Table 2. Table of symbols used in this paper.

Symbol	Definition
G = (V, E)	G: input graph, V: node set, E: edge set.
|V| = n	n: number of nodes.
v ∈ V	nodes in G.
xv ∈ Rc	xv: node feature vector, c: feature dimension.
X ∈ Rn×c	initial feature matrix.
$h_i^k$	node embedding of node i in the kth layer.
N(v)	set of one-hop neighbors of node v in G.
M(·)	message aggregation function.
H k	matrix of activations in the kth layer.
W k	a layer-specific trainable weight matrix.
$\tilde{A}$	adjacency matrix of the undirected graph G with added self-connections.
$\tilde{D}$	degree matrix of undirected graph G.
σ(·)	activation function.
f(i, j)	a function used to calculate the attention coefficient between i and j
g(·)	feature transformation function.
δ k	the weight matrix decay parameter of the kth layer
Z k	the kth layer output of DGCNNII.
AGAT ∈ Rn×n	adjacency matrix based on node attention coefficients.
α, β, γ	hyper-parameters for adjusting the proportion of information aggregation.
I n	identity matrix.
a	the weight vector that projects the concatenate vector to the scalar