This is a Matrix Factorization which can learn not only user-item rating but also user attributes and item attributes.
python3
numpy==1.14.5 numpydoc==0.7.0
please see how_to_use.py
# set training data
user_ids = [1,1,1,1,5,5,8,8]
item_ids = [1,2,3,4,2,4,1,2]
ratings = [5,5,4,4,3,3,2,2]
user_attributes = {1:[0,1,1], 5:[1,1,0], 8:[0,0,1]}
item_attributes = {1:[1.2, 2.3], 99:[3.7, 1.1]}
# load module
from MatrixFactorizationWithAddtionalData import MF
# train
n_epochs = 10000
mf = MF(n_latent_factor=2, learning_rate=0.005, regularization_weight=0.02, n_epochs=n_epochs, verbose=True)
mf.fit(user_ids, item_ids, ratings, user_attributes, item_attributes)
# predict on insample
preidict = mf.predict(
user_ids, item_ids,
user_attributes=user_attributes,
item_attributes=item_attributes
)
print(preidict)
# predict on outsample
preidict = mf.predict(
user_ids=[8,8,8], item_ids=[3,4,99],
user_attributes=user_attributes,
item_attributes=item_attributes
)
print(preidict)@article{koren2009matrix,
title={Matrix factorization techniques for recommender systems},
author={Koren, Yehuda and Bell, Robert and Volinsky, Chris},
journal={Computer},
number={8},
pages={30--37},
year={2009},
publisher={IEEE}
}