Concepts de base
The author proposes a motif-preserving Graph Neural Network with curriculum learning to improve financial default prediction by capturing both lower-order and higher-order structures from social graphs.
Résumé
User financial default prediction is crucial for credit risk management. The proposed MotifGNN method effectively captures motif-based graph structures for accurate predictions, outperforming existing methods on public and industrial datasets.
The paper introduces the challenges in traditional default prediction methods and proposes a novel approach using motifs in graph neural networks. By incorporating both lower-order and higher-order structures, the model achieves superior performance on various datasets.
The study highlights the importance of considering social relations in financial default prediction and demonstrates the effectiveness of the proposed MotifGNN method through extensive experiments.
Stats
Extensive experiments demonstrate that our proposed method achieves superior results on one public dataset and two industrial datasets.
The model is trained for a maximum of 60 epochs with a batch size of 256 using Pytorch for Cora dataset.
For ConsumeLn and CashLn datasets, the model is trained for a maximum of 3 epochs with a batch size of 512 using Tensorflow.
The time complexity of the model is O(m^1.5) where each motif-based graph can be processed parallelly.
Citations
"The purpose of the default prediction is to predict whether the user will fail to repay the money in the future."
"We propose a motif-preserving Graph Neural Network with curriculum learning to jointly learn lower-order structures from original graphs and higher-order structures from multi-view motif-based graphs."