Trustworthy-constraint Deep Graph Learning for Enterprise Financial Risk Prediction

• Published in: Journal of Applied Science and Engineering Vol. 29; no. 1; pp. 113 - 120
• Main Author: Wenting Ma
• Format: Journal Article
• Language: English
• Published: Tamkang University Press 2026
• Subjects: financial risk prediction; heterogeneous graph information aggregation; trustworthy decision-making mechanism
• DOI: 10.6180/jase.202601_29(1).0011
• ISSN: 2708-9967

Financial risk prediction for enterprises is a hot topic in the field of financial technology. Deep learning-Based methods achieve encouraging the financial risk prediction performance due to the power ability of the feature learning. However, there exists two issues in deep learning-Based methods. (1) Current methods fail to accurately model the complex relationships among listed companies in the real financial market. (2) Current methods lack credible estimates in the decision-making process, which leads to the questionable reliability of the decision-making results. To this end, a trustworthy-constraint deep graph learning network (TDGL-net) is proposed to achieve the above goal, which includes the multi-view feature encoding, the heterogeneous graph information aggregation, the trustworthy decision-making mechanism. Specifically, TDGL-net integrates a multi-dimensional bilinear neural tensor and Transformer into a unified multi-view feature encoding to learn comprehensive representation. Then, TDGL-net models heterogeneous graph information aggregation via the cross-category association and the intra-category association, to capture complex inter-enterprise relationships with momentum spillover effects for enhancing the representation discrimination. Additionally, TDGL-net incorporates a trustworthy decision-making mechanism to adaptively integrate information from deep embedding representations and graph embedding representations according to enterprise-specific contexts, improving decision reliability and accuracy. Ultimately, extensive experimental evaluations on the real-world dataset reveal that TDGL-net delivers state-of-the-art performance in predicting enterprise financial risk.