面向供应链金融风险控制的社区发现方法

The Community Detection Approach for Risk Control in Supply Chain Finance

针对现有图对比学习模型使用的随机数据增强策略可能会破坏网络中关键的社区结构、对比对选择策略可能将具有相似拓扑结构或属性的节点选为负样本等问题,本文提出了一种基于自适应图对比学习的社区发现算法,设计了一个考虑节点重要性的自适应图增强策略和一个基于K近邻的正负样本选择策略来解决以上问题。在真实网络上的实验表明,本文提出的方法在社区发现任务中取得了比现有最好算法高4%以上的准确性,在供应链金融风险控制任务中取得了比现有最好算法高3%以上的准确性。

Aiming to the problems that the most graph contrastive learning algorithms use random data augmentation strategies that may destroy the critical community structure in the network and the contrastive pair selection strategy that may select nodes with similar topology or attributes as negative samples,this paper proposes a community detection algorithm based on adaptive graph contrastive learning,and designs an adaptive graph enhancement strategy that considers the importance of nodes and a K-nearest neighbor-based positive sample selection strategy to solve the above problems.Experiments on real-world networks show that the method proposed in this paper is able to achieve an accuracy more than 4%higher than the best existing algorithms in community detection task and more than 3%higher accuracy than the best existing algorithms in the supply chain finance risk control task.