Analyzing credit risk among Chinese P2P-lending businesses by integrating text-related soft information

Text-related soft information effectively alleviates the information asymmetry associated with P2P lending and reduces credit risk. Most existing studies use nonsemantic text information to construct credit evaluation models and predict the borrower's level of risk. However, the semantic information also reflect the ability and willingness of borrowers to repay and might be able to explain borrowers' credit statuses. This paper examines whether semantic loan description text information helps predict the credit risk of different types of borrowers using a Chinese P2P platform. We use the 5P credit evaluation theory and the word embedding model to extract the semantic features of loan descriptions across five dimensions. Then, the AdaBoost ensemble learning strategy is applied to construct a credit evaluation model to improve the learning performance of an intelligent algorithm. The extracted semantic features are integrated into the evaluation model to study their explanatory ability with regard to the credit status of different types of borrowers. We conducted empirical research on the Renrendai P2P platform. Our conclusions show that the semantic features of textual soft information significantly improve the predictability of credit evaluation models and that the promotion effect is most significant for first-time borrowers. This paper has important practical significance for P2P platforms and the credit risk management of lenders. Furthermore, it has theoretical value for research concerning heterogeneous information-based credit risk analysis methods in big data environments.