A Novel GNN-Based Node Importance Ranking Method in a Heterogeneous Network

In a heterogeneous network, a reliable importance ranking of nodes can provide a useful reference for decision-making. Nevertheless, existing node importance ranking methods may not effectively integrate the characteristics of the edges and nodes in the graph, as well as the structural information. In this paper, a novel graph neural network-based heterogeneous network node importance ranking (GNN-HNNIR) method is proposed to rank the importance of different functional nodes in a heterogeneous network. GNN-HNNIR can adjust the node importance score based on those of neighboring nodes, which employ a multi-head attention mechanism to aggregate scores by pooling information from edges, nodes, and the graph structure. Finally, numerical experiments show that GNN-HNNIR can effectively evaluate the importance scores of different nodes with given small amount of data for training.