I built an app that recommends progressive artists

Magnum Vaeltaja wrote: Very cool!  The results were a bit mixed, with some selections seeming odd (Sloche not turning up Maneige right away?), but what can you expect with an algorithm? Also, some of the artists' iTunes links linked to the wrong artists. Bubu, for example, turned up a Brazilian pandeiro drum player dressed as a clown, as opposed to the Argentinian eclectic masterpiece of Anabelas.  So obviously some fine tuning could still be done, but I enjoyed it nonetheless.

I will definitely try to keep improving the app. That's why I published it here.

If you search for Maneige, Sloche actually appears in the first page. The problem here is that the results are not symmetric, something that will be fixed soon.

Thanks!

I tried to keep the algorithm simple. I do not assume any genre regarding bands. For every band, I check all their reviews for mentions of another band. I defined the similarity value between bands A and B as basically the condicional probability of finding a mention to A knowing it's a review from B. Then this value is scaled with the popularity of the band. We don't want to show very popular bands because the user may already know them, but very obscure bands with few reviews may also not be ideal.

If you meant n-grams, I did not use them at all. For minimizing the number of false-positives I take into account the number of users that like both band A and B. If this value is large, them it must mean an higher chance of the artists being similar.

You only have one written review on PA. That was not enough to calculate similar artists yet.