Treffer: Dynamic graph neural networks for financial crisis modeling and blocking in large-scale enterprise groups.

Facing the problem of dynamic financial risk propagation under the trend of enterprise group development, this paper proposes a graph neural network modelling method based on distributed training. Aiming at the shortcomings of traditional models that are difficult to capture the dynamic nonlinear multi-hop features of risk contagion, the study constructs a dual-path dynamic learning framework: firstly, a time-varying graph structure is established to characterise the topological evolution of the enterprise association network, and a multi-layer dynamic graph convolution network is used to realise the extraction of nonlinear risk propagation modes; and then, a GPU cluster-driven distributed training architecture is designed to break through the computational bottleneck of large-scale graph data. The experimental results show that the model improves 15%–20% over the sub-optimal model in the three core metrics of AUC-ROC (0.93), Precision@K (0.85) and F1-score (0.88). Combined with the targeted blocking strategy generated by interpretability analysis, the crisis node abatement rate reaches 72% at the 10th time step, forming a closed loop of management from risk prediction to proactive intervention. This study expands the application paradigm of graph neural networks in the field of financial risk control, and provides a whole-process solution for preventing systemic financial risks. [ABSTRACT FROM AUTHOR]

Copyright of Intelligent Decision Technologies is the property of Sage Publications Inc. and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)