You can also adjoin the matrices. Adjoining and adding are the two most important ways to handle this.
With adjoining, what happens is that you wind up using the content attributes to compute user similarity as well as the items. When you run through the cooccurrence math, you get content to item, content to content and item to content recommendations. When you put in a history consisting of content attributes and items, you get back a mix of recommended content attributes and items which you need to post filter. On Wed, Feb 22, 2012 at 1:05 AM, Lance Norskog <[email protected]> wrote: > This sounds like two item-item matrices, one from preferences and one > from shared attributes. > > You could combine these by adding the matrices, possibly with weights. > You could also multiply the matrices. > >
