Text Mining with the tm Library

library(tm)

• Loading in the data
• Cleaning text
• Making a document-term matrix
• Where you can go from there

• First make a source object
• Then a corpus object

reviews <- read.csv("reviews.csv", stringsAsFactors=FALSE)

review_source <- VectorSource(reviews$text)
corpus <- Corpus(review_source)

inspect(corpus)

[1] " SO ADDICTING  DEFF DOWNLAOD ITS EPIC YOU CAT LOVERS WILL FALL IN LOVE <3"

corpus <- tm_map(corpus, content_transformer(tolower))
inspect(corpus)

[1] " so addicting  deff downlaod its epic you cat lovers will fall in love <3"

corpus <- tm_map(corpus, removePunctuation)
inspect(corpus)

[1] " so addicting  deff downlaod its epic you cat lovers will fall in love 3"

corpus <- tm_map(corpus, stripWhitespace)
inspect(corpus)

[1] " so addicting deff downlaod its epic you cat lovers will fall in love 3"

corpus <- tm_map(corpus, removeWords, stopwords("english"))
inspect(corpus)

[1] "  addicting deff downlaod  epic  cat lovers will fall  love 3"

corpus <- tm_map(corpus, stemDocument)
inspect(corpus)

[1] "  addict deff downlaod  epic  cat lover will fall  love 3"

• Can do document-term matrix
• or term-document matrix

dtm <- DocumentTermMatrix(corpus)
tdm <- TermDocumentMatrix(corpus)

dim(dtm)

[1] 1000 2250

dtm_small <- removeSparseTerms(dtm, 0.99)
dim(dtm_small)

[1] 1000  205

frequency <- colSums(as.matrix(dtm))
frequency <- sort(frequency, decreasing=TRUE)
head(frequency)

 game  play great   fun  good  love 
  976   250   249   241   238   234 

model <- lm(reviews$rating ~ .,
            data = as.data.frame(as.matrix(dtm_small)))

by_country <- 
  reviews %>%
  group_by(location) %>%
  summarise(text = paste(text, collapse = ' '))

review_source <- VectorSource(by_country$text)
corpus <- Corpus(review_source)

• Read in data
• Clean text
• Make document term matrix
• Remove sparse terms
• Do anything else you like

Thanks for listening!

https://github.com/mhairi