A novel update and propagation-based dynamic graph neural network and its application to consumer finance data

In this paper, we propose a novel dynamic graph neural network, the UP-DGNN, which combines the DGNNs and the update-propagation framework. The main contributions of the proposed method are two-fold. On one hand, in the update phase, the model updates the embedding vector of interacting nodes using a T-GRU to extract temporal features, and on the other hand, in the propagation phase, the self-attention mechanism is introduced to update the embedding vector of the affected nodes. Compared with GRU and LSTM models, the proposed model can update node information more efficiently. Further, the performance of the model is demonstrated in public datasets on tasks of link prediction, edge classification and node classification, and we apply the proposed method on the prediction of peer to peer credit default, taking into account the social relationships, and achieve better prediction performance than SOTA methods in consumer finance area.

Keywords

consumer finance, dynamic graphs, graph neural networks, update-propagation components

2010 Mathematics Subject Classification

Primary 68Txx. Secondary 65S05.

The full text of this article is unavailable through your IP address: 172.17.0.1

Received 3 December 2022

Accepted 1 March 2023

Published 19 July 2024