Hi,

I am trying some experiments (mahout 0.5) with a new neighbourhood
model that includes implicit feedback as well as explicit feedback.
And I wanted to know, what are the clIasses, etc that I need to
modify.

To start, my input to the system should in the following format:

userid, itemid, exp_feedback, imp_feedback

where userid and itemid are integers and exp_feedback and imp_feedback
are floats between 0.0 and 1.0
for example:

101, 1, 0.67, 0.18

I've used the recommender system code in mahout before. But it will be
my first time trying to modify it in order to run some custom code. I
have skimmed through the Mahout in Action book.

The other point is that I want to make this modifications with as
little changes to the mahout distribution so that I can benefit from
the optimisations that are present in the distribution.

Definitely I need to have custom UserSimilarity and I need a custom
UserNeighborhood? What else? Can somebody please give me some pointers
where to start, e.g which classes i need to modify, etc. and what are
the things I should be wary of? Then I will have a go and report back
my progress.
Regards
Gawesh

You almost certainly need to modify a DataModel too to expose this extra
information.
But that and your proposed modification seems like all you need to worry
about. The rest should work as-is.
I would use code from Subversion, not just 0.5.

On Sat, Jan 28, 2012 at 4:42 PM, gj <[EMAIL PROTECTED]> wrote:

> Hi,
>
> I am trying some experiments (mahout 0.5) with a new neighbourhood
> model that includes implicit feedback as well as explicit feedback.
> And I wanted to know, what are the clIasses, etc that I need to
> modify.
>
> To start, my input to the system should in the following format:
>
> userid, itemid, exp_feedback, imp_feedback
>
> where userid and itemid are integers and exp_feedback and imp_feedback
> are floats between 0.0 and 1.0
> for example:
>
> 101, 1, 0.67, 0.18
>
> I've used the recommender system code in mahout before. But it will be
> my first time trying to modify it in order to run some custom code. I
> have skimmed through the Mahout in Action book.
>
> The other point is that I want to make this modifications with as
> little changes to the mahout distribution so that I can benefit from
> the optimisations that are present in the distribution.
>
> Definitely I need to have custom UserSimilarity and I need a custom
> UserNeighborhood? What else? Can somebody please give me some pointers
> where to start, e.g which classes i need to modify, etc. and what are
> the things I should be wary of? Then I will have a go and report back
> my progress.
>
>
> Regards
> Gawesh
>

Thanks Sean. My Java is very rusty. So, I'm brushing up on that before
diving in. :-)

I guess I also need to modify the PreferenceArray because there will
two preferences (explicit and implicit) for each item per user.

Gawesh
ps: Sean, looking forward to your talk at London RecSys.

On Sat, Jan 28, 2012 at 6:47 PM, Sean Owen <[EMAIL PROTECTED]> wrote:
> You almost certainly need to modify a DataModel too to expose this extra
> information.
> But that and your proposed modification seems like all you need to worry
> about. The rest should work as-is.
> I would use code from Subversion, not just 0.5.
>
> On Sat, Jan 28, 2012 at 4:42 PM, gj <[EMAIL PROTECTED]> wrote:
>
>> Hi,
>>
>> I am trying some experiments (mahout 0.5) with a new neighbourhood
>> model that includes implicit feedback as well as explicit feedback.
>> And I wanted to know, what are the clIasses, etc that I need to
>> modify.
>>
>> To start, my input to the system should in the following format:
>>
>> userid, itemid, exp_feedback, imp_feedback
>>
>> where userid and itemid are integers and exp_feedback and imp_feedback
>> are floats between 0.0 and 1.0
>> for example:
>>
>> 101, 1, 0.67, 0.18
>>
>> I've used the recommender system code in mahout before. But it will be
>> my first time trying to modify it in order to run some custom code. I
>> have skimmed through the Mahout in Action book.
>>
>> The other point is that I want to make this modifications with as
>> little changes to the mahout distribution so that I can benefit from
>> the optimisations that are present in the distribution.
>>
>> Definitely I need to have custom UserSimilarity and I need a custom
>> UserNeighborhood? What else? Can somebody please give me some pointers
>> where to start, e.g which classes i need to modify, etc. and what are
>> the things I should be wary of? Then I will have a go and report back
>> my progress.
>>
>>
>> Regards
>> Gawesh
>>

Hi Gawesh,
you could also implement an ensemble recommender by using two recommenders and then blending the results.

The following JIRA is discussing how to do this:

https://issues.apache.org/jira/browse/MAHOUT-810

/Manuel

On 28.01.2012, at 17:42, gj wrote:

> Hi,
>
> I am trying some experiments (mahout 0.5) with a new neighbourhood
> model that includes implicit feedback as well as explicit feedback.
> And I wanted to know, what are the clIasses, etc that I need to
> modify.
>
> To start, my input to the system should in the following format:
>
> userid, itemid, exp_feedback, imp_feedback
>
> where userid and itemid are integers and exp_feedback and imp_feedback
> are floats between 0.0 and 1.0
> for example:
>
> 101, 1, 0.67, 0.18
>
> I've used the recommender system code in mahout before. But it will be
> my first time trying to modify it in order to run some custom code. I
> have skimmed through the Mahout in Action book.
>
> The other point is that I want to make this modifications with as
> little changes to the mahout distribution so that I can benefit from
> the optimisations that are present in the distribution.
>
> Definitely I need to have custom UserSimilarity and I need a custom
> UserNeighborhood? What else? Can somebody please give me some pointers
> where to start, e.g which classes i need to modify, etc. and what are
> the things I should be wary of? Then I will have a go and report back
> my progress.
>
>
> Regards
> Gawesh

--
Manuel Blechschmidt
Dortustr. 57
14467 Potsdam
Mobil: 0173/6322621
Twitter: http://twitter.com/Manuel_B

Thank you Manuel. I will check it out.

On Sat, Jan 28, 2012 at 7:44 PM, Manuel Blechschmidt
<[EMAIL PROTECTED]> wrote:
> Hi Gawesh,
> you could also implement an ensemble recommender by using two recommenders and then blending the results.
>
> The following JIRA is discussing how to do this:
>
> https://issues.apache.org/jira/browse/MAHOUT-810
>
> /Manuel
>
> On 28.01.2012, at 17:42, gj wrote:
>
>> Hi,
>>
>> I am trying some experiments (mahout 0.5) with a new neighbourhood
>> model that includes implicit feedback as well as explicit feedback.
>> And I wanted to know, what are the clIasses, etc that I need to
>> modify.
>>
>> To start, my input to the system should in the following format:
>>
>> userid, itemid, exp_feedback, imp_feedback
>>
>> where userid and itemid are integers and exp_feedback and imp_feedback
>> are floats between 0.0 and 1.0
>> for example:
>>
>> 101, 1, 0.67, 0.18
>>
>> I've used the recommender system code in mahout before. But it will be
>> my first time trying to modify it in order to run some custom code. I
>> have skimmed through the Mahout in Action book.
>>
>> The other point is that I want to make this modifications with as
>> little changes to the mahout distribution so that I can benefit from
>> the optimisations that are present in the distribution.
>>
>> Definitely I need to have custom UserSimilarity and I need a custom
>> UserNeighborhood? What else? Can somebody please give me some pointers
>> where to start, e.g which classes i need to modify, etc. and what are
>> the things I should be wary of? Then I will have a go and report back
>> my progress.
>>
>>
>> Regards
>> Gawesh
>
> --
> Manuel Blechschmidt
> Dortustr. 57
> 14467 Potsdam
> Mobil: 0173/6322621
> Twitter: http://twitter.com/Manuel_B
>