A Hierarchical Network with User Memory Matrix for Long Sequence Recommendation

Algorithm 1

HNUM².

	input: triple < , , >,
	output: the prediction score .
(1)	group the session by users into .
(2)	initialize memory-matrix:
(3)	for in epoch:
(4)	for in user :
(5)	//Session-level
(6)	read by reader into as preference vector
(7)	if new session
(8)	use to initialize GRU hidden state
(9)	as the weight of user interest attention
(10)	, by equation (18)
(11)	//User-level
(12)	when the end of a time step
(13)	write state to by writer
(14)	, by equation (20)
(15)	computer the loss according equation (21)
(16)	end for
(17)	end for