[mb-devel] GSoC 2012 Proposal for Genre Integration to MusicBrainz

[Saurav]

Hello all,

I am Saurav, and I have a idea proposal for MusicBrainz which can be done
as a GSoC project.
The idea is to incorporate the attribute 'genre' for the Music Data
available at MusicBrainz.
For this, the tag universe (the existing tag data) is used as a base.


The "Genre List" is taken from [1] http://musicgenreslist.com/ , which
is the 3rd
result for "Official List of Music Genres" (1st and 2nd being Wikipedia.com).
This arguably means that this page is the "most popular" all over the
world, when
it comes to people referring to Music Genres' List. Of course, genre
classification
CAN be different on a more personal scale, but that is taken into consideration
by labelling an item to a particular genre solely on the basis of
People's choice.
This will become clear very soon.

Taking the list from [1], we now have a fixed list of genres. Also we
have the tag
universe, which will now be used to map items (artists, releases,
works) to their
respective genres. A particular entity will be assigned to a genre
(and sub-genre,
if there is one in our genre-list) depending upon the most voted tag associated
with it. e.g. a Release group "Wish You Were Here" has the tags:

progressive rock 6
rock 6
1975 1
classic pop and rock 1
prog rock 1
psychadelic 1
psychedelic rock 1
should be public domain 1

So this recording will be assigned to genre:: Rock>>Prog-Rock/Art Rock
When enough information is not available for sub genre classification,
the classification
would be(e.g):: Rock

Note that the classification system will be intelligent (Rock, ROCK,
Progressive,
PROgresiv etc. will be identified correctly)

So much for 'existing' tags(and assignment of genres for them). Now
the case when
the user is entering new tags:

Case 1:: Entities which haven't been genre-fied as yet:
When a user edits, an auto-completion feature will show results from
our Genre-List
and prompt him to enter a genre (this ensures that it is the users who
decide genre).
The user will choose one of the genres (and sub-genre too depending
upon how much he
knows). The user will also be able to enter additonal tags or vote on
existing tags, as before.

Case 2:: Entities which have been genre-fied before (by the user or
through our previously described scheme):
When the user edits, he will NOT be prompted for entering the genre.
He can only enter new
tags or vote on existing tags. The genre of the entity will change/not
depending on
how many votes the corresponding tags have and which is the currently
most preferred tag
(for genre). This ensures that the 'most-popular' tag is always chosen
for being called the
genre for the entity (and no additional 'power' is given to any one user).

e.g Suppose the tags for a particular entity (genre presently is pop)
were as follows:

rock 3
pop 4

Suppose now, two users voted for rock. The genre for the entity
changes to rock, as rock now
has 5 votes.

Note however that the genre classification depends only upon the tag
availability for an
entity (at least for the project in the summer). More work on this can
be looked into as
a future plan of action (having other attributes as well for entities
like beat-count, tempo
etc. which can be a more accurate measurer of genre). I am willing to
work in this even after summer.

Suggestions/Improvements?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.musicbrainz.org/pipermail/musicbrainz-devel/attachments/20120331/b8901187/attachment.htm

[Nine99]

I wouldn't call this list comprehensive. Also, new genres appear all the time. We should take the genres from Wikipedia, and edit (use the most often used version etc.).

- Nine99

-------- Original-Nachricht --------

Datum: Sat, 31 Mar 2012 09:49:56 +0530
Von: Saurav <online.saurav at gmail.com>
An: musicbrainz-devel at lists.musicbrainz.org
Betreff: [mb-devel] GSoC 2012 Proposal for Genre Integration to MusicBrainz
Hello all,
I am Saurav, and I have a idea proposal for MusicBrainz which can be done
as a GSoC project.
The idea is to incorporate the attribute 'genre' for the Music Data
available at MusicBrainz.
For this, the tag universe (the existing tag data) is used as a base.
The "Genre List" is taken from [1] http://musicgenreslist.com/ , which
is the 3rd
result for "Official List of Music Genres" (1st and 2nd being
Wikipedia.com).
This arguably means that this page is the "most popular" all over the
world, when
it comes to people referring to Music Genres' List. Of course, genre
classification
CAN be different on a more personal scale, but that is taken into consideration
by labelling an item to a particular genre solely on the basis of
People's choice.
This will become clear very soon.
Taking the list from [1], we now have a fixed list of genres. Also we
have the tag
universe, which will now be used to map items (artists, releases,
works) to their
respective genres. A particular entity will be assigned to a genre
(and sub-genre,
if there is one in our genre-list) depending upon the most voted tag associated
progressive rock 6
rock 6
1975 1
classic pop and rock 1
prog rock 1
psychadelic 1
psychedelic rock 1
should be public domain 1
So this recording will be assigned to genre:: Rock>>Prog-Rock/Art Rock
When enough information is not available for sub genre classification,
the classification
would be(e.g):: Rock
Note that the classification system will be intelligent (Rock, ROCK,
Progressive,
PROgresiv etc. will be identified correctly)
So much for 'existing' tags(and assignment of genres for them). Now
the case when
When a user edits, an auto-completion feature will show results from
our Genre-List
and prompt him to enter a genre (this ensures that it is the users who
decide genre).
The user will choose one of the genres (and sub-genre too depending
upon how much he
knows). The user will also be able to enter additonal tags or vote on
existing tags, as before.
Case 2:: Entities which have been genre-fied before (by the user or
When the user edits, he will NOT be prompted for entering the genre.
He can only enter new
tags or vote on existing tags. The genre of the entity will change/not
depending on
how many votes the corresponding tags have and which is the currently
most preferred tag
(for genre). This ensures that the 'most-popular' tag is always chosen
for being called the
genre for the entity (and no additional 'power' is given to any one user).
e.g Suppose the tags for a particular entity (genre presently is pop)
rock 3
pop 4
Suppose now, two users voted for rock. The genre for the entity
changes to rock, as rock now
has 5 votes.
Note however that the genre classification depends only upon the tag
availability for an
entity (at least for the project in the summer). More work on this can
be looked into as
a future plan of action (having other attributes as well for entities
like beat-count, tempo
etc. which can be a more accurate measurer of genre). I am willing to
work in this even after summer.
Suggestions/Improvements?

[Saurav]

Hi Nine99,

Sorry for the delay in replying. Had no internet access yesterday due to a
rare power failure in our area.

Are you referring to
http://en.wikipedia.org/wiki/List_of_popular_music_genres ?
This seems to be a more comprehensive list definitely, and is also
classified properly into genres and sub-genres (and more levels too). This
can be just perfect for our purposes.

So, the source of the genre-list can be from wikipedia. The rest of the
idea is similar to before.

Any ideas anyone?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.musicbrainz.org/pipermail/musicbrainz-devel/attachments/20120402/b9067f2a/attachment.htm

[Frederic Da Vitoria]

2012/3/31 Saurav <online.saurav at gmail.com>

Post by Saurav
Hello all,
I am Saurav, and I have a idea proposal for MusicBrainz which can be done
as a GSoC project.
The idea is to incorporate the attribute 'genre' for the Music Data
available at MusicBrainz.
For this, the tag universe (the existing tag data) is used as a base.
The "Genre List" is taken from [1] http://musicgenreslist.com/ , which is the 3rd
result for "Official List of Music Genres" (1st and 2nd being Wikipedia.com).
This arguably means that this page is the "most popular" all over the world, when
it comes to people referring to Music Genres' List. Of course, genre classification
CAN be different on a more personal scale, but that is taken into consideration
by labelling an item to a particular genre solely on the basis of People's choice.
This will become clear very soon.
Taking the list from [1], we now have a fixed list of genres. Also we have the tag
universe, which will now be used to map items (artists, releases, works) to their
respective genres. A particular entity will be assigned to a genre (and sub-genre,
if there is one in our genre-list) depending upon the most voted tag associated
progressive rock 6
rock 6
1975 1
classic pop and rock 1
prog rock 1
psychadelic 1
psychedelic rock 1
should be public domain 1
So this recording will be assigned to genre:: Rock>>Prog-Rock/Art Rock
When enough information is not available for sub genre classification, the classification
would be(e.g):: Rock
Note that the classification system will be intelligent (Rock, ROCK, Progressive,
PROgresiv etc. will be identified correctly)
So much for 'existing' tags(and assignment of genres for them). Now the case when
When a user edits, an auto-completion feature will show results from our Genre-List
and prompt him to enter a genre (this ensures that it is the users who decide genre).
The user will choose one of the genres (and sub-genre too depending upon how much he
knows). The user will also be able to enter additonal tags or vote on existing tags, as before.
When the user edits, he will NOT be prompted for entering the genre. He can only enter new
tags or vote on existing tags. The genre of the entity will change/not depending on
how many votes the corresponding tags have and which is the currently most preferred tag
(for genre). This ensures that the 'most-popular' tag is always chosen for being called the
genre for the entity (and no additional 'power' is given to any one user).
rock 3
pop 4
Suppose now, two users voted for rock. The genre for the entity changes to rock, as rock now
has 5 votes.
Note however that the genre classification depends only upon the tag availability for an
entity (at least for the project in the summer). More work on this can be looked into as
a future plan of action (having other attributes as well for entities like beat-count, tempo
etc. which can be a more accurate measurer of genre). I am willing to work in this even after summer.
Suggestions/Improvements?

Hello,

I am not a developer, but I'd like to comment from my user's point of view.

There are 2 parts in your proposition, one I like, and one which I don't. I
like the idea of normalizing genres and I believe there are other users who
wish there was a way to apply some order to folksonomy tags. But I don't
think single genre can work, too many works are borderline. Take any
borderline work or recording, there are users who would want to set it to
genre A, others to genre B and still others who would want both. I am a
member of the third group. If a work or recording is both classical and
jazz, I'd like to see it both in classical and jazz lists. If a work is
both rock and jazz, I'd wand to get it in rock and in jazz lists. Of
course, for many works or recordings, there could even be more than 2
genres.

Furthermore, there are genres and sub-genres. If I set a recording to rock,
then I am not using a more specific genre which could apply to this
recording. OTOH, if I use this sub-genre, I am "hiding" this recording for
users who may not be aware of the intricacies of genre hierarchies.

[Per Øyvind Øygard]

Genres were discussed at last year's MBz Summit, so I'll just briefly
repeat my thoughts here.

The problem with a straight voting system is that it promotes generic
genres. Knowing the precise genres of a release is not necessarily
that easy since the borders are quite fluid (jungle vs. drum & bass),
genres can pop up over night (chilllwave), or just be too
small/obscure that the editor might not have heard of it. The result
will be that everything is tagged as electronic, rock, pop, or equally
meaningless/wrong classifications (see last.fm).

A better solution is to copy the Discogs-approach of genres and
styles. For instance Kind of Blue[1] is classified as Jazz
(obviously), of Modal style (I don't even know what Modal jazz is, but
it sounds right). On last.fm it's tagged as: "jazz, cool jazz, albums
i own, favorite albums, modal jazz" To take another example, Bj?rk's
Post[2] is classified as Electronic, of Downtempo and Trip-Hop style,
while on last.fm it's "albums i own, electronic, alternative, female
vocalists, 90s".

Now Discogs' approach isn't necessarily perfect, but that's in my
opinion just a side-effect of dealing with genres which in general are
neither objective nor absolute. My suggestion would be to make main
genres be a regular edit (whether something is jazz, rock, or pop,
isn't particularly subjective), and styles be some kind of normalized
free-form voting system. Styles change too quickly to lock them down
to a fixed list, though obviously we can herd users into commonly used
styles and make free-form an advanced field somewhere. It probably
wouldn't hurt to seed the list with genres, though IMO that list
should be taken from Wikipedia since it's by far the easiest way of
determining whether a genre has sufficient notability.

Anyway, the biggest challenge with genres is going to be getting
buy-in from devs and style-people since this is kind of a big deal,
and something that most people are going to have opinions on. There's
definitely a lot of interest in it though, if you can pull it off.

- Per / Wizzcat

[1]: http://www.discogs.com/Miles-Davis-Kind-Of-Blue/master/5460
[2]: http://www.discogs.com/Bj?rk-Post/master/36251

Post by Saurav
Hello all,
I am Saurav, and I have a idea proposal for MusicBrainz which can be done as
a GSoC project.
The idea is to incorporate the attribute 'genre' for the Music Data
available at MusicBrainz.
For this, the tag universe (the existing tag data) is used as a base.
The "Genre List" is taken from [1] http://musicgenreslist.com/ , which is
the 3rd
result for "Official List of Music Genres" (1st and 2nd being
Wikipedia.com).
This arguably means that this page is the "most popular" all over the world,
when
it comes to people referring to Music Genres' List. Of course, genre
classification
CAN be different on a more personal scale, but that is taken into consideration
by labelling an item to a particular genre solely on the basis of People's
choice.
This will become clear very soon.
Taking the list from [1], we now have a fixed list of genres. Also we have
the tag
universe, which will now be used to map items (artists, releases, works) to
their
respective genres. A particular entity will be assigned to a genre (and
sub-genre,
if there is one in our genre-list) depending upon the most voted tag associated
progressive rock 6
rock 6
1975 1
classic pop and rock 1
prog rock 1
psychadelic 1
psychedelic rock 1
should be public domain 1
So this recording will be assigned to genre:: Rock>>Prog-Rock/Art Rock
When enough information is not available for sub genre classification, the
classification
would be(e.g):: Rock
Note that the classification system will be intelligent (Rock, ROCK,
Progressive,
PROgresiv etc. will be identified correctly)
So much for 'existing' tags(and assignment of genres for them). Now the case
when
When a user edits, an auto-completion feature will show results from our
Genre-List
and prompt him to enter a genre (this ensures that it is the users who
decide genre).
The user will choose one of the genres (and sub-genre too depending upon how
much he
knows). The user will also be able to enter additonal tags or vote on
existing tags, as before.
Case 2:: Entities which have been genre-fied before (by the user or through
When the user edits, he will NOT be prompted for entering the genre. He can
only enter new
tags or vote on existing tags. The genre of the entity will change/not
depending on
how many votes the corresponding tags have and which is the currently most
preferred tag
(for genre). This ensures that the 'most-popular' tag is always chosen for
being called the
genre for the entity (and no additional 'power' is given to any one user).
e.g Suppose the tags for a particular entity (genre presently is pop) were
rock 3
pop 4
Suppose now, two users voted for rock. The genre for the entity changes to
rock, as rock now
has 5 votes.
Note however that the genre classification depends only upon the tag
availability for an
entity (at least for the project in the summer). More work on this can be
looked into as
a future plan of action (having other attributes as well for entities like
beat-count, tempo
etc. which can be a more accurate measurer of genre). I am willing to work
in this even after summer.
Suggestions/Improvements?
_______________________________________________
MusicBrainz-devel mailing list
MusicBrainz-devel at lists.musicbrainz.org
http://lists.musicbrainz.org/mailman/listinfo/musicbrainz-devel

[Chad Wilson]

For reference, luks has some interesting ideas about how this could be
done discussed at the below links. I felt there was somewhere to go with
this; but of course understood luks possibly feeling dissuaded by the
indecision or mixed feelings in the community.

To be honest though, without prior community agreement, I don't
personally feel this is realistic as a GSoC project unless the idea is
to just prototype it to avoid the chicken+egg reaction that can boil
down to "i hate it because i don't know how it could work". Sometimes if
you show how it could work people change their minds :)

Anyway, getting it live I think requires a lot of Style and community
buy-in, and working to gain that isn't a development task per se. :-/

http://forums.musicbrainz.org/viewtopic.php?id=3189
http://musicbrainz.1054305.n4.nabble.com/Genre-support-td4186483.html
http://tickets.musicbrainz.org/browse/MBS-3738

Chad / voiceinsideyou

Post by Per Øyvind Øygard
Genres were discussed at last year's MBz Summit, so I'll just briefly
repeat my thoughts here.
The problem with a straight voting system is that it promotes generic
genres. Knowing the precise genres of a release is not necessarily
that easy since the borders are quite fluid (jungle vs. drum& bass),
genres can pop up over night (chilllwave), or just be too
small/obscure that the editor might not have heard of it. The result
will be that everything is tagged as electronic, rock, pop, or equally
meaningless/wrong classifications (see last.fm).
A better solution is to copy the Discogs-approach of genres and
styles. For instance Kind of Blue[1] is classified as Jazz
(obviously), of Modal style (I don't even know what Modal jazz is, but
it sounds right). On last.fm it's tagged as: "jazz, cool jazz, albums
i own, favorite albums, modal jazz" To take another example, Bj?rk's
Post[2] is classified as Electronic, of Downtempo and Trip-Hop style,
while on last.fm it's "albums i own, electronic, alternative, female
vocalists, 90s".
Now Discogs' approach isn't necessarily perfect, but that's in my
opinion just a side-effect of dealing with genres which in general are
neither objective nor absolute. My suggestion would be to make main
genres be a regular edit (whether something is jazz, rock, or pop,
isn't particularly subjective), and styles be some kind of normalized
free-form voting system. Styles change too quickly to lock them down
to a fixed list, though obviously we can herd users into commonly used
styles and make free-form an advanced field somewhere. It probably
wouldn't hurt to seed the list with genres, though IMO that list
should be taken from Wikipedia since it's by far the easiest way of
determining whether a genre has sufficient notability.
Anyway, the biggest challenge with genres is going to be getting
buy-in from devs and style-people since this is kind of a big deal,
and something that most people are going to have opinions on. There's
definitely a lot of interest in it though, if you can pull it off.
- Per / Wizzcat
[1]: http://www.discogs.com/Miles-Davis-Kind-Of-Blue/master/5460
[2]: http://www.discogs.com/Bj?rk-Post/master/36251

Post by Saurav
Hello all,
I am Saurav, and I have a idea proposal for MusicBrainz which can be done as
a GSoC project.
The idea is to incorporate the attribute 'genre' for the Music Data
available at MusicBrainz.
For this, the tag universe (the existing tag data) is used as a base.
The "Genre List" is taken from [1] http://musicgenreslist.com/ , which is
the 3rd
result for "Official List of Music Genres" (1st and 2nd being
Wikipedia.com).
This arguably means that this page is the "most popular" all over the world,
when
it comes to people referring to Music Genres' List. Of course, genre
classification
CAN be different on a more personal scale, but that is taken into consideration
by labelling an item to a particular genre solely on the basis of People's
choice.
This will become clear very soon.
Taking the list from [1], we now have a fixed list of genres. Also we have
the tag
universe, which will now be used to map items (artists, releases, works) to
their
respective genres. A particular entity will be assigned to a genre (and
sub-genre,
if there is one in our genre-list) depending upon the most voted tag associated
progressive rock 6
rock 6
1975 1
classic pop and rock 1
prog rock 1
psychadelic 1
psychedelic rock 1
should be public domain 1
So this recording will be assigned to genre:: Rock>>Prog-Rock/Art Rock
When enough information is not available for sub genre classification, the
classification
would be(e.g):: Rock
Note that the classification system will be intelligent (Rock, ROCK,
Progressive,
PROgresiv etc. will be identified correctly)
So much for 'existing' tags(and assignment of genres for them). Now the case
when
When a user edits, an auto-completion feature will show results from our
Genre-List
and prompt him to enter a genre (this ensures that it is the users who
decide genre).
The user will choose one of the genres (and sub-genre too depending upon how
much he
knows). The user will also be able to enter additonal tags or vote on
existing tags, as before.
Case 2:: Entities which have been genre-fied before (by the user or through
When the user edits, he will NOT be prompted for entering the genre. He can
only enter new
tags or vote on existing tags. The genre of the entity will change/not
depending on
how many votes the corresponding tags have and which is the currently most
preferred tag
(for genre). This ensures that the 'most-popular' tag is always chosen for
being called the
genre for the entity (and no additional 'power' is given to any one user).
e.g Suppose the tags for a particular entity (genre presently is pop) were
rock 3
pop 4
Suppose now, two users voted for rock. The genre for the entity changes to
rock, as rock now
has 5 votes.
Note however that the genre classification depends only upon the tag
availability for an
entity (at least for the project in the summer). More work on this can be
looked into as
a future plan of action (having other attributes as well for entities like
beat-count, tempo
etc. which can be a more accurate measurer of genre). I am willing to work
in this even after summer.
Suggestions/Improvements?