Reinforced Neighborhood Selection Guided Multi-Relational Graph Neural Networks

Graph Neural Networks (GNNs) have been widely used for the representation learning of various structured graph data, typically through message passing among nodes by aggregating their neighborhood information via different operations. While promising, most existing GNNs oversimplify complexity and diversity edges in thus are inefficient to cope with ubiquitous heterogeneous graphs, which form multi-relational representations. In this article, we propose RioGNN , a novel Reinforced, recursive, flexible selection guided Network architecture, navigate neural network structures whilst maintaining relation-dependent We first construct graph, according practical task, reflect heterogeneity nodes, edges, attributes, labels. To avoid embedding over-assimilation types employ label-aware similarity measure ascertain similar neighbors based on node attributes. A reinforced relation-aware neighbor mechanism is developed choose targeting within relation before all from relations obtain eventual embedding. Particularly, improve efficiency selecting, new recursive scalable reinforcement framework estimable depth width scales graphs. can learn more discriminative enhanced explainability due recognition individual importance each filtering threshold mechanism. Comprehensive experiments real-world data tasks demonstrate advancements effectiveness, efficiency, model explainability, as opposed other comparative GNN models.