How to write custom removePunctuation() function to better deal with Unicode chars?

Question

In the source code of the tm text-mining R-package, in file transform.R, there is the removePunctuation() function, currently defined as:

function(x, preserve_intra_word_dashes = FALSE)
{
    if (!preserve_intra_word_dashes)
        gsub("[[:punct:]]+", "", x)
    else {
        # Assume there are no ASCII 1 characters.
        x <- gsub("(\\w)-(\\w)", "\\1\1\\2", x)
        x <- gsub("[[:punct:]]+", "", x)
        gsub("\1", "-", x, fixed = TRUE)
    }
}

I need to parse and mine some abstracts from a science conference (fetched from their website as UTF-8). The abstracts contain some unicode characters that need to be removed, particularly at word boundaries. There are the usual ASCII punctuation characters, but also a few Unicode Dashes, Unicode Quotes, Math Symbols...

There are also URLs in the text, and there the punctuation the intra-word punctuation characters need to be preserved. tm's built-in removePunctuation() function is too radical.

So I need a custom removePunctuation() function to do removal according to my requirements.

My custom Unicode function looks like this now, but it does not work as expected. I am using R only rarely, so getting things done in R takes some time, even for the simplest tasks.

My function:

corpus <- tm_map(corpus, rmPunc =  function(x){ 
# lookbehinds 
# need to be careful to specify fixed-width conditions 
# so that it can be used in lookbehind

x <- gsub('(.*?)(?<=^[[:punct:]’“”:±</>]{5})([[:alnum:]])'," \\2", x, perl=TRUE) ;
x <- gsub('(.*?)(?<=^[[:punct:]’“”:±</>]{4})([[:alnum:]])'," \\2", x, perl=TRUE) ;
x <- gsub('(.*?)(?<=^[[:punct:]’“”:±</>]{3})([[:alnum:]])'," \\2", x, perl=TRUE) ;
x <- gsub('(.*?)(?<=^[[:punct:]’“”:±</>]{2})([[:alnum:]])'," \\2", x, perl=TRUE) ;
x <- gsub('(.*?)(?<=^[[:punct:]’“”:±</>])([[:alnum:]])'," \\2", x, perl=TRUE) ; 
# lookaheads (can use variable-width conditions) 
x <- gsub('(.*?)(?=[[:alnum:]])([[:punct:]’“”:±]+)$',"\1 ", x, perl=TRUE) ;

# remove all strings that consist *only* of punct chars 
gsub('^[[:punct:]’“”:±</>]+$',"", x, perl=TRUE) ;

}

It does not work as expected. I think, it doesn't do anything at all. The punctuation is still inside the terms-document matrix, see:

 head(Terms(tdm), n=30)

  [1] "<></>"                      "---"                       
  [3] "--,"                        ":</>"                      
  [5] ":()"                        "/)."                       
  [7] "/++"                        "/++,"                      
  [9] "..,"                        "..."                       
 [11] "...,"                       "..)"                       
 [13] "“”,"                        "(|)"                       
 [15] "(/)"                        "(.."                       
 [17] "(..,"                       "()=(|=)."                  
 [19] "(),"                        "()."                       
 [21] "(&)"                        "++,"                       
 [23] "(0°"                        "0.001),"                   
 [25] "0.003"                      "=0.005)"                   
 [27] "0.006"                      "=0.007)"                   
 [29] "000km"                      "0.01)" 
...

So my questions are:

1. Why doesn't the call to my function(){} have the desired effect? How can my function be improved?
2. Are Unicode regex pattern classes such as if \P{ASCII} or \P{PUNCT} supported in R's perl-compatible regular expressions? I think they aren't (by default) by PCRE:: " Only the support for various Unicode properties with \p is incomplete, though the most important ones are supported."

Answer 1

As much as I like Susana's answer it is breaking the Corpus in newer versions of tm (No longer a PlainTextDocument and destroying the meta)

You will get a list and the following error:

Error in UseMethod("meta", x) : 
no applicable method for 'meta' applied to an object of class "character"

Using

tm_map(your_corpus, PlainTextDocument)

will give you back your corpus but with broken $meta (in particular document ids will be missing.

Solution

Use content_transformer

toSpace <- content_transformer(function(x,pattern)
    gsub(pattern," ", x))
your_corpus <- tm_map(your_corpus,toSpace,"„")

Source: Hands-On Data Science with R, Text Mining, Graham.Williams@togaware.com http://onepager.togaware.com/

Update

This function removes everything that is not alpha numeric (i.e. UTF-8 emoticons etc.)

removeNonAlnum <- function(x){
  gsub("[^[:alnum:]^[:space:]]","",x)
}