How is similarity calculated

[aceakash]

How is similarity calculated

[dangh]

Hey @18672786357 It's 6 months old and you might found the answer yourself but I like to post my answer here for the future me 😛

The Sørensen–Dice algorithm is calculate the number of matching items in 2 sets using the formular DSC = 2(𝚇⋂𝚈)/(|𝚇|+|𝚈|) where |𝚇| and |𝚈| are the number of items in each set.

This library convert each strings to multiple bigrams (2 characters substring) and apply the formular above. If you break the string manually you will see that the number of items is the length of string minus 1.

"abide" -> ['ab', 'bc', 'cd', 'de']

We have two strings so |𝚇|+|𝚈| will be (s1.length - 1) + (s2.length - 1). That's where the number 2 come from.

[18672786357]

return (2.0 * intersectionSize) / (first.length + second.length - 2);，，，，Why is this line of code minus 2

[Rabbitzzc]

let str = '12345'  // len = 5
let set = {'12', '23', '34', '45'} len =4
// so one string represents the denominator that needs to:
X = len - 1 = str.length - 1

[zscaiosi]

Hey @18672786357 It's 6 months old and you might found the answer yourself but I like to post my answer here for the future me stuck_out_tongue

The Sørensen–Dice algorithm is calculate the number of matching items in 2 sets using the formular DSC = 2(𝚇⋂𝚈)/(|𝚇|+|𝚈|) where |𝚇| and |𝚈| are the number of items in each set.

This library convert each strings to multiple bigrams (2 characters substring) and apply the formular above. If you break the string manually you will see that the number of items is the length of string minus 1.

"abide" -> ['ab', 'bc', 'cd', 'de']

We have two strings so |𝚇|+|𝚈| will be (s1.length - 1) + (s2.length - 1). That's where the number 2 come from.

@dangh the bigrams are applied to the Map structure (firstBigram), not to the variables first and second !
first.substring(i, i + 2); and second.substring(i, i + 2); DO NOT modify the objects (strings), but return new strings (substrings).

It does not make any sense to subtract 2!

[aceakash]

@dangh has the right explanation and the link that I used the formula from.
The result is as expected, 'night' and 'nacht' are similar to a degree of 0.25
https://jsbin.com/kidibosato/edit?js,console