blend(boosting) recommender models

User 2570 | 11/30/2015, 3:48:44 PM

I am trying to blend two recommender model(popularity and item-item similarity) , can one do the same using graph create


User 954 | 11/30/2015, 9:08:29 PM

Hi erigits,

Can you provide us with more information what you exactly want to do? Do you want to somehow blend the top-k results from popularity and item-item similarity models?

Thank you

User 1207 | 12/2/2015, 2:05:01 AM

Hello erigits,

The best way to blend two recommender models at the moment is to pass the output of the recommend() method of one model into the items argument of the other recommend() call. For example, if you do something like:

candidates = m1.recommend(..., k = 25) final_rec = m2.recommend(..., k = 5, items = candidates[['users', 'items']])

then the recommendations in final_rec is chosen from the top 25 items recommended by model m1, but ranked according to the m2 model.

-- Hoyt