R tm removeWords function not removing words

Question

I am trying to remove some words from a corpus I have built but it doesn't seem to be working. I first run through everything and create a dataframe that lists my words in order of their frequency. I use this list to identify words I am not interested in and then try to create a new list with the words removed. However, the words remain in my dataset. I am wondering what I am doing wrong and why the words aren't being removed? I have included the full code below:

install.packages("rvest")
install.packages("tm")
install.packages("SnowballC")
install.packages("stringr")
library(stringr) 
library(tm) 
library(SnowballC) 
library(rvest)

# Pull in the data I have been using. 
paperList <- html("http://journals.plos.org/plosone/search?q=nutrigenomics&sortOrder=RELEVANCE&filterJournals=PLoSONE&resultsPerPage=192")
paperURLs <- paperList %>%
  html_nodes(xpath="//*[@class='search-results-title']/a") %>%
  html_attr("href")
paperURLs <- paste("http://journals.plos.org", paperURLs, sep = "")
paper_html <- sapply(1:length(paperURLs), function(x) html(paperURLs[x]))

paperText <- sapply(1:length(paper_html), function(x) paper_html[[1]] %>%
                      html_nodes(xpath="//*[@class='article-content']") %>%
                      html_text() %>%
                      str_trim(.))
# Create corpus
paperCorp <- Corpus(VectorSource(paperText))
for(j in seq(paperCorp))
{
  paperCorp[[j]] <- gsub(":", " ", paperCorp[[j]])
  paperCorp[[j]] <- gsub("\n", " ", paperCorp[[j]])
  paperCorp[[j]] <- gsub("-", " ", paperCorp[[j]])
}

paperCorp <- tm_map(paperCorp, removePunctuation)
paperCorp <- tm_map(paperCorp, removeNumbers)

paperCorp <- tm_map(paperCorp, removeWords, stopwords("english"))

paperCorp <- tm_map(paperCorp, stemDocument)

paperCorp <- tm_map(paperCorp, stripWhitespace)
paperCorpPTD <- tm_map(paperCorp, PlainTextDocument)

dtm <- DocumentTermMatrix(paperCorpPTD)

termFreq <- colSums(as.matrix(dtm))
head(termFreq)

tf <- data.frame(term = names(termFreq), freq = termFreq)
tf <- tf[order(-tf[,2]),]
head(tf)

# After having identified words I am not interested in
# create new corpus with these words removed.
paperCorp1 <- tm_map(paperCorp, removeWords, c("also", "article", "Article", 
                                              "download", "google", "figure",
                                              "fig", "groups","Google", "however",
                                              "high", "human", "levels",
                                              "larger", "may", "number",
                                              "shown", "study", "studies", "this",
                                              "using", "two", "the", "Scholar",
                                              "pubmedncbi", "PubMedNCBI",
                                              "view", "View", "the", "biol",
                                              "via", "image", "doi", "one", 
                                              "analysis"))

paperCorp1 <- tm_map(paperCorp1, stripWhitespace)
paperCorpPTD1 <- tm_map(paperCorp1, PlainTextDocument)
dtm1 <- DocumentTermMatrix(paperCorpPTD1)
termFreq1 <- colSums(as.matrix(dtm1))
tf1 <- data.frame(term = names(termFreq1), freq = termFreq1)
tf1 <- tf1[order(-tf1[,2]),]
head(tf1, 100)

If you look through tf1 you will notice that plenty of the words that were specified to be removed have not actually been removed.

Just wondering what I am doing wrong, and how I might remove these words from my data?

NOTE: removeWords is doing something because the output from head(tm, 100) and head(tm1, 100) are not exactly the same. So removeWords seems to removing some instances of the words I am trying to get rid of, but not all instances.

Solution 1

I switched some code around and added tolower. The stopwords are all in lowercase, so you need to do that first before you remove stopwords.

paperCorp <- tm_map(paperCorp, removePunctuation)
paperCorp <- tm_map(paperCorp, removeNumbers)
# added tolower
paperCorp <- tm_map(paperCorp, tolower)
paperCorp <- tm_map(paperCorp, removeWords, stopwords("english"))
# moved stripWhitespace
paperCorp <- tm_map(paperCorp, stripWhitespace)

paperCorp <- tm_map(paperCorp, stemDocument)

Upper case words no longer needed, since we set everything to lower case. You can remove these.

paperCorp <- tm_map(paperCorp, removeWords, c("also", "article", "Article", 
                                               "download", "google", "figure",
                                               "fig", "groups","Google", "however",
                                               "high", "human", "levels",
                                               "larger", "may", "number",
                                               "shown", "study", "studies", "this",
                                               "using", "two", "the", "Scholar",
                                               "pubmedncbi", "PubMedNCBI",
                                               "view", "View", "the", "biol",
                                               "via", "image", "doi", "one", 
                                               "analysis"))

paperCorpPTD <- tm_map(paperCorp, PlainTextDocument)

dtm <- DocumentTermMatrix(paperCorpPTD)

termFreq <- colSums(as.matrix(dtm))
head(termFreq)

tf <- data.frame(term = names(termFreq), freq = termFreq)
tf <- tf[order(-tf[,2]),]
head(tf)

           term  freq
fatty     fatty 29568
pparα     ppara 23232
acids     acids 22848
gene       gene 15360
dietary dietary 12864
scholar scholar 11904

tf[tf$term == "study"]


data frame with 0 columns and 1659 rows

And as you can see, the outcome is that study is no longer in the corpus. The rest of the words are also gone

Solution 2

If someone gets error like me and above solution still doesn't work, try use: paperCorp <- tm_map(paperCorp, content_transformer(tolower)) instead of paperCorp <- tm_map(paperCorp, tolower) because tolower() is a function from base package and returns different structure (I mean changes something in the result type) so you can't use for example paperCorp[[j]]$content but only paperCorp[[j]]. It's just a digression, maybe halpful to someone.