From: Graph convolutional and attention models for entity classification in multilayer networks
Notations | Description |
---|---|
G | A simple (i.e., monoplex) network graph |
V,Ā E | Set of nodes and set of edges in G |
\(G_{{\mathcal {L}}}\) | A multilayer network graph |
\({\mathcal {V}}\) | Set of entities in \(G_{{\mathcal {L}}}\) |
\({\mathcal {L}}, \ell , L_l\) | Set of layers, number of layers, l-th layer of \(G_{{\mathcal {L}}}\) |
\(V_{{\mathcal {L}}}, E_{{\mathcal {L}}}\) | Set of nodes and set of edges in \(G_{{\mathcal {L}}}\) |
i | Index of node \(v_i\) in G, resp. entity \(v_i\) in \(G_{{\mathcal {L}}}\) |
\(\Gamma (i)\) | Neighborhood of node \(v_i\) in G |
\(\Gamma (i,l)\) | Within-layer neighborhood of \(v_i\) in layer \(L_l\) of \(G_{{\mathcal {L}}}\) |
\(\Psi (i,l)\) | Outside-layer neighborhood of \(v_i\) in layers of \(G_{{\mathcal {L}}}\) different from \(L_l\) |
\({\mathbf {X}}, {\mathbf {X}}_l\) | Attribute (input feature) matrix in G, resp. in the l-th layer of \(G_{{\mathcal {L}}}\) |
\({\mathbf {x}}_i\), \({\mathbf {x}}_{(i,l)}\) | Attribute (input feature) vector for node \(v_i\) in G, resp. entity \(v_i\) in layer \(L_l\) of \(G_{{\mathcal {L}}}\) |
\({\mathbf {Z}}, {\mathbf {Z}}_{l}\) | Embedding (output feature) matrix in G, resp. in the l-th layer of \(G_{{\mathcal {L}}}\) |
\({\mathbf {z}}_i\) | Embedding (output feature) vector for node \(v_i\) |
\(\widetilde{{\mathbf {Z}}}\) | Embedding (output feature) matrix for the entities in \(G_{{\mathcal {L}}}\) |
\({\mathbf {h}}_i\) | Hidden-layer vector for node \(v_i\) |
\({\mathbf {h}}^{(k)}_{(i,l)}\) | Hidden-layer vector at the k-th layer of the GNN for entity \(v_i\) in layer \(L_l\) of \(G_{{\mathcal {L}}}\) |
f | Number of attributes (input features) |
d | Size of the embedding |
K,Ā k | Number of GNN layers, index of layer |
Q,Ā q | Number of attention heads, index of attention head |
\({\mathbf {W}}, {\mathbf {W}}^{(k)}\) | Weight matrix of a generic, resp. k-th, layer of a GNN |
\({\mathbf {A}}, {\mathbf {A}}_l\) | Adjacency matrix in G, resp. in the l-th layer of \(G_{{\mathcal {L}}}\) |
\({\mathbf {A}}^{\text {sup}}\) | Supra-adjacency matrix in \(G_{{\mathcal {L}}}\) |
\(\widetilde{{\mathbf {A}}}, \widetilde{{\mathbf {A}}}^{\text {sup}}\) | Adjacency matrix, resp. supra-adjacency matrix, with self-loops |
\(\sigma (\cdot )\) | Activation function |
\(e_{ij}\) | Attention coefficient for edge between nodes \(v_i\) and \(v_j\) |
\(\alpha _{ij}\) | Normalized attention coefficient for edge between nodes \(v_i\) and \(v_j\) |