10.1 Heterogeneous & Knowledge Graph Embedding

10.1 Heterogeneous & Knowledge Graph Embedding

Recap: A Single GNN Layer

回顾

一个单层 GNN 层应该包括：

(1) Message

\(\mathbf{m}_{u}^{(l)}=\operatorname{MSG}^{(l)}\left(\mathbf{h}_{u}^{(l-1)}\right), u \in\{N(v) \cup v\}\)

每个节点计算一个消息

(2) Aggregation

\(\mathbf{h}_{v}^{(l)}=\mathrm{AGG}^{(l)}\left(\left\{\mathbf{m}_{u}^{(l)}, u \in N(v)\right\}, \mathbf{m}_{v}^{(l)}\right)\)

从邻居节点聚合消息