Reproduce “Corporate Fraud Detection in Rich-yet-Noisy Financial Graph” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shiqi Wang, Zhibo Zhang, Libing Fang, Cam-Tu Nguyen, Wenzhong Li
Nanjing University researchers developed KeGCN_R_, a Knowledge-enhanced Graph Convolutional Network with Robust Two-stage Learning, to detect corporate fraud in complex financial graphs. The model mitigates information overload from abundant support nodes and addresses hidden fraud (mislabeled true fraud cases), achieving improved detection accuracy on newly released real-world Chinese stock market datasets.
Key insights
- Information overload in heterogeneous financial graphs is a significant challenge for GNNs, but knowledge graph embeddings combined with processing on company-centric subgraphs effectively distill useful information from support nodes.
- The proposed robust two-stage learning framework, specifically designed to model asymmetric, instance-, and neighborhood-dependent noise, is crucial for effectively handling hidden fraud and outperforms other noise-robust methods.
- While rich relational data is valuable, financial attributes from annual statements remain fundamental features for corporate fraud detection and are not fully replaceable by knowledge embeddings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.