Modeling leadership influence propagation in cross-departmental collaboration with graph neural networks

Abstrak

To address the issue that traditional methods cannot accurately model non Euclidean structured data in cross departmental collaboration networks or effectively capture the propagation patterns of leadership influence, this study constructs graph convolutional network (GCN) and graph attention network (GAT) models, and proposes a GCN plus GAT fusion model to explore the propagation of leadership influence in cross departmental collaboration. Experimental results on a real enterprise dataset show that the GCN plus GAT fusion model achieves an AUC of 84.72%, which is 9.04% higher than logistic regression (LR), 7.79% higher than support vector machine (SVM), 4.29% higher than the graph structured CNN (PSCN), 3.47% higher than the single GCN, 2.36% higher than the single GAT, and 2.85% higher than GraphSAGE. The model also achieves an F1 score of 61.95%, which is 2.71% higher than the single GAT. The model shows a performance drop of only 0.97% on a simulated dataset, demonstrating stronger generalization ability and noise resistance. These results provide theoretical support for quantitative modeling of leadership influence propagation in cross departmental collaboration and offer new ideas for improving leadership effectiveness in a digital environment.

Identifier

• DOI: https://doi.org/10.18282/hrms5887
• OpenAlex: https://openalex.org/W7164396496