1.7 Text Processing and Regular Expressions

The learning objectives for this section are to:

  • Transform non-tidy data into tidy data
  • Manipulate and transform a variety of data types, including dates, times, and text data

Most common types of data are encoded in text, even if that text is representing numerical values, so being able to manipulate text as a software developer is essential. R provides several built-in tools for manipulating text, and there is a rich ecosystem of packages for R for text based analysis. First let’s concentrate on some basic text manipulation functions.

1.7.1 Text Manipulation Functions in R

Text in R is represented as a string object, which looks like a phrase surrounded by quotation marks in the R console. For example "Hello!" and 'Strings are fun!' are both strings. You can tell whether an object is a string using the is.character() function. Strings are also known as characters in R.

You can combine several strings using the paste() function:

paste("Square", "Circle", "Triangle")
[1] "Square Circle Triangle"

By default the paste() function inserts a space between each word. You can insert a different string between each word by specifying the sep argument:

paste("Square", "Circle", "Triangle", sep = "+")
[1] "Square+Circle+Triangle"

A shortcut for combining all of the string arguments without any characters in between each of them is to use the paste0() function:

paste0("Square", "Circle", "Triangle")
[1] "SquareCircleTriangle"

You can also provide a vector of strings as an argument to paste(). For example:

shapes <- c("Square", "Circle", "Triangle")
paste("My favorite shape is a", shapes)
[1] "My favorite shape is a Square"   "My favorite shape is a Circle"  
[3] "My favorite shape is a Triangle"
two_cities <- c("best", "worst")
paste("It was the", two_cities, "of times.")
[1] "It was the best of times."  "It was the worst of times."

As you can see, all of the possible string combinations are produced when you provide a vector of strings as an argument to paste(). You can also collapse all of the elements of a vector of strings into a single string by specifying the collapse argument:

paste(shapes, collapse = " ")
[1] "Square Circle Triangle"

Besides pasting strings together, there are a few other basic string manipulation functions you should be made aware of. The nchar() function counts the number of characters in a string:

nchar("Supercalifragilisticexpialidocious")
[1] 34

The toupper() and tolower() functions make strings all uppercase or lowercase respectively:

cases <- c("CAPS", "low", "Title")
tolower(cases)
[1] "caps"  "low"   "title"
toupper(cases)
[1] "CAPS"  "LOW"   "TITLE"

1.7.2 Regular Expressions

Now that we’ve covered the basics of string manipulation in R, let’s discuss the more advanced topic of regular expressions. A regular expression is a string that defines a pattern that could be contained within another string. A regular expression can be used for searching for a string, searching within a string, or replacing one part of a string with another string. In this section I might refer to a regular expression as a regex, just know that they’re the same thing.

Regular expressions use characters to define patterns of other characters. Although that approach may seem problematic at first, we’ll discuss meta-characters (characters that describe other characters) and how you can use them to create powerful regular expressions.

One of the most basic functions in R that uses regular expressions is the grepl() function, which takes two arguments: a regular expression and a string to be searched. If the string contains the specified regular expression then grepl() will return TRUE, otherwise it will return FALSE. Let’s take a look at one example:

regular_expression <- "a"
string_to_search <- "Maryland"

grepl(regular_expression, string_to_search)
[1] TRUE

In the example above we specify the regular expression "a" and store it in a variable called regular_expression. Remember that regular expressions are just strings! We also store the string "Maryland" in a variable called string_to_search. The regular expression "a" represents a single occurrence of the character "a". Since "a" is contained within "Maryland", grepl() returns the value TRUE. Let’s try another simple example:

regular_expression <- "u"
string_to_search <- "Maryland"

grepl(regular_expression, string_to_search)
[1] FALSE

The regular expression "u" represents a single occurrence of the character "u", which is not a sub-string of "Maryland", therefore grepl() returns the value FALSE. Regular expressions can be much longer than single characters. You could for example search for smaller strings inside of a larger string:

grepl("land", "Maryland")
[1] TRUE
grepl("ryla", "Maryland")
[1] TRUE
grepl("Marly", "Maryland")
[1] FALSE
grepl("dany", "Maryland")
[1] FALSE

Since "land" and "ryla" are sub-strings of "Maryland", grepl() returns TRUE, however when a regular expression like "Marly" or "dany" is searched grepl() returns FALSE because neither are sub-strings of "Maryland".

There’s a dataset that comes with R called state.name which is a vector of Strings, one for each state in the United States of America. We’re going to use this vector in several of the following examples.

head(state.name)
[1] "Alabama"    "Alaska"     "Arizona"    "Arkansas"   "California"
[6] "Colorado"  

Let’s build a regular expression for identifying several strings in this vector, specifically a regular expression that will match names of states that both start and end with a vowel. The state name could start and end with any vowel, so we won’t be able to match exact sub-strings like in the previous examples. Thankfully we can use metacharacters to look for vowels and other parts of strings. The first metacharacter that we’ll discuss is ".". The metacharacter that only consists of a period represents any character other than a new line (we’ll discuss new lines soon). Let’s take a look at some examples using the peroid regex:

grepl(".", "Maryland")
[1] TRUE
grepl(".", "*&2[0+,%<@#~|}")
[1] TRUE
grepl(".", "")
[1] FALSE

As you can see the period metacharacter is very liberal. This metacharacter is most userful when you don’t care about a set of characters in a regular expression. For example:

grepl("a.b", c("aaa", "aab", "abb", "acadb"))
[1] FALSE  TRUE  TRUE  TRUE

In the case above grepl() returns TRUE for all strings that contain an a followed by any other character followed by a b.

You can specify a regular expression that contains a certain number of characters or metacharacters using the enumeration metacharacters. The + metacharacter indicates that one or more of the preceding expression should b present and * indicates that zero or more of the preceding expression is present. Let’s take a look at some examples using these metacharacters:

# Does "Maryland" contain one or more of "a" ?
grepl("a+", "Maryland")
[1] TRUE
# Does "Maryland" contain one or more of "x" ?
grepl("x+", "Maryland")
[1] FALSE
# Does "Maryland" contain zero or more of "x" ?
grepl("x*", "Maryland")
[1] TRUE

You can also specify exact numbers of expressions using curly brackets {}. For example "a{5}" specifies “a exactly five times,” "a{2,5}" specifies “a between 2 and 5 times,” and "a{2,}" specifies “a at least 2 times.” Let’s take a look at some examples:

# Does "Mississippi" contain exactly 2 adjacent "s" ?
grepl("s{2}", "Mississippi")
[1] TRUE
# This is equivalent to the expression above:
grepl("ss", "Mississippi")
[1] TRUE
# Does "Mississippi" contain between 1 and 3 adjacent "s" ?
grepl("s{2,3}", "Mississippi")
[1] TRUE
# Does "Mississippi" contain between 2 and 3 adjacent "i" ?
grepl("i{2,3}", "Mississippi")
[1] FALSE
# Does "Mississippi" contain between 2 adjacent "iss" ?
grepl("(iss){2}", "Mississippi")
[1] TRUE
# Does "Mississippi" contain between 2 adjacent "ss" ?
grepl("(ss){2}", "Mississippi")
[1] FALSE
# Does "Mississippi" contain the pattern of an "i" followed by 
# 2 of any character, with that pattern repeated three times adjacently?
grepl("(i.{2}){3}", "Mississippi")
[1] TRUE

In the last three examples I used parentheses () to create a capturing group. A capturing group allows you to use quantifiers on other regular expressions. In the last example I first created the regex "i.{2}" which matches i followed by any two characters (“iss” or “ipp”). I then used a capture group to to wrap that regex, and to specify exactly three adjacent occurrences of that regex.

You can specify sets of characters with regular expressions, some of which come built in, but you can build your own character sets too. First we’ll discuss the built in character sets: words ("\\w"), digits ("\\d"), and whitespace characters ("\\s"). Words specify any letter, digit, or a underscore, digits specify the digits 0 through 9, and whitespace specifies line breaks, tabs, or spaces. Each of these character sets have their own compliments: not words ("\\W"), not digits ("\\D"), and not whitespace characters ("\\S"). Each specifies all of the characters not included in their corresponding character sets. Let’s take a look at a few exmaples:

grepl("\\w", "abcdefghijklmnopqrstuvwxyz0123456789")
[1] TRUE
grepl("\\d", "0123456789")
[1] TRUE
# "\n" is the metacharacter for a new line
# "\t" is the metacharacter for a tab
grepl("\\s", "\n\t   ")
[1] TRUE
grepl("\\d", "abcdefghijklmnopqrstuvwxyz")
[1] FALSE
grepl("\\D", "abcdefghijklmnopqrstuvwxyz")
[1] TRUE
grepl("\\w", "\n\t   ")
[1] FALSE

You can also specify specific character sets using straight brackets []. For example a character set of just the vowels would look like: "[aeiou]". You can find the complement to a specific character by putting a carrot ^ after the first bracket. For example "[^aeiou]" matches all characters except the lowercase vowels. You can also specify ranges of characters using a hyphen - inside of the brackets. For example "[a-m]" matches all of the lowercase characters between a and m, while "[5-8]" matches any digit between 5 and 8 inclusive. Let’s take a look at some examples using custom character sets:

grepl("[aeiou]", "rhythms")
[1] FALSE
grepl("[^aeiou]", "rhythms")
[1] TRUE
grepl("[a-m]", "xyz")
[1] FALSE
grepl("[a-m]", "ABC")
[1] FALSE
grepl("[a-mA-M]", "ABC")
[1] TRUE

You might be wondering how you can use regular expressions to match a particular punctuation mark since many punctuation marks are used as metacharacters! Putting two backslashes before a punctuation mark that is also a metacharacter indicates that you are looking for the symbol and not the metacharacter meaning. For example "\\." indicates you are trying to match a period in a string. Let’s take a look at a few examples:

grepl("\\+", "tragedy + time = humor")
[1] TRUE
grepl("\\.", "http://www.jhsph.edu/")
[1] TRUE

There are also metacharacters for matching the beginning and the end of a string which are "^" and "$" respectively. Let’s take a look at a few examples:

grepl("^a", c("bab", "aab"))
[1] FALSE  TRUE
grepl("b$", c("bab", "aab"))
[1] TRUE TRUE
grepl("^[ab]+$", c("bab", "aab", "abc"))
[1]  TRUE  TRUE FALSE

The last metacharacter we’ll discuss is the OR metacharacter ("|"). The OR metacharacter matches either the regex on the left or the regex on the right side of this character. A few examples:

grepl("a|b", c("abc", "bcd", "cde"))
[1]  TRUE  TRUE FALSE
grepl("North|South", c("South Dakota", "North Carolina", "West Virginia"))
[1]  TRUE  TRUE FALSE

Finally we’ve learned enough to create a regular expression that matches all state names that both begin and end with a vowel:

  1. We match the beginning of a string.
  2. We create a character set of just capitalized vowels.
  3. We specify one instance of that set.
  4. Then any number of characters until:
  5. A character set of just lowercase vowels.
  6. We specify one instance of that set.
  7. We match the end of a string.
start_end_vowel <- "^[AEIOU]{1}.+[aeiou]{1}$"
vowel_state_lgl <- grepl(start_end_vowel, state.name)
head(vowel_state_lgl)
[1]  TRUE  TRUE  TRUE FALSE FALSE FALSE
state.name[vowel_state_lgl]
[1] "Alabama"  "Alaska"   "Arizona"  "Idaho"    "Indiana"  "Iowa"    
[7] "Ohio"     "Oklahoma"

Below is a table of several important metacharacters:

Metacharacter Meaning
. Any Character
\w A Word
\W Not a Word
\d A Digit
\D Not a Digit
\s Whitespace
\S Not Whitespace
[xyz] A Set of Characters
[^xyz] Negation of Set
[a-z] A Range of Characters
^ Beginning of String
$ End of String
\n Newline
+ One or More of Previous
* Zero or More of Previous
? Zero or One of Previous
| Either the Previous or the Following
{5} Exactly 5 of Previous
{2, 5} Between 2 and 5 or Previous
{2, } More than 2 of Previous

1.7.3 RegEx Functions in R

So far we’ve been using grepl() to see if a regex matches a string. There are a few other built in reged functions you should be aware of. First we’ll review our workhorse of this chapter, grepl() which stands for “grep logical.”

grepl("[Ii]", c("Hawaii", "Illinois", "Kentucky"))
[1]  TRUE  TRUE FALSE

Then there’s old fashioned grep() which returns the indices of the vector that match the regex:

grep("[Ii]", c("Hawaii", "Illinois", "Kentucky"))
[1] 1 2

The sub() function takes as arguments a regex, a “replacement,” and a vector of strings. This function will replace the first instance of that regex found in each string.

sub("[Ii]", "1", c("Hawaii", "Illinois", "Kentucky"))
[1] "Hawa1i"   "1llinois" "Kentucky"

The gsub() function is nearly the same as sub() except it will replace every instance of the regex that is matched in each string.

gsub("[Ii]", "1", c("Hawaii", "Illinois", "Kentucky"))
[1] "Hawa11"   "1ll1no1s" "Kentucky"

The strsplit() function will split up strings according to the provided regex. If strsplit() is provided with a vector of strings it will return a list of string vectors.

two_s <- state.name[grep("ss", state.name)]
two_s
[1] "Massachusetts" "Mississippi"   "Missouri"      "Tennessee"    
strsplit(two_s, "ss")
[[1]]
[1] "Ma"        "achusetts"

[[2]]
[1] "Mi"   "i"    "ippi"

[[3]]
[1] "Mi"   "ouri"

[[4]]
[1] "Tenne" "ee"   

1.7.4 The stringr Package

The stringr package, written by Hadley Wickham, is part of the Tidyverse group of R packages. This package takes a “data first” approach to functions involving regex, so usually the string is the first argument and the regex is the second argument. The majority of the function names in stringr begin with str_.

The str_extract() function returns the sub-string of a string that matches the providied regular expression.

library(stringr)
state_tbl <- paste(state.name, state.area, state.abb)
head(state_tbl)
[1] "Alabama 51609 AL"     "Alaska 589757 AK"     "Arizona 113909 AZ"   
[4] "Arkansas 53104 AR"    "California 158693 CA" "Colorado 104247 CO"  
str_extract(state_tbl, "[0-9]+")
 [1] "51609"  "589757" "113909" "53104"  "158693" "104247" "5009"  
 [8] "2057"   "58560"  "58876"  "6450"   "83557"  "56400"  "36291" 
[15] "56290"  "82264"  "40395"  "48523"  "33215"  "10577"  "8257"  
[22] "58216"  "84068"  "47716"  "69686"  "147138" "77227"  "110540"
[29] "9304"   "7836"   "121666" "49576"  "52586"  "70665"  "41222" 
[36] "69919"  "96981"  "45333"  "1214"   "31055"  "77047"  "42244" 
[43] "267339" "84916"  "9609"   "40815"  "68192"  "24181"  "56154" 
[50] "97914" 

The str_order() function returns a numeric vector that corresponds to the alphabetical order of the strings in the provided vector.

head(state.name)
[1] "Alabama"    "Alaska"     "Arizona"    "Arkansas"   "California"
[6] "Colorado"  
str_order(state.name)
 [1]  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
[24] 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46
[47] 47 48 49 50
head(state.abb)
[1] "AL" "AK" "AZ" "AR" "CA" "CO"
str_order(state.abb)
 [1]  2  1  4  3  5  6  7  8  9 10 11 15 12 13 14 16 17 18 21 20 19 22 23
[24] 25 24 26 33 34 27 29 30 31 28 32 35 36 37 38 39 40 41 42 43 44 46 45
[47] 47 49 48 50

The str_pad() function pads strings with other characters which is often useful when the string is going to be eventually printed for a person to read.

str_pad("Thai", width = 8, side = "left", pad = "-")
[1] "----Thai"
str_pad("Thai", width = 8, side = "right", pad = "-")
[1] "Thai----"
str_pad("Thai", width = 8, side = "both", pad = "-")
[1] "--Thai--"

The str_to_title() function acts just like tolower() and toupper() except it puts strings into Title Case.

cases <- c("CAPS", "low", "Title")
str_to_title(cases)
[1] "Caps"  "Low"   "Title"

The str_trim() function deletes whitespace from both sides of a string.

to_trim <- c("   space", "the    ", "    final frontier  ")
str_trim(to_trim)
[1] "space"          "the"            "final frontier"

The str_wrap() function inserts newlines in strings so that when the string is printed each line’s length is limited.

pasted_states <- paste(state.name[1:20], collapse = " ")

cat(str_wrap(pasted_states, width = 80))
Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware Florida
Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine
Maryland
cat(str_wrap(pasted_states, width = 30))
Alabama Alaska Arizona
Arkansas California Colorado
Connecticut Delaware Florida
Georgia Hawaii Idaho Illinois
Indiana Iowa Kansas Kentucky
Louisiana Maine Maryland

The word() function allows you to index each word in a string as if it were a vector.

a_tale <- "It was the best of times it was the worst of times it was the age of wisdom it was the age of foolishness"

word(a_tale, 2)
[1] "was"
word(a_tale, end = 3)
[1] "It was the"
word(a_tale, start = 11, end = 15)
[1] "of times it was the"

1.7.5 Summary

String manipulation in R is useful for data cleaning, plus it can be fun! For prototyping your first regular expressions I highly recommend checking out http://regexr.com/. If you’re interested in what some people call a more “humane” way of constructing regular expressions you should check out the rex package by Kevin Ushey and Jim Hester. If you’d like to find out more about text analysis I highly recommend reading Tidy Text Mining in R by Julia Silge and David Robinson.