Week2-5Spelling similarity:edit distance

Spelling similarity

• Typos
• Variants in spelling

Edit operations

• Insertion
• Deletion
• Substitution
• Multiple edits

Levenstein method

• Based on dynamic programming
• Insertions, deletions and substitutions usually have a cost of 1

Example

we want to calculate the edit distance of strength and trend.

Definitions

• $s_1(i)$: $i^{th}$ character in string $s_1$
• $s_2(j)$: $j^{th}$ character in string $s_2$
• $D(i, j)$: edit distance between a prefix of $s_1$ of length i and a prefix of $s_2$ of length j
• $t(i, j)$: cost of aligning the $i^{th}$ character in string $s_1$ with the $j^{th}$ character in string $s_2$

Recursive dependencies

D(i, 0) = i
D(0, j) = j
D(i, j) = min{D( i - 1, j ) + 1, 
              D( i, j - 1 ) + 1,
              D( i - 1, j - 1 ) + t( i, j )}

Simple edit distance

t( i, j ) = 0 iff s1( i ) = s2( j )
t( i, j ) = 1 otherwise 

Initialization

Recursion

Other costs

• Damerau modification
  
  • swaps of 2 adjacent characters also have cost of 1( people are likely to swap the adjacent characters)
  • Lev( cats, cast ) = 2
  • Dam( cats, cast ) = 1

Other edit distance

• Dist( sit down, sit clown ) = 1?

  • model the errors common with optical character recognition(OCR), i.e. d is likely to be writen as cl

• Dist( qeather, weather ) = 1, Dist( leather, weather ) = 2?

  • model spelling errors introduced by fat finger, i.e. q and w are near on the keyboard whereas l and w is far(unlikely to be a typo)

Edit distance is also used in genetic sequence and amino acid sequence in bioinformatics.