smith_waterman {text.alignment}

Align text using Smith-Waterman

Description

Align text using the Smith-Waterman algorithm. The Smith–Waterman algorithm performs local sequence alignment. It finds similar regions between two strings.
Similar regions are a sequence of either characters or words which are found by matching the characters or words of 2 sequences of strings.
If the word/letter is the same in each text, the alignment score is increased with the match score, while if they are not the same the local alignment score drops by the gap score. If one of the 2 texts contains extra words/letters, the score drops by the mismatch score.

Usage

smith_waterman(
  a,
  b,
  type = c("characters", "words"),
  match = 2L,
  mismatch = -1L,
  gap = -1L,
  lower = TRUE,
  similarity = function(x, y) ifelse(x == y, 2L, -1L),
  tokenizer,
  collapse,
  edit_mark = "#",
  implementation = c("R", "Rcpp"),
  ...
)

Arguments

a

a character string of length one

b

a character string of length one

type

either 'characters' or 'words' indicating to align based on a sequence of characters or a sequence of words. Defaults to 'characters'.

match

integer value of a score to assign a match (a letter/word from a and b which are the same during alignment). This value should be bigger than zero. Defaults to 2.

mismatch

integer value of a score to assign a mismatch (leave out 1 word / 1 letter from 1 of the 2 input strings during alignment). This value should be smaller or equal to zero. Defaults to -1.

gap

integer value of a score to assign a gap (drop 1 word / letter from each of the 2 input strings during alignment). This value should be smaller or equal to zero. Defaults to -1.

lower

logical indicating to lowercase text before doing the alignment. Defaults to TRUE.

similarity

optionally, a function to compare 2 characters or words. This function should have 2 arguments x and y with the 2 letters / words to compare and should return 1 number indicating the similarity between x and y. See the examples.

tokenizer

a function to tokenise text into either a sequence of characters or a sequence of words. Defaults to tokenize_letters if type is 'characters' and tokenize_spaces_punct if type is 'words'

collapse

separator used to combined characters / words back together in the output. Defaults to ” for type 'characters' and a space for type 'words'

edit_mark

separator to indicated a gap/mismatch between sequences. Defaults to the hashtag symbol.

implementation

either 'R' or 'Rcpp' indicating to perform the alignment in Rcpp or with plain R code. Defaults to 'R'.

...

extra arguments passed on to tokenizer

Details

The code uses similar code as the textreuse::local_align function and also allows to align character sequences next to aligning word sequences

Value

an object of class smith_waterman which is a list with elements

• type: The alignment type

• edit_mark:The edit_mark

• sw: The Smith-Waterman local alignment score

• similarity: Score between 0 and 1, calculated as the Smith-Waterman local alignment score / (the number of letters/words in the shortest text times the match weight)

• weights: The list of weights provided to the function: match, mismatch and gap

• matches: The number of matches found during alignment

• mismatches: The number of mismatches found during alignment

• a: A list with alignment information from the text provided in a. The list elements documented below

• b: A list with alignment information from the text provided in b. The list elements documented below

Elements a and b are both lists which contain

• text: The provided character string of either a or b

• tokens: A character vector of the tokenised texts of a or b

• n: The length of tokens

• similarity: The similarity to a calculated as the Smith-Waterman local alignment score / (the number of letters/words in the a or b text times the match weight)

• alignment: A list with the following elements

  • text: The aligned text from either a or b where gaps/mismatches are filled up with the edit_mark symbol

  • tokens: The character vector of tokens which form the aligned text

  • n: The length of the aligned text

  • gaps: The number of gaps during alignment

  • from: The starting position in the full tokenised tokens element from either a or b where the aligned text is found. See the example.

  • to: The end position in the full tokenised tokens element from either a or b where the aligned text is found. See the example.

See Also

https://en.wikipedia.org/wiki/Smith-Waterman_algorithm

Examples

## align sequence of letters
smith_waterman("Joske Vermeulen", "Jiske Vermoelen")
smith_waterman("Joske Vermeulen", "Ik zoek naar Jiske Versmoelen, waar is die te vinden")
smith_waterman("Joske", "Jiske")
smith_waterman("Joske", "Jiske",
               similarity = function(x, y) ifelse(x == y | (x == "o" & y == "i"), 2L, -1L))

## align sequence of words
a <- "The answer is blowin' in the wind."
b <- "As the Bob Dylan song says, the answer is blowing in the wind."
smith_waterman(a, b)
smith_waterman(a, b, type = "characters")
smith_waterman(a, b, type = "words")
smith_waterman(a, b, type = "words", similarity = function(x, y) adist(x, y))
smith_waterman(a, b, type = "words", 
               tokenizer = function(x) unlist(strsplit(x, "[[:space:]]")))
x <- smith_waterman(a, b, type = "words")
x$b$tokens[x$b$alignment$from:x$b$alignment$to]            
               
# examples on aligning text files
a <- system.file(package = "text.alignment", "extdata", "example1.txt")
a <- readLines(a)
a <- paste(a, collapse = "\n")
b <- system.file(package = "text.alignment", "extdata", "example2.txt")
b <- readLines(b)
b <- paste(b, collapse = "\n")
smith_waterman(a, b, type = "characters")
smith_waterman(a, b, type = "words")
smith_waterman("Gistel Hof", b, type = "characters")
smith_waterman("Bailiestraat", b, type = "characters")
smith_waterman("Lange rei", b, type = "characters")

# examples on extracting where elements were found
x <- smith_waterman("Lange rei", b)
x$b$tokens[x$b$alignment$from:x$b$alignment$to]
as.data.frame(x)
as.data.frame(x, alignment_id = "alignment-ab")

x <- lapply(c("Lange rei", "Gistel Hof", NA, "Test"), FUN = function(a){
  x <- smith_waterman(a, b)
  x <- as.data.frame(x)
  x
})
x <- do.call(rbind, x)
x