Extract a sample of words around a particular word using stringr in R

Question

I've seen a couple of similar questions posted on SO regarding this topic, but they seem to be worded improperly (example) or in a different language (example).

In my scenario, I consider everything that is surrounded by white space to be a word. Emoticons, numbers, strings of letters that aren't really words, I don't care. I just want to get some context around the string that was found without having to read the entire file to figure out if it's a valid match.

I tried using the following, but it takes awhile to run if you've got a long text file:

text <- "He served both as Attorney General and Lord Chancellor of England. After his death, he remained extremely influential through his works, especially as philosophical advocate and practitioner of the scientific method during the scientific revolution. Bacon has been called the father of empiricism.[6] His works argued for the possibility of scientific knowledge based only upon inductive and careful observation of events in nature. Most importantly, he argued this could be achieved by use of a skeptical and methodical approach whereby scientists aim to avoid misleading themselves. While his own practical ideas about such a method, the Baconian method, did not have a long lasting influence, the general idea of the importance and possibility of a skeptical methodology makes Bacon the father of scientific method. This marked a new turn in the rhetorical and theoretical framework for science, the practical details of which are still central in debates about science and methodology today. Bacon was knighted in 1603 and created Baron Verulam in 1618[4] and Viscount St. Alban in 1621;[3][b] as he died without heirs, both titles became extinct upon his death. Bacon died of pneumonia in 1626, with one account by John Aubrey stating he contracted the condition while studying the effects of freezing on the preservation of meat."

stringr::str_extract(text, "(.*?\\s){1,10}Verulam(\\s.*?){1,10}")

I'm assuming there is a much, much faster/more efficient way in which to do this, yes?

Answer 1

Try this:

stringr::str_extract(text, "([^\\s]+\\s){3}Verulam(\\s[^\\s]+){3}")
# alternately, if you like " " more than \\s:
# stringr::str_extract(text, "(?:[^ ]+ ){3}Verulam(?: [^ ]+){3}")

#[1] "and created Baron Verulam in 1618[4] and"

Change the number inside the {} to suit your needs.

You can use non-capture (?:) groups, too, though I'm not sure yet whether that will improve speed.

stringr::str_extract(text, "(?:[^\\s]+\\s){3}Verulam(?:\\s[^\\s]+){3}")

Answer 2

I'd use unlist(strsplit) and then index the resulting vector. You could make it a function so that the number of words to fetch pre and post is a flexible parameter:

getContext <- function(text, look_for, pre = 3, post=pre) {
  # create vector of words (anything separated by a space)
  t_vec <- unlist(strsplit(text, '\\s'))

  # find position of matches
  matches <- which(t_vec==look_for)

  # return words before & after if any matches
  if(length(matches) > 0) {
    out <- 
      list(before = ifelse(m-pre < 1, NA, 
                           sapply(matches, function(m) t_vec[(m - pre):(m - 1)])), 
           after = sapply(matches, function(m) t_vec[(m + 1):(m + post)]))

    return(out)
  } else {
    warning('No matches')
  }
}

Works for a single match

getContext(text, 'Verulam')

# $before
#      [,1]     
# [1,] "and"    
# [2,] "created"
# [3,] "Baron"  
# 
# $after
#      [,1]     
# [1,] "in"     
# [2,] "1618[4]"
# [3,] "and"   

Also works if there's more than one match

getContext(text, 'he')

# $before
#      [,1]     [,2]           [,3]          [,4]     
# [1,] "After"  "nature."      "in"          "John"   
# [2,] "his"    "Most"         "1621;[3][b]" "Aubrey" 
# [3,] "death," "importantly," "as"          "stating"
# 
# $after
#      [,1]          [,2]     [,3]      [,4]        
# [1,] "remained"    "argued" "died"    "contracted"
# [2,] "extremely"   "this"   "without" "the"       
# [3,] "influential" "could"  "heirs,"  "condition" 

getContext(text, 'fruitloops')
# Warning message:
#   In getContext(text, "fruitloops") : No matches