See todo in readme

[ukrbublik]

[ukrbublik]

You can:

use matrix factorization as base algo
use item-to-item recommendations for items rated by user (todo)
use kNN as secondary algo (use users - nearest neighbours) (todo)
use social info (friends, clubs) (todo)
You can not:

include in recommendations items that user already plans to watch or watched but not rated
repeat items of same mediafranchise (todo)

knn sparse - falconn (lsh), nsmlib (https://github.com/nmslib/nmslib/tree/master/python_bindings/notebooks)

knn dense - n2, annoy, faiss; flann, mrpt