I believe it's less intensive than training the whole set and throwing away data. As an example: if you have 25 items across five categories, you would do 25^2 = 625 comparisons to train an item-item model for all items regardless of category. If you train only within categories, you train 5*(5^2) = 125 operations. I don't think it's possible to do recommendation with less than O(n^2) complexity, so it's a safe bet that running many smaller recommenders will be cheaper than one big one.
On Fri, Jun 21, 2013 at 4:02 PM, Mouthgalya Ganapathy < [email protected]> wrote: > Thanks Alan for the Reply!! > if we have item based recommender model for each product category > wouldn't that be computationally intensive? > > > -----Original Message----- > From: Alan Gardner [mailto:[email protected]] > Sent: Friday, June 21, 2013 3:49 PM > To: [email protected] > Subject: Re: Query in Mahout > > If you're doing item-based recommendation, doesn't it make sense to use > multiple recommenders, one per category? This should reduce your training > time, instead of training on all items then throwing away most of your > results. > > If it was user-based recommendation I could see the value in finding > similar users across multiple categories, then paring down the > recommendations afterwards. > > > On Fri, Jun 21, 2013 at 3:42 PM, Mouthgalya Ganapathy < > [email protected]> wrote: > > > Hi, > > For item based recommendations in Mahout, is there a way to get > > recommendations from a selected Product category? > > For example: > > Product category A: Product1, Product2, Product3 Product Category B: > > Product4, Product5, Product6 > > > > Recommendations for product 2 should be only from category A i.e. > > Product > > 1 and 3.So recommendations should be only within the product category. > > Is this possible? > > > > > > Thanks, > > Mouthgalya > > > > > > -- > Alan Gardner > Solutions Architect - CTO Office > > [email protected] | LinkedIn: > http://www.linkedin.com/profile/view?id=65508699 | @alanctgardner< > https://twitter.com/alanctgardner> > Tel: +1 613 565 8696 x1218 > Mobile: +1 613 897 5655 > > -- > > > -- > > > > -- Alan Gardner Solutions Architect - CTO Office [email protected] | LinkedIn: http://www.linkedin.com/profile/view?id=65508699 | @alanctgardner<https://twitter.com/alanctgardner> Tel: +1 613 565 8696 x1218 Mobile: +1 613 897 5655 -- --
