DCKH‐CNN: A Multimetric Graph‐Based Convolutional Neural Network for Identifying Key Influential Nodes in Earth Surface Data Linked Networks

ABSTRACT Identifying key influential nodes in Earth surface data association networks is crucial for optimizing the use of surface scientific data. However, challenges such as network size, data complexity, and dynamic node influence make this task difficult. While deep learning methods have improved recognition accuracy and reduced computational costs in complex networks, they still struggle with balancing efficiency and accuracy. To address this, we propose the DCKH‐CNN, a novel Multimetric Graph‐Based Convolutional Neural Network framework. Based on the LCNN model, it integrates global and local node features by calculating metrics such as degree centrality, K ‐shell, H ‐index, and near‐centrality. One‐hop and two‐hop adjacency matrices are used to represent internode relationships, enhancing feature representation. Trained on small‐scale Earth surface data networks, the model captures unique network characteristics. Experimental results using the SIR model demonstrate that DCKH‐CNN surpasses state‐of‐the‐art algorithms on the vast majority of the Earth Surface Data Linked Network (ESSDLN) datasets and real‐world networks in accuracy, while demonstrating moderate time consumption. This method offers a more efficient approach for identifying key nodes in Earth surface data networks, supporting more accurate recommendations and intelligent analysis of surface scientific data.