Leveraging Deep Learning and Multimodal Signals from Social Media to Enhance Credit Risk Prediction

With the rise of Internet-based finance, microlending (m-lending) firms such as Prosper, Funding Circle, Welab, and so on have increasingly tapped into online social media to extract vital signals to enhance credit risk prediction. On one hand, microlending firms may not have comprehensive financial records of their Internet-based clients. On the other hand, these m-lending firms also want to significantly expand their customer bases by evaluating the credit risk of their clients out of purely traditional quantitative features and signals. However, systematic studies about the effectiveness of leveraging multimodal social media signals for online credit scoring in the context of microlending are rare. Our study just tries to fill such a research gap by proposing a deep learning-based credit scoring model which utilizes multimodal signals extracted from online social media to enhance the credit scoring processes. Based on the real-world client data provided by a listed microlending firm, our experimental results show that the proposed deep learning-based credit scoring model that leverages multimodal social media signals can significantly improve credit risk prediction by 26.07% in terms of accuracy when compared to the same model that utilizes traditional quantitative features alone. Our research opens the door to apply deep learning and multimodal social media signals to enhance an array of Internet-based financial applications.