Recommender system with: multiple instances of side data

User 3246 | 2/25/2016, 9:09:27 AM

I am trying to set up a model for a factorization recommender and I need to use side data. The problem I am facing is that some of the items to recommend have, for instance, more than one genre. How can I represent this properly so that the system uses both genres? As of now I have tried to represent this as a list, but I see the system still treats them as one genre.


User 19 | 2/25/2016, 5:50:53 PM

Hi Valentina,

You should be able to have side data that includes a feature column that contains a list of genres. This will be valid input into the model.

For example,

items_data = gl.SFrame({'item_id': [0, 1], 'genres': [['Action', 'Romance'], ['Action', 'Drama', 'Western']]})

then the factorization recommender will learn factors for each of the unique 'genre' items.

Please let me know if that helps, Chris

User 3246 | 2/26/2016, 1:04:20 PM

Dear Chris,

thanks for the quick reply! It works great!