Recommend for a group of users

[vycon <we...@gmail.com>]

Hi experts,

Instead of taking user id as an input, I am thinking whether mahout will
support demographic information as input, e.g., gender, age and occupation
like in movielens-100k dataset. In this case, I can recommend movie to a
group of similar people. 

I think I need to implement my own user similarity, but the method in mahout
is taking two user id as input.

Any experts can give me a hint on this?

Thanks

--
View this message in context: http://lucene.472066.n3.nabble.com/Recommend-for-a-group-of-users-tp2994342p2994342.html
Sent from the Mahout User List mailing list archive at Nabble.com.

Re: Recommend for a group of users

[Sean Owen <sr...@gmail.com>]

Yes, you can implement your own UserSimilarity metric using whatever
information you want. That could easily be based on factors like gender or
age. It's possible, but entirely up to you. Yes, it takes two user IDs as
input, since that is the question it answers: how similar are two users? Of
course, your implementation will also read from other sources of information
in order to return an answer.

It plugs into a GenericUserBasedRecommender then, and you recommend items to
users via user ID. That does not change.

On Fri, May 27, 2011 at 9:37 PM, vycon <we...@gmail.com> wrote:

> Hi experts,
>
> Instead of taking user id as an input, I am thinking whether mahout will
> support demographic information as input, e.g., gender, age and occupation
> like in movielens-100k dataset. In this case, I can recommend movie to a
> group of similar people.
>
> I think I need to implement my own user similarity, but the method in
> mahout
> is taking two user id as input.
>
> Any experts can give me a hint on this?
>
> Thanks
>
> --
> View this message in context:
> http://lucene.472066.n3.nabble.com/Recommend-for-a-group-of-users-tp2994342p2994342.html
> Sent from the Mahout User List mailing list archive at Nabble.com.
>

Re: Recommend for a group of users

["petar.mitrovic" <pe...@gmail.com>]

Hi vycon,

I am not an expert in Mahout, but seems like you could represent user
demographic data as vectors (age, gender, occupation, etc), weighting the
values eventually and returning the distance between these vectors from your
own implementation of UserSimilarity.

Regards,
Petar

--
View this message in context: http://lucene.472066.n3.nabble.com/Recommend-for-a-group-of-users-tp2994342p3554833.html
Sent from the Mahout User List mailing list archive at Nabble.com.