mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Charly Lizarralde <charly.lizarra...@gmail.com>
Subject Re: Few Users, Too Many Items
Date Thu, 09 Dec 2010 13:31:07 GMT
Thanks a lot for the answers!

On Wed, Dec 8, 2010 at 7:33 AM, Steven Bourke <sbourke@gmail.com> wrote:

> If the matrix is to sparse collab filtering just dont work. Try some of the
> cluster based recommendation techniques in mahout if you really don't want
> to use content based.
>
>
> On Wed, Dec 8, 2010 at 8:18 AM, Ted Dunning <ted.dunning@gmail.com> wrote:
>
> > You can do a quick SVD on the matrix, but if you don't have any
> significant
> > overlap, this is very likely to fail.
> >
> > If you augment this matrix with content you might have a very interesting
> > thing to do recommendations with.  Essentially, you would be recommending
> > meta-data items.  An example of this is if you have a music
> recommendation
> > problem, you might do better especially with small data to recommend
> > artists
> > instead of songs.  From the artists, you can then recommend the songs
> that
> > artist did.
> >
> > On Tue, Dec 7, 2010 at 6:05 PM, Charly Lizarralde <
> > charly.lizarralde@gmail.com> wrote:
> >
> > > Hi, I'm currently working on a small dataset ( about 16k ratings) with
> > only
> > > 1500 users and 9800 items. Most users do not have a rich neighbourhood
> > and
> > > item similarity does not work either as items are rated only a couple
> of
> > > times.
> > >
> > > Is there any suggestion on how we can adresss this issue besides
> Content
> > > Based Recommendations?
> > >
> > > Regards,
> > > Charly
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message