Modelling user pictures with hierarchical Dirichlet process of P2P lending market

The emergence of peer-to-peer (P2P) lending has drawn a lot of attention. The enormous data generated from this billions level market bring us a lots of challenges and opportunities. One interesting question of modelling this data is that can we discover the hidden pattern of users' characteristics from it? Currently, few works have been made to this area. In this article, we try to build a Bayesian probabilistic model to discover the latent user pictures. Especially, we build a user picture model via hierarchical Dirichlet process from the data of one of the biggest market, lending club. The discovered user picture is interpretable and can be evaluated from many perspectives. To demonstrate the usage of user picture, we also proposed a method to predict the loan status. The experimental results show our approach outperformed the comparison methods.