How to search for specific n-grams in a corpus using R

Question

I'm looking for specific n-grams in a corpus. Let's say I want to find 'asset management' and 'historical yield' in a collection of documents.

This is how I loaded the corpus

my_corpus <- VCorpus(DirSource(directory, pattern = ".pdf"), 
                 readerControl = list(reader = readPDF)

I cleaned the corpus and did some basic calculations using document term matrices. Now I want to look for particular expressions and put them in a dataframe. This is what I use (thanks to phiver):

ngrams <- c('asset management', 'historical yield')
dtm_ngrams <- DocumentTermMatrix(my_corpus, control = list(dictionary = ngrams))
df_ngrams <- data.frame(Docs = dtm$dimnames$Docs, as.matrix(dtm_ngrams), row.names = NULL )

This code runs, but the result is 0 for both n-grams. So, I'm guessing the problem is that the library is not defined correctly because R doesn't pick up the space between the words. So far, I tried to put '' between the words, or [:space:] and some other solutions.

score 1 · Accepted Answer · answered Aug 02 '18 at 10:08

A document term matrix without any further manipulation contains only single words (and words of nchar 3 or more). If you want to have bigrams, you need to create a term matrix of bigrams (or uni and bigrams).

Based on your example and using just tm and NLP which is loaded as soon as you call tm, we can make a bigram tokenizer. Or multi-gram, see comment in code.

Using the built in crude data set again.

library(tm)

data("crude")
crude <- as.VCorpus(crude)
crude <- tm_map(crude, content_transformer(tolower))

# This tokenizer is built on NLP and creates bigrams. 
# If you want multi-grams specify 1:2 for uni- and bi-gram, 
# 2:3 for bi- and trigram, 1:3 for uni-, bi- and tri-grams.
# etc. etc. ...(ngrams(words(x), 1:3)...

bigram_tokenizer <- function(x) {
  unlist(lapply(ngrams(words(x), 2), paste, collapse = " "), use.names = FALSE)
}

my_words <- c("crude oil", "west texas")

dtm <- DocumentTermMatrix(crude, control=list(tokenizer = bigram_tokenizer, dictionary = my_words))

inspect(dtm)
<<DocumentTermMatrix (documents: 20, terms: 2)>>
Non-/sparse entries: 11/29
Sparsity           : 72%
Maximal term length: 10
Weighting          : term frequency (tf)
Sample             :
     Terms
Docs  crude oil west texas
  127         2          1
  144         0          0
  191         2          0
  194         1          2
  211         0          0
  273         2          0
  349         1          0
  353         1          0
  543         1          1
  708         1          0

After this you can put the dtm into a data.frame again as mentioned in your previous question

How to search for specific n-grams in a corpus using R

1 Answers1

Linked