You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@mahout.apache.org by Pradeep Pujari <pp...@gmail.com> on 2010/07/09 01:57:03 UTC
Recommendation Algorithms
Hi,
Recommendation Algorithms: Can it be used for a case like, people who viewed
this item also viewed these other items? I read the taste recommendation
framework which talks about collaborative filtering. Looks to me this above
use case is not a collaborative filtering subject. We know the click data
and math lib can able to help. Please advise.
Thanks,
Pradeep.
Re: Recommendation Algorithms
Posted by Robin Anil <ro...@gmail.com>.
For this simple case FPGrowth can work as well. Top Frequent patterns for a
given item could infact be a good list of recommended items
On Fri, Jul 9, 2010 at 1:52 PM, Sean Owen <sr...@gmail.com> wrote:
>
Re: Recommendation Algorithms
Posted by Sean Owen <sr...@gmail.com>.
Yes this is simpler than a recommendation problem. You probably want
to use LogLikelihoodSimilarity with GenericItemBasedRecommender and
call mostSimilarItems(). The input will just be user ID / item ID
pairs, one for each click, with no preference value.
On Fri, Jul 9, 2010 at 12:57 AM, Pradeep Pujari <pp...@gmail.com> wrote:
> Hi,
>
> Recommendation Algorithms: Can it be used for a case like, people who viewed
> this item also viewed these other items? I read the taste recommendation
> framework which talks about collaborative filtering. Looks to me this above
> use case is not a collaborative filtering subject. We know the click data
> and math lib can able to help. Please advise.
>
> Thanks,
> Pradeep.
>
Re: Recommendation Algorithms
Posted by Pradeep Pujari <pp...@gmail.com>.
Thanks Robin and Sean, I will experiment with both the approaches and update
you.
Thanks
Pradeep
On Fri, Jul 9, 2010 at 9:59 PM, Sean Owen <sr...@gmail.com> wrote:
> Either approach could work. In essence they are doing something
> similar. What works best for your problem will depend on the exact
> data.
>
> On Sat, Jul 10, 2010 at 12:37 AM, Pradeep Pujari <pp...@gmail.com>
> wrote:
> > Hi Ted,
> >
> > I want to build a prototype for "people who view this item also viewd
> these
> > other items"
> > using Mahout. I am exploring how Mahout could help. I have data like
> > user_id --> item_id--->no_of_clicks. Looks to me this is not a
> collaborative
> > filtering problem.
> > Because, this is neither finding users having similar taste not
> similarilty
> > between items.
> > I think this is a problem of Co-occurrence discovery and can be solved by
> > Association Rules Mining
> > algorithms like FP Growth. Any comment on this is highly appriciated.
> >
> > Thanks in advance.
> > Pradeep
> >
> >
> > On Thu, Jul 8, 2010 at 5:15 PM, Ted Dunning <te...@gmail.com>
> wrote:
> >
> >> The answer to your first question is "yes".
> >>
> >> The answer to your second question (please advise) is "heh?"
> >>
> >> Can you explain what you are asking in a bit more detail?
> >>
> >> On Thu, Jul 8, 2010 at 4:57 PM, Pradeep Pujari <pp...@gmail.com>
> wrote:
> >>
> >> >
> >> > Recommendation Algorithms: Can it be used for a case like, people who
> >> > viewed
> >> > this item also viewed these other items? I read the taste
> recommendation
> >> > framework which talks about collaborative filtering. Looks to me this
> >> above
> >> > use case is not a collaborative filtering subject. We know the click
> data
> >> > and math lib can able to help. Please advise.
> >> >
> >> >
> >>
> >
>
Re: Recommendation Algorithms
Posted by Sean Owen <sr...@gmail.com>.
Either approach could work. In essence they are doing something
similar. What works best for your problem will depend on the exact
data.
On Sat, Jul 10, 2010 at 12:37 AM, Pradeep Pujari <pp...@gmail.com> wrote:
> Hi Ted,
>
> I want to build a prototype for "people who view this item also viewd these
> other items"
> using Mahout. I am exploring how Mahout could help. I have data like
> user_id --> item_id--->no_of_clicks. Looks to me this is not a collaborative
> filtering problem.
> Because, this is neither finding users having similar taste not similarilty
> between items.
> I think this is a problem of Co-occurrence discovery and can be solved by
> Association Rules Mining
> algorithms like FP Growth. Any comment on this is highly appriciated.
>
> Thanks in advance.
> Pradeep
>
>
> On Thu, Jul 8, 2010 at 5:15 PM, Ted Dunning <te...@gmail.com> wrote:
>
>> The answer to your first question is "yes".
>>
>> The answer to your second question (please advise) is "heh?"
>>
>> Can you explain what you are asking in a bit more detail?
>>
>> On Thu, Jul 8, 2010 at 4:57 PM, Pradeep Pujari <pp...@gmail.com> wrote:
>>
>> >
>> > Recommendation Algorithms: Can it be used for a case like, people who
>> > viewed
>> > this item also viewed these other items? I read the taste recommendation
>> > framework which talks about collaborative filtering. Looks to me this
>> above
>> > use case is not a collaborative filtering subject. We know the click data
>> > and math lib can able to help. Please advise.
>> >
>> >
>>
>
Re: Recommendation Algorithms
Posted by Pradeep Pujari <pp...@gmail.com>.
Hi Ted,
I want to build a prototype for "people who view this item also viewd these
other items"
using Mahout. I am exploring how Mahout could help. I have data like
user_id --> item_id--->no_of_clicks. Looks to me this is not a collaborative
filtering problem.
Because, this is neither finding users having similar taste not similarilty
between items.
I think this is a problem of Co-occurrence discovery and can be solved by
Association Rules Mining
algorithms like FP Growth. Any comment on this is highly appriciated.
Thanks in advance.
Pradeep
On Thu, Jul 8, 2010 at 5:15 PM, Ted Dunning <te...@gmail.com> wrote:
> The answer to your first question is "yes".
>
> The answer to your second question (please advise) is "heh?"
>
> Can you explain what you are asking in a bit more detail?
>
> On Thu, Jul 8, 2010 at 4:57 PM, Pradeep Pujari <pp...@gmail.com> wrote:
>
> >
> > Recommendation Algorithms: Can it be used for a case like, people who
> > viewed
> > this item also viewed these other items? I read the taste recommendation
> > framework which talks about collaborative filtering. Looks to me this
> above
> > use case is not a collaborative filtering subject. We know the click data
> > and math lib can able to help. Please advise.
> >
> >
>
Re: Recommendation Algorithms
Posted by Ted Dunning <te...@gmail.com>.
The answer to your first question is "yes".
The answer to your second question (please advise) is "heh?"
Can you explain what you are asking in a bit more detail?
On Thu, Jul 8, 2010 at 4:57 PM, Pradeep Pujari <pp...@gmail.com> wrote:
>
> Recommendation Algorithms: Can it be used for a case like, people who
> viewed
> this item also viewed these other items? I read the taste recommendation
> framework which talks about collaborative filtering. Looks to me this above
> use case is not a collaborative filtering subject. We know the click data
> and math lib can able to help. Please advise.
>
>