Research On P2P Personal Credit Scoring Model Combined With Social Data

P2P lending,that is,peer-to-peer lending,is an Internet financial product.P2 P lending has the potential to reduce financing costs and improve financial market transaction efficiency.Compared with traditional finance,P2 P lending is smaller in amount,shorter in term,higher in interest rate and requires no collateral.It is well known that due to the poor customer base,P2 P companies often face problems with loan repayment and collection.The P2 P market and company survival conditions in developed countries such as the United States have stabilized due to mature risk management policies and technologies.By contrast,P2 P market is at the exploratory stage in some developing countries including China.At present,the risk control of P2 P lending has become a problem that must be overcome for the sustainable development of P2 P enterprises.In recent years,because of the popularity of the Internet finance industry,more and more scholars have begun to study P2 P lending,and have achieved remarkable results.But these studies still have some deficiencies.First of all,at present,there are few studies consider user behavior and social data in credit card models.In the era of big data,people's social data is diverse and huge.Social networks of P2 P lending users can greatly improve user portraits and reflect credit status.Secondly,most current studies use traditional logistic regression model to construct credit card.However,the prediction accuracy of the linear model is not yet satisfactory for P2 P lending.Innovation is mainly reflected in the following two aspects.Firstly,Dimensionreduction and embedding methods such as LDA and Deep Walk extract effective features from huge behavior and social data of users,and these features have been incorporated into the model training.Secondly,the innovative two-stage hybrid credit scorecard model of Extreme Gradient Boosting(XGBoost)combined Factorization Machine(FM)is proposed.Through the data analysis of Indonesian market data from multinational P2 P company,it's found that the ensemble learning two-stage hybrid model combined with social data can exceed the benchmark model in model stability and prediction accuracy.All in all,data is used to drive the company's operation and decision-making,and further guides the P2 P company's risk management through the combination of social data.This innovative approach can further improve the risk control capability of the P2 P company.