This new radio leads to interesting suggestions (artists I know and artists I never heard):

It's based on a really simple algorithm: we start from a given track, a and grab all the tracks that people played *after* this one. Then we pick a random track among those (with higher weight for more popular tracks).

Rinse and repeat.

The whole thing took less than 100 lines of code to write (tests included) and works suprising well!

Thank you @gordon for the idea :3

Haha, reminds memory of when I asked for "The Blaze" and "Tally Hall" to be uploaded :P (glad the upload feature improved a lot !)

@eliotberriot @gordon please... resist from implementing recommendation based on algorithms. Let’s support human recommendation and encourage curiosity and serendipity. Algorithms creates filter bubble effect and passivity.

@imacrea @gordon I'm also in favor of human recommendations (you can find a more detailed post about that here and I do think they are superior to purely algorithmic recommendations.

However, I do think algorithms can help you discover new things you'd never have listen before, and develop your curiosity, especially for people stuck in their bubble.

@imacrea @gordon in the current case, there is indeed an issue with the popularity-based weight though, because it can easily snowball. However, if we remove that bit, we could have a decent system promoting serendipity/discovery :)

@imacrea @gordon also, this algorithm should not create a filter bubble at the user level, because it yields the same recommendations for each user. It does not use your own personal history, but everyones history.

I guess you could say it's a filter buble at the instance / network level, but then human recommendations from the same sphere would also be a filter bubble.

@imacrea @gordon (I'm following my thoughts here, because your post raises interesting questions).

I don't think passivity or lack of serendipity are correlated with algorithmic recommendations.

When watching TV or listening to mainstream radios, you are totally passive, and have zero serendipity, while content is curated by humans. TV is a human filter bubble.

@imacrea @gordon
However, I do agree, that many (most?) of the recommendation algorithms YouTube, Netflix, Spotify…) out there encourage passivity and filter bubbles.

But it's more a business model issues : those services earn more the longer you use them. They have financial interest to keep you passive, not to hurt your habits, or to promote the most profitable content, for them.

It's not the case for Funkwhale, because users are not paying or selling their data to use the platform.

@eliotberriot @gordon I look forward to see what it'll be like ! Seems promising. I hope it'll federate, or little instance won't be able to have okayish recommendations.

@marsxyz @eliotberriot well, there is a "listen" ActivityPub activity type, but i don’t know if Funkwhale produces them at this time. It would be a nice way to achieve federation in this feature

@gordon @marsxyz we don't send those, but that's probably the first user activity we'll federate, yes!

Sign in to participate in the conversation is one server in the network