Identifying Key Nodes in Multi-layer Networks Through Relationship Strength Gravity Centrality Based on the Multi-feature Fusion

Identifying key nodes within multi-layer network has become a prominent research topic. The relationship strength between nodes is an important factor in identifying the criticality of nodes. Existing methods mainly measure the strength of relationships between nodes by edge weights on weighted graphs, which have limitations when applying to unweighted graphs. In addition, the different influence ranges of nodes on multi-layer networks can significantly affect their criticality. Therefore, this paper proposes a method called Intrinsic and Topological Features-based Relationship Strength Gravity Centrality (ITFRSGC) for identifying key nodes in multi-layer network. This method fuses multiple node features to calculate hidden relationship strength between nodes in unweighted multi-layer network, and introduces an improved weighted gravity centrality based on the relationship strength. ITFRSGC incorporates centrality based on dynamic influence range as the quality of nodes in the gravity model, which considers the heterogeneity of nodes and the importance of different network layers. Experimental results on multi-layer networks demonstrate that ITFRSGC can more accurately identify key nodes compared to latest methods, and the advantages of our method are particularly prominent on sparsely connected multi-layer networks. Ablation study further confirms the positive impact of incorporating relationship strength based on multi-feature fusion on the overall model performance.