Implement tag/genre browsing

It's worth considering an open source music recommendation neural network here also:

• Worth quickly glancing over: https://towardsdatascience.com/how-to-build-a-simple-song-recommender-296fcbc8c85
• Potential Python library: https://github.com/Coder-Yu/Yue

During the implementation of tag/genre browsing - so that recommendations could also be made (future feature) to users playlists, and perhaps the "Radio" feature - This is one of the key features that makes streaming services such as Spotify so successful with Social music, and what all other self-hosted services currently lack.

@n3ddy One thing I don't like with those algorithms is that they are basically black boxes, making it impossible for end users to understand why something is recommended, or not. Also, ML stuff is usually quite expensive in terms of resources and skills to develop, maintain and deploy.

In the same way I think chronological timelines are superior to algorithmic ones, I'm more inclined to simpler recommendations systems, based on explicit, understandable criterias:

• Person X, that you follow, has listened to track Y
• Person X, that you follow, has favorited to track Y
• Artist Z, that you favorited, has released a new album
• Person X, that you follow, has created a playlist
• Person W has recommended you track Y

For recommendations based on the music itself, someone suggested https://github.com/Polochon-street/bliss a while ago, which is able to extract features from the music. We could use it to build smarter radios (similar songs, different songs, calm songs, etc.).

One thing I don't like with those algorithms is that they are basically black boxes, making it impossible for end users to understand why something is recommended, or not. Also, ML stuff is usually quite expensive in terms of resources and skills to develop, maintain and deploy.

I understand, however the downside of this is that if you have a small "closed" Funkwhale instance, you're not going to get huge benefit's from this if you don't have lots of active users - where as if there was some sort of global recommendation Engine/API that benefited from the meta/scrobbing data of all Funkwhale users (with the ability to opt out, and not collecting personal information of course - just what tracks are popular and relate to other similar tracks etc) then the Recommendation engine would essentially be very powerful for smaller scale Funkwhale installs

not if we use the federation for that purpose: you can be on your own, small instance, and get recommendations from people you interact with, and suggstion based on other people activity and your own history?

I've been developing a recommendation client for MPD for many years now, the best compromise I settled on is the last.fm API.

MetaBrainz Foundation started an alternative to last.fm (cf. #497 (closed)), but it still lacks recommendation.

I'm not aware of any other alternative providing freely access to recommendations.

These are indeed a kinda "black box" recommendation but it is pretty efficient to feed a play queue. I believe it could be a very neat and attractive radio feature to add to funkwhale, especially with the potential of federated music libraries as mentioned @eliotberriot !

By the way, the issue should be renamed IMHO ;) What about use a "music recommendation brainstorming/smart radio" or something else more related to the discussion.

@kaliko thank you for sharing your experience!

For me, the discussion here slightly diverged from the initial issue (although it was not well specified): annotating tracks / albums / artists with genre / tags information and expose this data in the UI does not necessarily imply working or integrating a recommendation algorithm.

I agree both can be linked, and a good enough algorithm would probably leverage this data to offer recommendations anyway, but I still think those can be tackled as separate issues.

As for the last.fm integration, we will probaby integrate it at some point, because lot of users want to forward their listening history from Funkwhale to last.fm :)

I created a new ticket for algorithmic recommendations #524 (closed)

mentioned in issue #523

mentioned in issue #524 (closed)

mentioned in issue #525 (closed)

mentioned in issue #526 (closed)

mentioned in issue #527 (closed)

mentioned in issue #528 (closed)

Subtickets to this task :

• #525 (closed) : update model and API
• #526 (closed) : update import to get the genre from metadata
• #527 (closed) : update UI to show genre and allow to add tags
• #528 (closed) : implement genre-based radios

mentioned in issue #592 (closed)

changed milestone to %0.20

mentioned in commit 76bf5cca

mentioned in commit 07a959f4

mentioned in commit 07ed5c7c

mentioned in commit 8f4306f7