A Shilling Group Detection Framework Based on Deep Learning Techniques

Algorithm 3

GPSA-based user classification.

	Input: rating dataset
	sampling result SN
	training set Tr
	test set Te
	number of iterations loop
	Output: the set of group shilling attackers GSA
	Begin
(1)
(2)	for each do
(3)	obtain user feature vector using sparse autoencoder
(4)	end for
(5)	for k = 1 to loop do
(6)	for each do
(7)	calculate the user embeddings according to equations (16)–(20)
(8)	calculate the cross-entropy loss according to equations (21)–(23)
(9)	perform back-propagation to update the weight matrices
(10)	end for
(11)	end for
(12)	for eachdo
(13)	calculate the user label probabilities and according to equations (16)–(22)
(14)	ifthen
(15)
(16)	end if
(17)	end for
(18)	return GSA
	End