String similarity
Score how similar two texts are with Jaro-Winkler, Dice, Jaccard and Levenshtein ratio. Runs entirely on your device, nothing is uploaded.
String similarity scores two pasted texts with four measures at once: Jaro-Winkler, Dice coefficient, Jaccard on word sets, and a Levenshtein ratio. Levenshtein compares whole characters, walking every pair between the two texts to build its edit distance. Comparing texts whose combined character product tops 25 million is refused outright, since that cost grows with the square of the length.
Questions
What do the four scores mean?
Jaro-Winkler and the Levenshtein ratio both compare character by character and reward a shared prefix, so they are sensitive to typos and near-matches. Dice coefficient compares overlapping two-character sequences. Jaccard compares the two texts as sets of whole words, ignoring order and repeats entirely. All four print to four decimal places, from 0 for no similarity to 1 for identical.
Why did it refuse to compare my texts?
Because the product of their two lengths passed 25,000,000, which the tool refuses with "inputs too large for this comparison, trim one side". The Levenshtein calculation compares every character of one text against every character of the other, so its cost grows with the square of the combined length.
Does it care about capitalization?
It depends on the measure. Levenshtein and Jaro-Winkler compare the raw characters you typed, so Apple and apple count as different. Dice and Jaccard lowercase both texts before comparing, so case differences do not affect those two scores.
What counts as a match for the Jaccard score?
Whole words, split with a Unicode letter and digit pattern and lowercased, then compared as two sets. A word present in both texts counts once no matter how many times it repeats in either one, since sets have no duplicates.
Is there one overall similarity number?
No. All four scores are printed together with no combined average, since they measure different things and a single blended number would hide which kind of similarity you are looking at.