What i don't understand is why they locked themselves to log-linear model there in that paper.
There's a Yahoo research paper around where they generalize the case for GLMs and do very similar thing. Except it doesn't have to be a log-linear and actually could be a two-pass learning using whatever techniques in each pass (as long as architecture allows to plug those models one into another). Intuitively i feel that the most promising approach is somethnig like incremental SVD (with first 20 or so items being soft-limited by logit) and perhaps using wieghted regularization learning side information parameters subsequently in minibatches for online learning. but yeah i think this is a most common recommender problem and yes i think Mahout lacks here. I was looking for a similar off-the-shelf solution for this some time ago as well and eventually had to resort to half-baked in-house solution as it wasn't much of a priority. Also i am still not quite sure what the best way to do input normalization for both continuous and nominal inputs together with existing framework. Last thing i heard was it is not working well together in existing SGD solution. But that's one of the most common problem and actually not so hard to solve. It's just one has to dig for a solution, i haven't found it in the docs anywhere even if it exists in Mahout. But i am fairly sure Mahout doesn't support hierarchical plugs at the moment at all. On Wed, Dec 29, 2010 at 10:36 AM, Ted Dunning <[email protected]> wrote: > The latent factor log-linear work that I started, got to a barely working > state and then was unable to continue would provide hybrid recommendations. > > This is the MAHOUT-525 > <https://issues.apache.org/jira/browse/MAHOUT-525>work that Sebastien > alluded to. > > On Wed, Dec 29, 2010 at 9:49 AM, Andy Parsons <[email protected]> > wrote: > > > >> - What do you exactly mean by hybrid recommendations? Do you mean a > > >> combination of content based and collaborative filtering techniques? > > [ASP] Yes, precisely. >
