How suited is MusicBrainz for gathering and presenting subjective data? (such as genre)

Tags: #<Tag:0x00007f510ef84470> #<Tag:0x00007f510ef84330> #<Tag:0x00007f510ef841a0>

There is one aspect about managing music in the digital domain that never stops bugging me.
It’s what is called ‘genre’.

And to relieve my own brain a little bit, I thought to bother other people with it, here.

One thing that MusicBrainz is very good, smart and wise at, is trying to get the facts straight. Or at least get them as straight as possible.
That’s not up for too much debate where it concerns items such as: artist names, instruments, dates, locations, publishers, etc.
And I have learned to have great faith and trust that MusicBrainz offers reliable results in all these areas.

But MusicBrainz also has the option to enter and store a much more subjective item as ‘genre’.
And compared to the items I named before, the ‘genre’ category is the one I don’t consider to be very complete or reliable.

This is not me critisizing the people that set and decide on the genre list, nor is it me critisizing the editors that are making an effort to populate these ‘genre’ relationships.

But I am wondering:

  • Can it be expected of all (or even the slightest majority of) editors that they are willing and/or able to add complete and sensible ‘genre’ data to the releases that they enter?
  • Are the end-users that do have a good knowledge and understanding of the music involved, and have sensible opinions on what music genres could be applied to them, actually adding to, and improving on MusicBrainz’ database in this respect?

To be honest, I have severe doubts on both points.

Far as I can tell this is not causing any real problems. But it makes me wonder:

  1. What are the deeper thoughts and goals behind MusicBrainz having a subjective category such as ‘genre’ in the first place?
  2. Assuming that it was understood and accepted that at first database entry these ‘genre’ relationships would not get populated in the most satisfactory way (if even at all), how was it anticipated that end-users with valuable knowledge and opinions on music genres would improve on this database?
2 Likes

Big dump of unvarnished thoughts incoming :smiley:
Some of these @hiccup has already acknowledged but I’m putting everything in.

Editors of MB are very used to working towards identifying and applying ‘concrete’ data, and as a result this tends to represent the ‘perfect’ data model for many here. The reality is that there are many valuable and useable types of data, including subjective and messy. That a band of X name that was formed in X year is great info. But comments on the bands last.fm page may well be what a future music historian, trying to paint a picture of the band, would salivate over. Put both together and we have a delicious hot info broth.

I see these this attitude as counter-productive. It implies that some people are more qualified to have an opinion on music - valid in some cases, but related to an element that we acknowledge to be at least partially subjective? When it comes to sorting your music, if the plebian population (presumably most of them) would classify something as ‘jazz’, is it really useful to have a jazz expert step in and say ‘no it isn’t’ - particularly when the DB is here for everyone to tag and sort their music, not just experts?

To illustrate my thoughts a little bit more, I think these are both equally valid ways to approach the problem:

image

A: A small group of experts enters genre tags. ‘Accuracy’ (however you decide to measure that) is very high, with few outliers.
B: A large group of the public enters genres. We have a much larger dataset, with a lot of outliers. We identify and remove the outliers.

I would say that I prefer the second, but I’m sure it depends on the user and what they need out of the data or out of the tags. I personally don’t have confidence those ‘experts’ are going to get round to classifying my noisegrind releases. Classical editors with popular releases/artists might prefer the more stringent method.

Last.fm is an example of method B, with a system hugely open to abuse. Junk tags are common. So why is my last.fm ‘genres listened to’ list so accurate? It uses group consensus to discard (not display/assign) the outliers. I find last.fm much more accurate an comprehensive than RYM, for instance.

MusicBrainz also currently has option B, with the tag/genre voting system assigning importance to a tag via a higher number. I have seen it to be effective so far.

imo large datasets are only useful if you can group, display or analyse sets of data. Genre is one of the most common (if subjective) groupings for music, so it seems a logical step. On the more practical side, tagging with genre is a major/common user need, and someone was interested enough to code the feature.

I would love to see some more features added, particularly auto-grouping of different spellings of a genre (https://tickets.metabrainz.org/browse/MBS-10062), and perhaps a genre tree. But I think the current implementation is actually excellent at the basic function.

2 Likes

I understand what you mean about genre being too much of a personal opinion. What one person sees as folk-punk someone else will see as a noisy racket. Someone within Jazz will see all the fine differences with the styles, but someone from outside will stick anything with a saxophone in. Some of the ways I see Pink Floyd categorised is weird (had Arty-Rock(?) come up when ripping a Gilmour album at the weekend).

I personally think one of the better ways of genre tagging is how it happens now. Having that score alongside is useful. If 25 people think it is “Category A” and one person thinks “Category B” then it allows an algorithm like aerozol mentions to skim off a narrower result.

Though I don’t think the person who said category B should be ignored. It may upset someone from outside of Jazz having something referred to as Jazz, but that is how an outsider may well see it. And from there they may find real Jazz.

This also allows categories like “Rock” to dominate on a album with “Blues Rock” to be a sub-category with a lesser score.

Finding a release date for an album is easy as there is only (usually) one answer. Nailing a genre will always be opinion and there is no “right” just “most people think”.

Oh - and certainly should not be just a small team of “experts” doing genre. The wider the opinions the better.

1 Like

Why not? You can add what you think is the genre, and it will be upvoted or downvoted by others.
In this matter, to be frank, that’s the only way. Sometimes a band says they play “doom metal” and I often disagree, because well it’s not even “metal” (according to myself). But who can give an exact definition of a musical genre?
So MB went for a voting system, after, literally, years of discussions.

But why would an editor not add genres he thinks the music is? That’s exactly what I could call counter-productive.
If you think that’s jazz, then tag it as jazz. If 10 other editors upvote it, well, you were right, if no one votes, well, no one cares, if 10 editors vote for another genre, well, you were prolly wrong, and perhaps you can remove the genre you set.

Also, anything related to a small set of experts is void, because it requires millions of people to set genres on each recording produced in a year…

MusicBrainz team has to improve the genre feature in various ways:

  • aliases (rock’n’roll vs rock and roll …)
  • relationships (blues rock is related to blues and rock)
  • disambiguation (some genres spelled the same way, are in fact different, depending on the context)

But what is existing now is working pretty well. I’m trying to add genres based on artist/label qualification for the music (including Bandcamp tags, because those are set by the label or the artist themselves).
Other databases (metalarchives or wikipedia) are good sources too.

Also putting genres to Pink Floyd music is very easy, because it’s around since a long time, with millions of people listening and defining it. But genres are also evolving over the time, perhaps in few years it would be qualified as “traditional progressive rock” or “old-school stoner blues rock” or whatever, who knows.

5 Likes

This is where I :heart: the ‘loose’ system. That 100 people classified something as jazz tells me a lot about the music. That another 40 went on to classify it as free jazz gives me another interesting data point. But those 7 votes for electro pop?? Now we’re talking!

You can start to see where the music might sit in a spectrum, with (often) multiple influences. Even if it’s somehow ‘wrong’, I want to know that a minority thought an album sounded like electro pop.

Haha, I’ve heard “we don’t really like to box ourselves in, we don’t really have a genre” from bands so often. Or “we mix influences like jazz and soul and blues with extreme metal”. Then they play one song and, yup, that’s definitely just post-rock (or whatever). Genre is one of the few (only?) instances where I think the artist shouldn’t have any more say than Bob off the street.

2 Likes