Groups of small and medium enterprises (SMEs) can back each other to obtain loans from banks and thus form guarantee networks. If the loan repayment of a small company in the network defaults, its backers are required to repay the loan. Therefore, risk over networked enterprises may cause significant contagious damage. In real-world applications, it is critical to detect top vulnerable nodes in such complex financial network with near real-time performance. To address this challenge, we introduce VulnDS: a top-k vulnerable SME detection system for large-scale financial networks, which is deployed in our collaborated bank. First, we model the risks of the guaranteed-loan network by a probabilistic graph, which consists of the guarantee-loan network structure, self-risk probability for the nodes and diffusion probability for the edges. Moreover, to identify the vulnerable enterprises, we propose a sampling-based approach with tight theoretical guarantee. Novel optimization techniques are developed in order to scale for large networks. We conduct extensive experiments on 3 real financial datasets, in addition with 5 large-scale benchmark networks. The evaluation results show that the proposed method can achieve up to 100x speedup ratio compared with baseline methods. Case studies are further conducted in the deployed system to demonstrate the effectiveness of proposed model.

中文翻译：

---

VulnDS：网络贷款中的 Top-k 脆弱中小企业检测系统

中小企业集团可以相互支持，从银行获得贷款，从而形成担保网络。如果网络中的一家小公司的贷款偿还违约，其支持者需要偿还贷款。因此，网络企业的风险可能会造成重大的传染性损害。在实际应用中，以近乎实时的性能检测这种复杂金融网络中的顶级易受攻击节点至关重要。为了应对这一挑战，我们引入了 VulnDS：一种用于大型金融网络的 top-k 脆弱中小企业检测系统，部署在我们的合作银行中。首先，我们通过概率图对担保贷款网络的风险进行建模，该图由担保贷款网络结构、节点的自风险概率和边的扩散概率组成。而且，为了识别脆弱企业，我们提出了一种基于抽样的方法，具有严格的理论保证。开发了新的优化技术以扩展大型网络。我们对 3 个真实的金融数据集以及 5 个大型基准网络进行了广泛的实验。评估结果表明，与基线方法相比，所提出的方法可以实现高达 100 倍的加速比。在部署的系统中进一步进行案例研究，以证明所提出模型的有效性。评估结果表明，与基线方法相比，所提出的方法可以实现高达 100 倍的加速比。在部署的系统中进一步进行案例研究，以证明所提出模型的有效性。评估结果表明，与基线方法相比，所提出的方法可以实现高达 100 倍的加速比。在部署的系统中进一步进行案例研究，以证明所提出模型的有效性。