Graph Embedding-Based Sensitive Link Protection in IoT Systems

Table 1

Summary of notations.

Notations	Meanings
	The undirected original graph
	The set of nodes in
	The number of nodes
	The set of edges in
	The number of edges
	The node
	The edge between and
	The node feature matrix of
	The number of node attributes
	The adjacency matrix of
	The adjacency matrix of privacy graph
	The adjacency matrix of training graph
	The reconstructed adjacency matrix of
	The reconstructed adjacency matrix of privacy graph
	The link state between and in
	The link state between and in
	The number of categories for node labels
	The node label matrix predicted by softmax classifier with each row includes the predicted values of categories
	The privacy embedding of privacy graph
	The link protection graph embedding
	The higher dimensional graph embedding concatenated by and
	The maximum number of edges added for each sensitive link
	The sensitive links in
	Part of nonsensitive links in
	The links which are known to the attack models
	The reconstruction loss
	The node classification loss
	The distribution loss of the generator
	The total loss of the generator
	The distribution loss of the discriminator
	The classification accuracy of the attack models for sensitive links
	The classification accuracy of the attack models for nonsensitive links
	The link reconstruction accuracy of
	The link reconstruction recall of
	The node classification accuracy of