Mahout and PCA for a music analysis tool

[Nicolas Pied <ni...@gmail.com>]

> Hello,
> 
> I would like to implement an application like Like.fm / Pandora (but more simple) that suggests musics close to a given one.
> 
> I think extract a lot of information for each music file like the mood, the BPM, the style, etc. in order to define a matrice with a lot of columns.
> 
> I know that the PCA algorithm can be a solution to implement such a thing, but do you know how to define it with Mahout?
> 
> Maybe is there a better algorithm that the PCA too?
> 
> Thank in advance for your replies.
> 
> Regards,

Nicolas

Re: Mahout and PCA for a music analysis tool

[Ted Dunning <te...@gmail.com>]

Now that I have been all negative, if you want to go developing auditory
features, look up music information retrieval.  The ISMIR conferences have
a wealth of information.

http://www.ismir.net/

On Tue, Apr 17, 2012 at 11:03 AM, Nicolas Pied <ni...@gmail.com>wrote:

> > Hello,
> >
> > I would like to implement an application like Like.fm / Pandora (but
> more simple) that suggests musics close to a given one.
> >
> > I think extract a lot of information for each music file like the mood,
> the BPM, the style, etc. in order to define a matrice with a lot of columns.
> >
> > I know that the PCA algorithm can be a solution to implement such a
> thing, but do you know how to define it with Mahout?
> >
> > Maybe is there a better algorithm that the PCA too?
> >
> > Thank in advance for your replies.
> >
> > Regards,
>
> Nicolas
>
>

Re: Mahout and PCA for a music analysis tool

[Ted Dunning <te...@gmail.com>]

If you really want to recommend music that people will like, you have to
start from the realization that most of musical appreciation is social, not
auditory.  This has been substantiated in controlled tests where as much as
60% of appreciation was driven by very weak social cues in a test.  In my
experience, the social evidence also reflects a vast amount of the acoustic
information as well so that even in a case where the response is acoustic,
you can derive that preference if you have good social signals.

This correlates very well with my experience with building large music
(MusicMatch) and video (Veoh) recommendation systems.  It also correlates
with the hallway conversations I have had with other folks who have done
similar things at last and Pandora.

In addition, if you are building a streaming recommender, you have to worry
a lot about injecting reasonable amounts diversity into the stream to
maintain user interest.  This is exactly at odds with what most auditory
features will lead you to do.

My summary of the situation is that for most purposes, automated or manual
musical feature extraction is a complete waste of time in terms of
recommending music.  It can be fun, but it isn't likely to be very
profitable. You would generally do much better putting your time into
negotiating deals to have the right to place the music in the stream and
building a better user experience so that you get better social data.



On Tue, Apr 17, 2012 at 11:03 AM, Nicolas Pied <ni...@gmail.com>wrote:

> > I would like to implement an application like Like.fm / Pandora (but
> more simple) that suggests musics close to a given one.
> >
> > I think extract a lot of information for each music file like the mood,
> the BPM, the style, etc. in order to define a matrice with a lot of columns.
> >
> > I know that the PCA algorithm can be a solution to implement such a
> thing, but do you know how to define it with Mahout?
> >
> > Maybe is there a better algorithm that the PCA too?
> >
> > Thank in advance for your replies.
> >
> > Regards,
>
> Nicolas
>
>

Re: Mahout and PCA for a music analysis tool

[Nicolas Pied <ni...@gmail.com>]

Is it a question ;-)

Thanks

Nicolas

Le 17 avr. 2012 à 20:42, Ted Dunning <te...@gmail.com> a écrit :

> Nicolas,
> 
> Are you replying to this?  Or asking these questions?
> 
> On Tue, Apr 17, 2012 at 11:03 AM, Nicolas Pied <ni...@gmail.com>wrote:
> 
>>> Hello,
>>> 
>>> I would like to implement an application like Like.fm / Pandora (but
>> more simple) that suggests musics close to a given one.
>>> 
>>> I think extract a lot of information for each music file like the mood,
>> the BPM, the style, etc. in order to define a matrice with a lot of columns.
>>> 
>>> I know that the PCA algorithm can be a solution to implement such a
>> thing, but do you know how to define it with Mahout?
>>> 
>>> Maybe is there a better algorithm that the PCA too?
>>> 
>>> Thank in advance for your replies.
>>> 
>>> Regards,
>> 
>> Nicolas
>> 
>> 

Re: Mahout and PCA for a music analysis tool

[Ted Dunning <te...@gmail.com>]

Nicolas,

Are you replying to this?  Or asking these questions?

On Tue, Apr 17, 2012 at 11:03 AM, Nicolas Pied <ni...@gmail.com>wrote:

> > Hello,
> >
> > I would like to implement an application like Like.fm / Pandora (but
> more simple) that suggests musics close to a given one.
> >
> > I think extract a lot of information for each music file like the mood,
> the BPM, the style, etc. in order to define a matrice with a lot of columns.
> >
> > I know that the PCA algorithm can be a solution to implement such a
> thing, but do you know how to define it with Mahout?
> >
> > Maybe is there a better algorithm that the PCA too?
> >
> > Thank in advance for your replies.
> >
> > Regards,
>
> Nicolas
>
>