Loan Recommendation in P2P Lending Investment Networks: A Hybrid Graph Convolution Approach

Abstract

Low successful rate of loan money has become a significant challenge for P2P Lending platforms. In this paper, we propose a hybrid deep learning algorithm combining strengths of both supervised graph convolution network and unsupervised community discovery, to accurately match the loan requirements and investing lenders on P2P Lending platforms. Our hybrid deep architecture learns predictive node embedding from both local neighborhood information and global structural information through complex investment networks with minimal information loss. We evaluate our method on large-scale dataset collected from a real-world P2P platform. Compared with strong baselines, our proposed method provides optimal loan recommendation performance, generates efficient solutions for “Cold-Start” problem and features fast computing speed.