Show HN: A Go library for approximate string matching using character n-grams

github.com

2 pointsqrv3w2 comments

I have only looked at the README.

> If you run go test the tests will pass which validate that Levenshtein performs < 60% accuracy and closestmatch performs with > 98% accuracy.

If you've set those tests up yourself, what you mean by "accuracy" may be what your library achieves.

You're absolutely right. You can check out the tests [1] - its a lot of re-ordering and inserting random letters in random places, which basically trying to emulate sloppy user input.

Your comment is also getting at a caveat: this library works better than Levenshtein for long strings (multiple words) but worse than Levenshtein for short strings (e.g. single words) as there is too much overlap in character n-grams.

[1] https://github.com/schollz/closestmatch/blob/master/closestm...

Reply on news.ycombinator.com