list() # creats an empty list
  i <- 1 
  for (w in ls(dict)) { # transform dictionary into a list
    occlist[[i]] <- list(name=w, value=dict[[w]]$value) # extracts words and their corresponding number of occurrences from the environment hash
    i <- i+1
  }
  
  occlist[order(as.numeric(sapply(occlist, "[", "value")), decreasing=T)] # applies the '[' function through the sapply to the named list created
# and extract its values, which will be used by the as.numeric to generate a simple numeric vector which, in turn, will be sorted by the 
# order function to ordenate the values returning their positions that will be "given" back to occlist and will automatically order it
}

wcount <- wordcounter(txtw) # calling the function that turns a list of lists into a dictionary 
wcountsort <- ordnamedlist(wcount) # transforming the wcount (dictionary) into an ordered list


## Creating a file containing the words and their counts ##

namefile <- "wordscounts.txt" # just to make it easier to change the file name if one prefer

write("words\tcounts", file=namefile) # generates a file containing 2 column-headers: words and counts. avoids problems if the user call this function more than once (function countsfile right below calls a write function which is using append=T)

print("A file named 'wordscounts.txt', containing your text words and their counts has been generated. Feel free to explore it (:") # so the user will know a file has been generated

countsfile <- function(l){ # function to be used by the lapply to print properly in the file words and their counts
  write(paste(c(l$name, l$value), collapse="\t"), file=namefile, append=T)
}  

trash <- lapply(wcountsort, countsfile) # calling the function to generate a file. the variable 'trash' is being attributed the lapply output for this was generating NULL values and, apparently, R has nothing like a procedure function (a function that returns nothing).


## Finding the co-occurrences ##

hastarget <- function(wlist, target){ # checks if the lists (sentences) in wlist has the target word
  target %in% wlist # the test is made using the logical function %in%
}

sbin <- which(sapply(txtw, hastarget, targetword)) # function 'which' returns the positions of the lists from txtw tested by the hastarget function and that were marked "TRUE"
tsentences <- txtw[sbin] # tsentences receives only those sentences containing the target word

cooccurrences <- wordcounter(tsentences) # creates a dicitionary with the counts of the words from the sentences which contains the target word
occlistsort <- ordnamedlist(cooccurrences) # sorts the created dictionary of co-occurrences


## Calculating cooccurrences words frequencies ##

wtotal <- sum(sapply(txtw, length)) # total number of words in the text
stotal <- sum(sapply(tsentences, length)) # total number of words in the subtext (sentences which contained the target word)


## Creating a dictionary for the frequencies of the co-occurrences ##

wfreqs <- function(countlist, total){ # countlist is a named list containing names and associated values and total is the amount of words in the portion of text from which this countlist was generated
  freqs  <- new.env(hash=TRUE, parent = emptyenv())
  
  for (i in seq(1:length(countlist))){
    eachword <- countlist[[i]]$name 
    freqs[[eachword]]$value <- (countlist[[i]]$value/total) # calculating the frequencies of the words
  }
  return(freqs)
}

tfreqs <- wfreqs(wcountsort, wtotal) # calling the functions that creates the frequency dictionary for the whole text
cfreqs <- wfreqs(occlistsort, stotal) # calling the functions that creates the frequency dictionary for the sentences of the text that contains the target


## Creating a dictionary for the frequencies of the relevant words ##

testocc <- function(ftotal, fsent){ 

  ratio  <- new.env(hash=TRUE, parent = emptyenv())
  
  for (w in ls(fsent)){
     if(tfreqs[[w]] > 0.0001 & wcount[[w]]$value > 5){ # discarding words with frequencies in the text below 0.01% or, for a text with less than 5000 words, discarding those words with freqs less than 2
      ratio[[w]]$value <- fsent[[w]]$value/ftotal[[w]]$value # this important parameter will be used to judge whether or not a word w is forming a collocation with the target. if this value is 0.5, it means there is no diference between the distribution of the word w considered alone and its distribution given the target (and that's the null model). the max of this ratio will be 'wtotal/stotal'. the minimum will be 1 (when every occurrence of w meets every occurrence of the target)
     }
  }  
  return(ratio)
}

relevfreqs <- testocc(tfreqs, cfreqs) # dictionary containing words and their ratios
relevfreqslist <- ordnamedlist(relevfreqs) # ordered list od the dictionary created


## Extracting collocates ##

threshold <- 2 # only words which appear twice as often will be considered collocates candidates

selectcoll <- function(l){ # will be used to select the collocations comparing the candidate words' ratios and the chosen threshold
  l$value > threshold
}  

collindex <- which(sapply(relevfreqslist, selectcoll)) # selecting the collocates' indexes
collocates <- relevfreqslist[collindex] # using the extracted indexes, collocates access the correct collocates from the relevfreqlist (list of relevant words and their frequencies)

n <- as.numeric(ncollmax)
if(length(collocates) > n){ # test to check if there are more collocates than the number the user wants to display on the graph
  collocates <- collocates[1:n]
}

if(length(collocates) == 0) {
  print(paste("Couldn't find any collocate for the '", targetword,"' chosen. Maybe you can try a more frequent word."))
  return(invisible())
} 

## Plotting a barplot displaying the n most probable collocates ## 

collnames <- sapply(collocates, function(l) l$name) # generating a vector containing the words which were selected as being collocates
collratios <- sapply(collocates, function(l) as.numeric(l$value)) # collocates ratio (or the degree of the collocations extracted)

png("barplot-collocates.png")
barplot(collratios, names.arg = collnames, las=2, main=paste("Collocates of", targetword), ylab="Collocations ratios", ylim=c(0, max(collratios)+0.5)) # graph dysplaying the results of the function
dev.off()

}