Read more
This book integrates corporate financial risk research with graph neural network (GNN) technology to address the challenges of analyzing complex financial data and the interconnections between enterprises. It explores three key areas: 1. Dynamic Graph Representation: A framework for learning dynamic graph representations based on structural roles is proposed, capturing temporal evolution and global topological dependencies, marking the first use of recurrent learning in this context.2. Momentum Spillover Effects: A dual GNN algorithm is introduced to model the dynamic, complex inter-enterprise relationships and momentum spillover effects, offering a new approach to analyzing their impact on securities market volatility.3. Financial Risk Interpretability: To overcome the black-box nature of deep learning models, a heterogeneous GNN framework is developed to generate evidence subgraphs that reveal internal and external factors affecting enterprise financial risk, enhancing model transparency. Experimental results validate the proposed methods, showing improvements across multiple tasks, while also significantly enhancing model interpretability with faster inference times.
About the author
Huaming Du received his Ph.D. degree from Southwestern University of Finance and Economics. His research interests include Causal inference, LLMs, Fintech, and graph representation learning. He has published papers in conferences and journals such as KDD, IEEE TKDE, IEEE TETC, etc.
