Hacker News Reader

Comment by StavrosK | Hacker News Reader

Oh man... I haven't been able to find any recommendation algorithms that recommended songs I liked even 1% of. Spotify's is a trash fire. Last.fm's wasn't great. Pandora is okay. I am very interested in this project, but the setup instructions are... long. Not hard if you're good with computers, but long.

Hopefully it'll be worth the hassle, but please try to make it easier to install somehow?

View on HN

It would probably be better if these sites found a way to compile songs from DJ mix tracklists and then scan for the highest quality audio from sites like YouTube and Spotify and then stream entire mix tracklists by genre as individual tunes, that way matching songs within music genres would be more properly targeted genre-wise. I fear a future where AI picks all the music because what I like to hear usually is never what I hear on streaming cues.

Licensing within the music industry is the #1 enemy of getting good music cue suggestions across all music artists though, as many streaming providers can't just access any song they want to, because the agreements to play on each service are so limited and segmented.

Have you tried putting 100 songs you really like that are in similar sub genres into a new playlist in Spotify, then playing the last song and letting it generate new suggestions from there? Kind of a weird hack, but I’ve been getting good results from it that have improved my #2 daily list selections a lot.

Pandora’s concept of a music genome is quite fascinating too. For whatever reason the repeat rate gets too high though, even if you seed it with multiple songs.

Unfortunately I like a little bit of everything, so I don't have 100 songs I like in the same genre. I have 1000 songs from wildly different genres, so I'm very hard to generate suggestions for.

Ah. You might get some benefit even doing 10 song playlists and letting them run after the last song, with ml driving the song selection. This has worked well for me personally. Might be worth experimenting with.

Have you tried Gnoosic (http://www.gnoosic.com)? That is the one that works best for me.

I'm thinking of incorporating gnoosic (or something similar--haven't looked yet to see if they have an API) into Lagukan for recommending new artists. Mainly my work has focused on re-recommending songs you already know.

No, I didn't know about it, thank you!

When I use the "Start Radio" function on a song in Google Music it often leads to good stuff.

Yes, definitely! I'm working on mobile apps right now which will be super easy to use, and then I'll likely come back and improve the desktop app.

The current app was optimized for development time, just so I could get a prototype working at least for my own use as quickly as possible.

Oh, if it's a prototype, then that's understandable. Disregard, I will beta-test and report back!

Thanks!

I think you need to add a step to install libspotify-dev on Ubuntu, mopidy-spotify-web from your repo wouldn't install for me unless I installed that first.

Also, did you already install `mopidy-spotify` (https://github.com/mopidy/mopidy-spotify#installation) already? I think that package is supposed to have all the needed dependencies, maybe it's out of date.

Yes I did, alas.

Oh, good to know! I'll update the docs. I'm on arch linux so I wasn't aware. Should've spun up a VM to test it first or something.

Reply on news.ycombinator.com