Known As I

We're planning a major overhaul of the music genres

  • 3 weeks ago
  • 42
  • 10
announcement

We're planning a major overhaul of the music genres

TL;DR Genres will be replaced by classification tags which can be assigned by the artist and all listeners.

The introduction of the genres was a big step forward to find the music you like. It enabled us to provide you with a fresh stream of related tracks and weekly charts.

About one year after the introduction, we all made experiences showing some shortcomings of the current implementation: important genres are missing from the list, some genres are overcrowded while others seem to be dead, users were starting to abuse other genres to chart without any competition, lack of understanding of the genres, ...

We thought about several updates to mitigate those problems but every solution raised new questions. So we stepped back to re-evaluate the entire situation.

Some of our main challenges:

  • Genres are never sharp categories and their definition even changes over time
  • Avoid single-track niche-genres
  • Include sub-genres and fusion-genres in search
  • Add requested genres easily
  • Charts
  • Abuse

Taking one more step back we realized that the main function of the genres is to find similar music. But there are even more measures to group tracks together: moods, situations, XY type beat, ... and they all overlap. Lots of users already added those additional information to their tracks using the tags.

To cut a long story short: We are going to convert everything into tags - even the genres. While this seems to be a step backwards, we also introduce mechanisms to find related tags and allow everyone to assign tags to any track they want.

So in future one can express: "Hey that makes me wanna #dance", "Sounds like #hip hop and I like the #rap part", "I like that for #workout" (the full text expression is just for explanation - you'd just click on the #tag)

The more input you provide the better the results when looking for #dance music or a specific genre. We can even provide #dance charts or #workout playlists.

There are still some open challenges and there will be lots of WTF moments during the transition. But we believe we're heading towards a globally better solution.

Q/A

How to combat tag trolls? Users will get some kind a trust-level which affects the weight of their tag-choice. If we recognize that someone is abusing the system they will get their trust-level reduced. We can even automatically detect if users are consequently diverging from the community's choice and reduce their trust-level based on that. That also works vice versa: users that are known to have a good sense for the right tags may get a boost of their level. We might also want to enable tagging only if the user actually listened to n seconds or x % of the track.

What happens if 100 listeners select 1000 different tags? Only the most popular tags will be shown. So if 10 users add #sad and one user adds #melancholic, all will be taken into account internally but #melancholic might be hidden. (both tags are considered to be similar btw)

(I'll fill this according to your feedback)

Create an account or Login to write a comment.

  • User Avatar

    Known As I

    1 week ago

    me: "computer: what means #rain to our community?"

    computer: "soothing, quiet, calm, peace, calming, lullaby, nature, soft, dreaming, gentle, relaxing, bird, wind, water, emotion, dream, rose, thunder, sea, sleep, chillax, hopeful, peaceful, slow, crazy bout u, beautiful, chilling, relax, beauty, atmosphere, days, eyes, dr, lost, static, inspiring, thoughts, zen, ocean, pretty, reverb, paradise, memories, morning, aeiter, kevinwire, lovely, dramatic, melancholic, idea, hope, echo, opaqity, depressing, emotional, warm, serenity, winter, pad, story, falling, past, mellow, minor, spring, mr, chimes, tears, ..."

    me: "Wow! Well done, community!"

  • User Avatar

    ᅠᅠᅠ

    1 week ago

    this is pretty good :)

    i was thinking charts needed an update, and i agree with everything.

    but i do have one problem: how will people find new music?

    in the current charts, only a few artists rule the charts because of their popularity. how will other artists compete? just a question.

    3 more
  • User Avatar

    Celcius

    1 week ago

    I'm glad the staff is taking serious steps to address this issue.

    How will this affect the charts system? Are these going to be completely overhauled?

    • User Avatar

      Known As I

      1 week ago

      It boils down to the question which tracks shall participate in a certain chart list (e.g. #trap):

      1. a track which is good but only 10 % #trap

      2. a track which is bad but definitely 100 % #trap

      The solution will be something between those extreme cases.

    • User Avatar

      Known As I

      1 week ago

      Technically yes - they will be completely overhauled. But they will look similar.

      The new thing is that charts may be bound to genres or other criteria (like moods or situations). The number and kind for those charts will be moderated (whatever makes sense). As a consequence a single track can appear in several charts.

      If you manage to create a good edm track integrating some guitar skills you might chart in #dance, #rock, #house, #happy at the same time.

      Note to me (thinking loudly): I just found a problem with that: while the association with multiple charts is fine to me, the placement should not be coupled. If your listeners agree that a track is a #1 in #dance it's not necessarily better in #rock. Just counting the plays/likes whatever might be insufficient. One possible solution is to limit the participants in the charts by their associated tag weights. Revisiting my example above: the track can be at maximum: 25 % #dance, 25 % #rock, 25 % #happy, 25 % #house. If other tracks are > 40 % #house (more pure house) they will take precedence over my example track.

      We need to limit the number of tracks which can participate in the charts then.

  • User Avatar

    naswalt

    3 weeks ago

    I like the amount of involvement that this puts on the community for categorizing tracks properly. There will definately be a really rough start on this (I foresee much whining and people not reading this post to understand the change) but after people get used to this it'll be much more stable than the current system.

    I'm all for it and I can't wait to see #rock on audiotool!

    • User Avatar

      naswalt

      1 week ago

      The elders have blessed you with good good

      related tags. May they have mercy on your continued endeavors.

    • User Avatar

      Known As I

      2 weeks ago

      I just asked the elders of the tag library what #rock means to them: #metal, #indie, #guitar, #classic, #alternative, #acoustic, #world, #electric, #string, #synth, #drum, #pulverisateur, #song, #upbeat, #heisenberg, #original, ...

      (that's just based on the status quo of tags assigned by the artists)

    • User Avatar

      Balikna

      3 weeks ago

      #rock ftw

  • User Avatar

    naut

    3 weeks ago

    with disregard to the community driven part, will this necessitate more more involvement from the team/community moderators? Seems complicated to me, but I am still excited to be seeing it in action.

    • User Avatar

      Known As I

      2 weeks ago

      That's quite possible. I believe there will be some some initial work in figuring out how the community reacts and which tools and rules will be needed to compensate for any abuse.

      We can also nominate tag moderators, which simply have a boosted weight for the tags they assign.

  • User Avatar

    FRED. but a CLOWN.

    3 weeks ago

    At least people could finally put trap kids in their place

  • User Avatar

    Reverend Reverb

    3 weeks ago

    So the system against tag trolls kinda works like Trust Factor in CSGO, right?

  • User Avatar

    s_maurice

    3 weeks ago

    Wow this is going to be a massive thing! Still would be a good thing to be able to distinguish between the tags provided by the creator and those from the community I guess.

    I guess that tags (especially in comments) are only relevant to upcoming releases after the function being implemented? One possible option would be parsing comments by a routine generating tags from old comments to be able to include them in a generated playlist where those tracks would otherwise fall out. The parser could be based on a library generated from a list of user generated tags with a specific count >= x. That would probably be lots of extra work I guess but that wouldn't exclude old tracks from user generated taxonomy playlists. Just my two cents, need to think about that more, but I see a lot of potential in this!

    • User Avatar

      G3ar_bot

      3 weeks ago

      welp, mankind can wait :/

    • User Avatar

      Known As I

      3 weeks ago

      @Balikna yes, this should even work as expected :)

    • User Avatar

      Known As I

      3 weeks ago

      Good topic: I recently checked which tags were found to be similar to #lmao. Those were there results: #xd, #lit, #lol, #idk, #oof, #xd, #l m a o, #lit af, #n o i c e, #litaf, #oo, #l i t, #ooo, #xddd, #noice, #lol

      Looks promising - not for mankind, though ;)

    7 more
  • User Avatar

    DrD

    3 weeks ago
  • User Avatar

    Velocistar

    3 weeks ago

    I think enabling tagging after listening to a certain percent of a track is a good thing. It incentivizes users to actually listen to people's music (if they want to tag it), but if the system doesn't recognize when a person skips ahead in a track, it kinda defeats the purpose, since people could still tag troll by skipping ahead like 2 minutes in a track, and tagging it with something stupid or irrelevant.

    • User Avatar

      Known As I

      3 weeks ago

      Those values were just an example. "needs to listen for 30 s or 80 % (whichever comes first)" would solve this.

    • User Avatar

      G3ar_bot

      3 weeks ago

      what happens if a track is under 30 seconds long?

      i.e. shitposts/experiments

    • User Avatar

      Known As I

      3 weeks ago

      I believe 30 seconds of listening would be enough to determine the genre and mood of most tracks. It will be harder to do perform a mass-trolling with this restriction.

      I added the section "What happens if 100 listeners select 1000 different tags?" to address some concerns about stupid/irrelevant tags. (loosely related to you question)

  • User Avatar

    Balikna

    3 weeks ago

    so we, the listeners, get to #tag things ourselves?

    • User Avatar

      Known As I

      3 weeks ago

      Yeah - I'm pretty sure that this will happen.

      I added my suggestions in the section "How to combat tag trolls?"

    • User Avatar

      Jetdarc

      3 weeks ago

      I honestly have a really bad feeling that people will abuse the tagging system

    • User Avatar

      Known As I

      3 weeks ago

      Yes. The artist suggests tags they find appropriate (to have something to start with). The listeners may add more.

00:00
06:03