Context Attention Heterogeneous Network Embedding

Algorithm 2

Generating strategy of context embedding.

	Input: network G, context node sequences S, dimensionality , learning rate η, EWMA parameter γ, and values
	Output: -dimensional embedding results C
(1)	Normalize context node sequences S layer by layer with values;
(2)	Apply EWMA on normalized context nodes with parameter γ to obtain a weight for each context node;
(3)	Encode text contents of nodes in the context node sequence and input them into the CNN;
(4)	while not convergence
(5)	Update the value of loss function and node representations C by the Adam algorithm with learning rate η;
(6)	end while
(7)	Return C;