Overlap Similarity Score

Introduction:

This page describes the algorithm of calculating similarity score for leading and trailing characters overlapping. This algorithm is enhanced for better ranking from the Ensemble by adding weights on the leading and ending overlapping (see examples at the end) and use it as tiebreaker when the score of edit distance and phonetic are tie.

Algorithm:

• not case sensitive
• Similarity score is between 0.00 and 1.00
  • Enhanced Overlap Algorithm:

    condition	formula	example
    leadOverLap = minLen	Score = (1.0*leadOverlap + 0.1*trailOverlap)/(1.0*maxLen)	123 and 123123 leadOverLap = 3 (123) trailOverLap = 3 (123) minLength = 3 (123) maxLength = 6 (123123) score = (3 + 0.3)/6 = 0.55
    trailOverLap = minLen	Score = (0.1*leadOverlap + 1.0*trailOverlap)/(1.0*maxLen)	ends and leadends leadOverLap = 0 trailOverLap = 4 (ends) minLength = 4 (ends) maxLength = 8 (leadends) score = (0 + 4)/8 = 0.50
    else	Score = (1.0*leadOverLap + 1.0*trailOverLap)/(1.0*maxLen)	dianosed and diagnosed leadOverLap = 3 (dia) trailOverLap = 5 (nosed) minLength = 8 (dianosed) maxLength = 9 (diagnosed) score = (3 + 5)/9 = 0.89

    where:

    • minLen is the shorter length of two input strings
    • maxLen is the longer length of two input strings
    • leadOverLap is the length of matching characters at the beginning
    • trailOverLap is the length of matching characters at the end

  • More examples:

    Example	String 1	string 2	Calculation	Score	Notes
    Ex-1	spel	spell	=(4+0.1)/5	0.82	spel is closer to spell than speil
    	spel	speil	=(3+1)/5	0.80
    Ex-2	spell	sspell	=(0.1+5)/6	0.85	spell is closer to sspell than nspell
    	spell	nspell	=(0+5)/6	0.83

Source code:

• OverlapSimilarity.java