基于图卷积网络的中小企业信用风险预测
Credit Risk Prediction for Small and Medium-Sized Enterprises Based on Graph Convolutional Networks

长期以来中小企业一直存在融资难的问题，供应链金融被提出作为解决该问题的一个重要方案。如何更准确地评估供应链金融背景下的中小企业的信用问题成为一大难点。本文提出了一种基于图卷积网络的中小企业信用风险预测方法，该方法充分考虑了整个供应链网络的信息。首先，以化学制药行业为背景构建了包含420家核心企业、一级供应商和二级供应商的供应链网络，然后构建两层图卷积神经网络，将企业财务、基本信息等特征以及供应关系数据作为输入，企业是否ST或破产清算等作为标签训练，使其学习供应链网络间的复杂非线性关系并对末端中小企业进行风险预测。最后对比了SVM、AdaBoost等传统机器学习模型，实验结果表明，本文所提出的模型预测效果较好，为供应链金融背景下的中小企业信用风险预测提供了新思路。
For a long time, small and medium-sized enterprises (SMEs) have faced challenges in accessing financing, with supply chain finance proposed as a significant solution to address this issue. Accurately assessing the credit of SMEs within the context of supply chain finance has been a major challenge. This paper proposes a credit risk prediction method for SMEs based on graph convolutional networks, which takes into account information across the entire supply chain network. Firstly, this paper constructed a supply chain network comprising 420 core enterprises, primary suppliers, and secondary suppliers within the context of the pharmaceutical industry. Secondly, a two-layer graph convolutional neural network was developed. It utilized features such as financials, basic information, and supply relationship data as inputs, while training on labels indicating whether companies are flagged as “ST” (Special Treatment) or subjected to bankruptcy liquidation. This enabled the model to learn complex nonlinear relationships among the supply chain networks and predict risks for SMEs at the end of the chain. Comparative analyses were conducted with traditional machine learning models like SVM and AdaBoost. Experimental results demonstrated the effectiveness of the proposed model, offering new insights into credit risk prediction for SMEs within the domain of supply chain finance.