Combining UserSimilarity and ItemSimilarity for a recommendation in Mahout

Question

I have two separate Mahout recommenders set up, one providing recommendations based on user similarity and one based on item similarity. What I really want is a recommender that would base its recommendations on both dimensions, rather than just one.

Going through the documentation, I haven't been able to find any Recommender implementation which takes into account multiple dimensions. I could implement a basic version myself by taking the set intersection from the UserSimilarity recommendation set and the ItemSimilarity recommendation set, but it definitely wouldn't be the best way.

Answer 1

No it doesn't exist in the project. I think you could piece it together with some work. It would ultimately be based on a weighted average, where weights are a product of user-user and item-item similarity or something. You may find this just gets too slow to compute at run-time, or, if you want a more 'holistic' model you may find latent factor models more interesting anyway. But I have not tried it.