The financial system is a complex system with many components and sophisticated relations, which may be frequently updated. To represent the relational data in the financial domain, graphs are commonly constructed. In this work we will focus on multi-relational and/or temporal graphs that may contain multiple type of nodes and edges and apply recent methodologies, such as graph neural networks, on graphs constructed from financial data.
As a first step, the use of financial data graphs and the state-of-the-art models deployed will be analyzed. Then, these approaches will be extended and modified to be used on specific financial graphs (such as transactions graphs, dynamic financial KGs etc.) that may also contain external information (e.g. text about the nodes in the graph). The end goal will be to create an end-to-end system that can handle multi-relational/dynamic graphs and perform competitively on graph analytic tasks, mainly link prediction and node classification.
Indicative Bibliography:
Wang, Jianian, et al. “A review on graph neural network methods in financial applications.” arXiv preprint arXiv:2111.15367 (2021).