search

aceakash / string-similarity

Finds degree of similarity between two strings, based on Dice's Coefficient, which is mostly better than Levenshtein distance.
MIT License
2.52k stars 124 forks source link

This is not the Dice coefficient #114

Open vibl opened 3 years ago

vibl commented 3 years ago

Your algorithm is not the Dice coefficient. It counts all bigram duplicates, whereas the Dice coefficient only counts distinct bigrams (as defined in Wikipedia).

As an example, let's compare two versions of the main file of this repo (https://github.com/aceakash/string-similarity/blob/2718c82bbbf5190ebb8e9c54d4cbae6d1259527a/compare-strings.js and the latest https://github.com/aceakash/string-similarity/blob/eaeec5d74c98a6f6fcb1b06fad44ad7f3d8c2965/src/index.js. They have a Dice coefficient of 0.90, but this lib string-similarity outputs 0.74 when comparing these two files.

Please have a look at the implementations in Talisman, NLTK or in many languages in https://en.wikibooks.org/wiki/Algorithm_Implementation/Strings/Dice%27s_coefficient

aimeeaidanu commented 1 year ago

frr bruh like "dollar' and "money" return a match of 0 :((( like dawg I want semantic similarity who needs string similarity anyways 🤷