Collaborative Metapath Enhanced Corporate Default Risk Assessment on Heterogeneous Graph

Zheng Zhang, yingsheng Ji, Jiachen Shen, Yushu Chen, Xi Zhang, Guangwen Yang

Keywords: default risk, finance, heterogeneous graph, graph neural network, attention mechanism

Abstract:

Default risk assessment for small companies is a tough problem in financial services. Most small businesses expose fragility to external impacts and few information on their insecure finances. Recent efforts utilize advanced Heterogeneous Graph Neural Networks (HGNNs) with metapaths to exploit interactive features in corporate activities for risk analysis. However, few works are proposed for traditional commercial banks. Given a real financial graph, how to detect corporate default risks? We identify two challenges for the task. (1) Massive noisy connections hinder HGNNs to achieve strong results. (2) Multiple semantic connections greatly increase transitive default risk, while existing hierarchical aggregation schemes do not leverage such connection patterns.

In this work, we propose a novel Heterogeneous Graph Co-Attention Network (HetCAN) for corporate default risk assessment. HetCAN aims to take advantage of collaborative metapaths to distill effective risky features by a co-attentive aggregation mechanism, consisting of two attention scores and pairwise importance learning. First, the local attention score models the importance of neighbors under each metapath by considering holistic metapath context. Second, the global attention score further adjusts the importance of neighbors by combining these local attention scores to filter valuable/noisy signals. Then, HetCAN employs pairwise importance learning to enhance attention scores of multi-metapath neighbors for risky feature distillation. Extensive experiments verify that HetCAN outperforms state-of-the-art methods in accurately predicting default risks on large-scale banking datasets.

Track: Graph Algorithms and Learning for the Web

Submission Guidelines Scope: Yes

Submission Guidelines Blind: Yes

Submission Guidelines Format: Yes

Submission Guidelines Limit: Yes

Submission Guidelines Authorship: Yes

Student Author: No

Submission Number: 505