Emerging phenomena of large-scale network data provide opportunities for enhancing the financial risk prediction of SMEs. While existing studies have explored network-based features from the perspective of structure and risk propagation, they fail to capture heterogeneous interactions among entities, resulting in an inaccurate prediction. This research aims to fill the gap by proposing a node-importance based graph contrastive model (NIGCM) to construct firm embeddings that projects heterogeneous interactions from high-dimensional to low-dimensional space. NIGCM organically integrates three modules, i.e., a proposed node-importance aware sampling module, a heterogeneous information aggregation encoder module, and a proposed loss function module based on canonical correlation analysis. Empirical results demonstrate the superior predictive power of NIGCM over existing state-of-the-art methods. This study contributes an effective model for predicting SMEs’ financial risk for stakeholders in the financial market. It also presents a promising avenue for constructing features by leveraging large-scale network data and graph deep learning.
Jiang, Cuiqing; wang, jianfei; Wang, Zhao; and Liu, Ximing, "Capturing Heterogeneous Interactions for Financial Risk Prediction of SMEs" (2022). PACIS 2022 Proceedings. 56.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.