Credit analysis is required in a wide variety of decision of a modern economy.
It includes understanding the credit risk of small-medium enterprises (SMEs),
which today is the most significant contributor to the economy of almost every
nation. Creditors usually use credit scoring as a tool to predict the probability of
the SMEs to default in the future. The existing methods of SMEs credit scoring
still rely on traditional data, which may require high cost and have low scalability.
This thesis proposed an alternative approach of credit scoring for small-medium
enterprises (SMEs), which incorporate a novel set of features extracted from social
media data.
As a study case, we generate the credit scoring dataset which contains 20
traditional features and 35 social media features to quantify the creditworthiness
of more than 20,000 SMEs. The social media features are formulated based
on the previous studies in the adoption of social media data for personal credit
scoring and the social media metrics for quantifying business social perception.
To build the dataset, we develop the method to collect the information from some
public websites and SMEs’ Facebook page.
We conduct some experiments to develop credit scoring model for SMEs.
We found that using only the social media features insufficient to model SMEs
default in the future. However, by combining both social media features to build
the credit scoring model, we will get better performance compared to the model
developed using only traditional data.