Yes, if I understand your meaning. Both steps will be computed in Hadoop. The second step involves loading one row of A'A at a time, and dotting it with the user vector to get a recommendation score for one item.
On Fri, Dec 4, 2009 at 9:31 AM, Gökhan Çapan <[email protected]> wrote: > Ok, so we will implement and item based recommender; > - mapred jobs for constructing A'A first. > - the online part of algorithm.((A'A)h). > > What do you think about my comments on the online step? Using a small subset > of A'A
