A Social Relationships Enhanced Credit Risk Assessment Approach

Abstract

With the rapid growth of personal loan applications, credit risk assessment has become very crucial both in academic and industrial domain. Research literatures show that besides “hard” information, such as individual socio-demographic information and loan application information, “soft” information such as social relationships of the borrowers is a key factor to the credit risk assessment as social capital. In social networks, a user’s position and its influence are affected not only by the direct relationships (its friends) but also the indirect relationships (friends’ friends). A user’s importance and influence in his communities are attractive and valuable for credit assessment. But due to data deficiency in real life, social relationships are rarely considered in lending markets. By leveraging data from various sources, we proposed a social relationship enhanced credit risk assessment system, by building a social network from users’ geolocation data, extracting social relationship features at three different levels: ego, community and global level to capture a user’s position and influence from direct relationships, community and whole network perspectives. A real-life loan granting dataset is utilized for verifying the performance of the system. The experiment results show that, by combining the conventional financial indicators along with the proposed social network features, our system outperforms benchmark methods. Novel social network features we proposed make a good contribution to the loan default prediction. The research highlights the power of social relationships in detecting the default loans.