Learning to Make Document Context-Aware Recommendation with Joint Convolutional Matrix Factorization

Algorithm 1

The learning algorithm of JCMF.

	Input:
	The hyperparameters , , and ; user-item rating matrix ; item relation matrix ; user social matrix ; user review text ; item review text ; threshold
	Output:
	Latent vectors ; internal weights of CNNs and ;
	for each item j do
	Preprocess and represent review content as a word embedding based input: ;
	end for
	for each user i do
	Preprocess and represent review content as a word embedding based input: ;
	end for
	Initialize ;
	while do
	for each user i do
	Get from user CNN model;
	Update according to equation (18);
	end for
	for each item j do
	Get from item CNN model;
	Update according to equation (19);
	end for
	for each item i in do
	Update according to equation (20);
	end for
	for each user j in do
	Update according to equation (21);
	end for
	Update according to equation (22);
	Update according to equation (23);
	Compute according to equation (17);
	end while
	return ;