Context Attention Heterogeneous Network Embedding

Algorithm 1

Structure-based embedding with node importance.

	Input: network G, context node sampling parameter k, dimensionality , and learning rate η
	Output: -dimensional embedding results H
(1)	Initialize nodes’ relational trees by performing BFS on G starting from each node;
(2)	Obtain a context node sequence S by sampling context nodes layer by layer for each anchor node according to k;
(3)	for i = 1 to do
(4)	Calculate by equation (4);
(5)	end for
(6)	while not convergence do
(7)	Update the value of loss function equation (8) and node representations H by the Adam algorithm with learning rate η;
(8)	end while
(9)	Return H;