Live data from Hacker News

Show HN: Selecta – Tune Your Own Spotify Recommendation Algorithm

selecta.ciaran.co.za

11–20 of 32 posts

Re: Show HN: Selecta – Tune Your Own Spotify Recommendation Algorithm

#11

A feature I've long desired would be to be able to ask what lies between or along the average or sum of the vectors representing two arbitrary songs or bands. For example what music is midway between "Florence and the Machine" and "Primus"? Or, what happens if you take the vector from Green Day to Modest Mouse and extend it - do you land on another (presumably obscure indie) band? If Spotify won't give me those toys,…

There's a webapp called Boil the Frog that does exactly this!

Re: Show HN: Selecta – Tune Your Own Spotify Recommendation Algorithm

#12

A feature I've long desired would be to be able to ask what lies between or along the average or sum of the vectors representing two arbitrary songs or bands. For example what music is midway between "Florence and the Machine" and "Primus"? Or, what happens if you take the vector from Green Day to Modest Mouse and extend it - do you land on another (presumably obscure indie) band? If Spotify won't give me those toys,…

I mean... Spotify tracks a bunch of cool (algorithmically measured) sets of metadata for each song. Basically the ones you have on this site. And there are APIs to retrieve them for any song, and you can use them all as arguments for search. I don't see why you wouldn't be able to find a thing in between two things.

Keep in mind that the actual accuracy of those metrics is debatable. They are cool... But mostly for funzies, or for approximate sorting in 3rd party tools.

Re: Show HN: Selecta – Tune Your Own Spotify Recommendation Algorithm

#13
Seems to ask for quite a lot of data:

> Your email > The type of Spotify subscription you have, your account country and your settings for explicit content filtering > Your name and username, your profile picture, how many followers you have on Spotify and your public playlists

> Take actions in Spotify on your behalf > Create, edit, and follow private playlists > Create, edit, and follow playlists

Re: Show HN: Selecta – Tune Your Own Spotify Recommendation Algorithm

#14
post #13

Seems to ask for quite a lot of data: > Your email > The type of Spotify subscription you have, your account country and your settings for explicit content filtering > Your name and username, your profile picture, how many followers you have on Spotify and your public playlists > Take actions in Spotify on your behalf > Create, edit, and follow private playlists > Create, edit, and follow playlists

Hey, thanks for the feedback. Sign-in and saving to a profile was the most requested feature across the board when I previously launched Selecta, and as such it needs access to enable features that were requested by those who enjoyed using it.

I'd be more than happy to share what each of those datapoints is used for (and, in some cases where I only need an auth scope for one permission, but they're grouped together). Selecta is strictly open-source and anonymous analytics.

You can find the auth scopes I use in this file: https://github.com/knightspore/selecta/blob/main/src/app/api...

and the related access they provide here: https://developer.spotify.com/documentation/web-api/concepts...

Re: Show HN: Selecta – Tune Your Own Spotify Recommendation Algorithm

#15

A feature I've long desired would be to be able to ask what lies between or along the average or sum of the vectors representing two arbitrary songs or bands. For example what music is midway between "Florence and the Machine" and "Primus"? Or, what happens if you take the vector from Green Day to Modest Mouse and extend it - do you land on another (presumably obscure indie) band? If Spotify won't give me those toys,…

If I understand correctly, what I'm getting back from their API is basically what lies between, but without any of the interesting data they're using the do the calculation. One can hope, but I'd imagine that data is priceless to them and won't ever be available to us.

Re: Show HN: Selecta – Tune Your Own Spotify Recommendation Algorithm

#16
This is great! I've done a similar thing with knn clustering and it had fun results. Would you mind sharing a bit about the data science behind this project? One issue I ran into was the numeric features I clustered with didn't care about genre, so the resulting playlists had some interesting mixes.

Re: Show HN: Selecta – Tune Your Own Spotify Recommendation Algorithm

#17

A feature I've long desired would be to be able to ask what lies between or along the average or sum of the vectors representing two arbitrary songs or bands. For example what music is midway between "Florence and the Machine" and "Primus"? Or, what happens if you take the vector from Green Day to Modest Mouse and extend it - do you land on another (presumably obscure indie) band? If Spotify won't give me those toys,…

There's a webapp called Boil the Frog that does exactly this!

Cheers, going to give it a look.

Re: Show HN: Selecta – Tune Your Own Spotify Recommendation Algorithm

#19
post #8

Interesting! Too late for me; Spotify's declining recommendation capability was one of several issues that allowed me to cancel my subscription. I really wish streaming services would just let us give conscious direction instead of trying to hack our brains. They're not as good at as they think.

Did you end up switching to a different platform? If so, which one?

Re: Show HN: Selecta – Tune Your Own Spotify Recommendation Algorithm

#20
I have been playing with this and I really like the concept. A few nitpicky points of feedback

1) It would be nice if the button to play a sample of the song was located on each song or somewhere close to it, so that you wouldn't have to move your mouse to the bottom

2) The volume slider isn't super obvious as to its function, I originally thought it was a seek bar. Maybe a Volume icon would help here

3) Most importantly, it would be great if the app could play full tracks for users with Spotify Premium, preferably automatically as well, so that this could be useful as a "radio". One feature I really want is to have a "tunable recommendation radio" of songs that all sound a certain way and this comes very close.

Post reply on HN