Differentially Private Attributed Network Releasing Based on Early Fusion

Algorithm 3

Differentially private attributed network publishing by late fusion (LN).

	Input: node dataset V, attribute dataset , group number K, privacy parameter , , , maximum iterations .
	Output: sanitized attribute network .
(1)	Initialize the attribute network from V, and K.
(2)	Initialize the parameters
(3)	for t = 1,…,do
(4)	E step: given , compute .
(5)	M step: compute
(6)	.
(7)	Compute and choose .
(8)	Given , compute the number of vertices from , called .
(9)	Compute .
(10)	end for
(11)	Compute , where denotes the number of the vertices which take the attribute value as .
(12)	Compute attribute parameter , where denotes the sensitivity of u.
(13)	Compute correlation parameter using and .
(14)	Sample attribute with probability .
(15)	Sample edge .