多模块注意力协同的异构图神经网络产业链风险识别模型

Industrial Chain Risk Identification Model Based on Heterogeneous Graph Neural Network with Multi-module Attention Collaboration

摘要

风险识别通过精准定位产业链的脆弱节点以阻断风险传导，对于维护产业链完整性至关重要。然而，现有方法缺乏对产业链节点进行静态指标级、动态时序级和结构空间级的协同建模。因此，本文提出一种多模块注意力协同的异构图神经网络产业链风险识别模型（MGRI）。MGRI 首先根据企业所属行业动态学习并分配各属性维度的权重，以突出关键指标的重要性；同时捕捉低频财务时间序列蕴含的风险演变信息，形成全局动态时序表征；最后以动静态融合特征作为初始表示聚合关系感知的邻域信息，生成包含产业链结构空间依赖的最终嵌入并用于风险识别，实现了跨维度多层级特征的协同建模。实验验证了 MGRI在多个真实产业链数据集上准确识别风险企业方面优于最先进的方法。

Abstract

Risk identification is essential for maintaining the integrity of industry chains by precisely locating vulnerable nodes to inter- rupt risk propagation.However，existing methods lack collaborative modeling of industry nodes across static-indicator level，dynamic- time-series level，and structural-spatial level representations.Therefore，this paper proposes an industrial chain risk identification model based on a heterogeneous graph neural network with multi-module attention collaboration（MGRI）.MGRI first dynamically learns and assigns weights to attribute dimensions based on a company＇s industry to emphasize the importance of key indicators；it simultaneously captures the risk evolution information embedded in low-frequency financial time series to form a global dynamic temporal representa- tion；finally，using the fused static and dynamic features as initial representations，it aggregates relation-aware neighborhood informa- tion to generate the final embeddings that contain the structural-spatial dependencies of the industry chain for risk identification，thereby achieving collaborative modeling of cross-dimensional，multi-level features.Experiments on two real-world industry chain datasets demonstrate that MGRI outperforms state-of-the-art methods in accurately identifying at-risk enterprises.

关键词

产业链 / 风险识别 / 异构图神经网络 / 注意力机制 / 多模块协同

Key words

industry chain / risk identification / heterogeneous graph neural network / attention mechanism / multi-module collaboration

参考文献

[1] GUO Z M, QIU Y. Reconstruction of the global value chain in the digital economy era;typical facts,theoretical mechanisms and Chi- nese strategies[J]. Reform, 2020,(10 ):73-85, doi: 10.1109/ECIT52743.2021.00046.

[2] CHEN X D, YANG X X. Does digital transformation improve the autonomous and controllable capabilities of the industrial chain?[J]. Business Management Journal, 2022, 44(8):23-39.

[3] Wen C, Yang J, Gan L, et al. Big data driven internet of things for credit evaluation and early warning in finance[J]. Future Genera- tion Computer Systems, 2021, 124:295-307,doi:10.1016/j.fu-turc.2021.06.003.

[4] Hewage C, Yasakethu L, Jayakody D N K. Data protection:the wake of ai and machine learning[M]. Cham: Springer Nature Switzerland, 2024.

[5] Bi W, Xu B, Sun X, et al. Company-as-tribe:company financial risk assessment on tribe-style graph with hierarchical graph neural net- works[C]// Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2022:2712-2720.

[6] Song L, Li H, Tan Y, et al. Enhancing enterprise credit risk assess- ment with cascaded multi-level graph representation learning[J]. Neural Networks, 2024, 169:475-484.

[7] Zhou Z, Bi K, Zhong Y, et al. HKTGNN:hierarchical knowledge transferable graph neural network-based supply chain risk assess- ment[C]// Parallel Distributed Processing with Applications,Big Data Cloud Computing,Sustainable Computing Communications, Social Computing Networking,2023:772-782.

[8] Mahmoud A, Mohammed A. A survey on deep learning for time-se- ries forecasting[M].Machine Learning and Big Data Analytics Par- adigms: Analysis, Applications and Challenges,2021:365-392.

[9] DONG S Y. The research of the sparse time series prediction based on empirical mode decomposition and attention mechanism[D]. Changchun: Jilin University, 2024.

[10] Zhou J, Cui G, Hu S, et al. Graph neural networks:a review of methods and applications[J]. AI Opcn, 2020, 1:57-81,doi: 10.1016/j.aiopen.2021.01.001.

[11] Kipf T N, Welling M. Semi-supervised classification with graph convolutional networks[J]. arXiv:1609.02907, 2017,doi: 10.48550/arXiv.1609.02907.

[12] Veličković P, Cucurull G, Casanova A, et al. Graph attention net- works[C]// International Conference on Learning Representations, 2018,doi:10.48550/arXiv.1710.10903.

[13] Bing R, Yuan G, Zhu M, et al. Heterogeneous graph neural net- works analysis:a survey of techniques,evaluations and applications [J].Artificial Intelligence Review, 2023, 56(8):8003-8042.

[14] Zhang C, Song D, Huang C, et al. Heterogeneous graph neural net- work[C]// Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery ＆ Data Mining, 2019:793-803.

[15] Wang X, Ji H, Shi C, et al. Heterogeneous graph attention network[C]// The World Wide Web Conference, 2019:2022-2032.

[16] Hu Z, Dong Y, Wang K, et al. Heterogeneous graph transformer[C]// Proceedings of The Web Conference, 2020:2704-2710.

[17] Altman E I. Financial ratios, discriminant analysis and the predic- tion of corporate bankruptcy[J]. The Journal of Finance, 1968, 23 (4):589-609.

[18] Ohlson J A. Financial ratios and the probabilistic prediction of bank- ruptcy[J]. Journal of Accounting Research, 1980, 18(1):109-131.

[19] Lee Y C. Application of support vector machines to corporate credit rating prediction[J]. Expert Systems with Applications, 2007, 33 (1):67-74.

[20] Hua Z, Wang Y, Xu X, et al. Predicting corporate financial distress based on integration of support vector machine and logistic regres- sion[J]. Expert Systems with Applications, 2007, 33(2):434-440.

[21] Xu K, Hu W, Leskovec J, et al. How powerful are graph neural net- works?[C]// International Conference on Learning Representa- tions, 2019,doi:10.48550/arXiv.1810.00826.

[22] Wu Z, Pan S, Chen F, et al. A comprehensive survey on graph neu- ral networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(1):4-24.

[23] Zhang Z, Ji Y, Shen I, et al. Collaborative metapath enhanced cor- porate default risk assessment on heterogeneous graph[C]// Pro- ceedings of the ACM Web Conference, Association for Computing Machinery,2024:446-456.

[24] Wei S, Lv J, Guo Y, et al. Combining intra-risk and contagion risk for enterprise bankruptcy prediction using graph neural networks[J]. Information Sciences, 2024,659:120081,doi:10.1016/j.ins.2023.120081.

[25] Bi Kejun, Sun Pengzhao, Wang Ruijin, et al. Industry chain risk as- sessment model combining graph fusion and attribute completion[J/OL]. Journal of Chinese Computer Systems, 2025:1-11, http://kns.cnki.nct/kcms/dctail/21.1106.tp.20240731.1001.002.html.

[26] Chang Y C, Chang K H, Wu G J. Application of eXtreme gradient boosting trees in the construction of credit risk assessment models for financial institutions[J]. Applied Soft Computing, 2018, 73: 914-920,doi:10.1016/j.asoc.2018.09.029.

[27] Lv Q, Ding M, Liu Q, et al. Are we really making much progress? Revisiting,benchmarking and refining heterogeneous graph neural networks[C]// Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery ＆ Data Mining, 2021:1150-1160.

[28] 郭周明, 裘莹. 数字经济吋代全球价值链的重构:典型事实、理论机制与中国策略[J]. 改革, 2020,(10):73-85.

[29] 陈晓东, 杨晓霞. 数宁化转型是否提升了产业链自立可控能力?[J]. 经济管理, 2022, 44(8):23-39.

[30] 莗思逾. 基于经验模态分解和注意力机制的稀疏时间序列预测研究[D]. 长恚: 吉林大学, 2024.

[31] 毕可骏, 孙鹏钊, 王瑞锦, 等. 结合图融合和属性补全的产业链风险评估模型[J/OL]. 小型微型计算机系统, 2025:1-11, http://kns.cnki.net/kcms/detail/21.1106.tp.20240731.1001.002.html.

基金资助

国家重点研发计划项目(2022YFB3304300)

国家自然科学基金项目(62472329)

国家自然科学基金项目(62072342)